Downloaded 09/30/21 07:11 PM UTC

JANUARY 2008 H O T E I T E T A L . 131

Impact of Resolution and Optimized ECCO Forcing on Simulations of the Tropical Pacific

I. HOTEIT AND B. CORNUELLE Scripps Institution of Oceanography, University of California, San Diego, La Jolla, California

V. THIERRY Laboratoire de Physique des Océans, IFREMER, Brest, France

D. STAMMER Institut für Meereskunde, Hamburg, Germany

(Manuscript received 5 September 2006, in final form 24 April 2007)

ABSTRACT

The sensitivity of the dynamics of a tropical Pacific Massachusetts Institute of Technology (MIT) general circulation model (MITgcm) to the surface forcing fields and to the horizontal resolution is analyzed. During runs covering the period 1992–2002, two different sets of surface forcing boundary conditions are used, obtained 1) from the National Centers for Environmental Prediction/National Center for Atmo- spheric Research (NCEP/NCAR) reanalysis project and 2) from the Estimating the Circulation and Climate of the Ocean (ECCO) assimilation consortium. The “ECCO forcing” is the “NCEP forcing” adjusted by a state estimation procedure using the MITgcm with a 1° ϫ 1° global grid and the adjoint method assimilating a multivariate global ocean dataset. The skill of the model is evaluated against ocean observations available in situ and from satellites. The model domain is limited to the tropical Pacific, with open boundaries located along 26°S, 26°N, and in the Indonesian throughflow. To account for large-scale changes of the ocean circulation, the model is nested in the global time-varying ocean state provided by the ECCO consortium on a 1° grid. Increasing the spatial resolution to 1/3° and using the ECCO forcing fields significantly improves many aspects of the circulation but produces overly strong currents in the western model domain. Increasing the resolution to 1/6° does not yield further improvements of model results. Using the ECCO heat and freshwater fluxes in place of NCEP products leads to improved time-mean model skill (i.e., reduced biases) over most of the model domain, underlining the important role of adjusted heat and freshwater fluxes for improving model representations of the tropical Pacific. Combinations of ECCO and NCEP wind forcing fields can improve certain aspects of the model solutions, but neither ECCO nor NCEP winds show clear overall superiority.

1. Introduction 2002; Durand and Delcroix 2000; Lagerloef et al. 1999; Vialard et al. 2001). The global climate impacts associ- Since the strong influence of the Southern Oscillation ated with the strong 1997–98 El Niño event further un- (El Niño) on the extratropical climate began to unfold derscored the need for a better understanding of tropi- (Bjerknes 1966), much attention focused on the inves- cal Pacific dynamics and for accurate descriptions of the tigation of the tropical Pacific ocean circulation space–time structure of the tropical Pacific Ocean cir- (Johnson et al. 2002; Kessler et al. 2003; McPhaden et culation. Many studies have been based primarily on al. 1998; McCreary et al. 2002) using both observations surface datasets such as sea surface temperature (SST), and numerical models (e.g., Bonjean and Lagerloef wind stress, and sea surface height (SSH) fields. With the advent of Argo (Gould et al. 2004) and the presence Corresponding author address: I. Hoteit, Scripps Institution of of the Tropical Ocean Global Atmosphere-Tropical Oceanography, 9500 Gilman Dr., MC 0230, La Jolla, CA 92093. Atmosphere Ocean (TOGA-TAO) buoy network E-mail: [email protected] (McPhaden et al. 1998), the availability of subsurface

DOI: 10.1175/2007JTECHO528.1

JTECH2109 132 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 25 data is now also becoming more complete. However, a In cooperation with the Consortium for the Ocean’s detailed description of the tropical Pacific dynamics still Role in Climate (CORC) project, we intend to con- requires the support of model simulations driven by strain the model by all available data of the tropical surface fluxes of heat, momentum, and freshwater. Pacific and thereby to study the dynamics of the tropi- Complete spatial and temporal coverage of surface cal Pacific circulation. As a prerequisite to such a data fluxes of momentum, heat, and freshwater is available assimilation effort, we present a detailed model–data from atmospheric analysis centers. However, their forc- comparison without assimilation to test the model skill ing fields are prone to random uncertainties and signifi- and to demonstrate that the model is consistent with cant model biases (Yang et al. 1999; Putman et al. the observations, within the expected uncertainties of 2000). Wind stress errors can be partially remedied by both model and data. For that purpose we explore the using satellite wind stress fields (Grima et al. 1999). sensitivity of the model to the horizontal resolution and However, Auad et al. (2001) concluded from a study of to the forcing fields and study its dynamics as a function a tropical Pacific model response to National Centers of model resolution. A first assimilation study based on for Environmental Prediction (NCEP) and Compre- the model is provided by Hoteit et al. (2005). hensive Ocean–Atmosphere Data Set/Florida State Forcing fields used during our study were drawn University (COADS/FSU) forcing fields that NCEP from the National Centers for Environmental Predic- fluxes overall yielded a better basinwide ocean hind- tion/National Center for Atmospheric Research cast, except for serious problems related to weak in- (NCAR) reanalysis project (Kalnay et al. 1996) as well terannual variability in the NCEP forcing fields. A as from the ECCO results (Köhl et al. 2007). The number of GCM representations of the tropical Pa- NCEP forcing is available on a 1° ϫ 1° global grid and cific show bias in the cold tongue SST (Sun et al. 2003). contains twice-daily wind stress vectors and daily net Improvement of model simulations, such as reduc- heat flux, net shortwave radiation, and water flux at the tion of model bias errors, is expected to arise by im- sea surface. The ECCO forcing is the NCEP forcing proving the model physics, or by adjusting forcing adjusted by a global state estimation procedure using and boundary conditions to match the model to the Massachusetts Institute of Technology general cir- ocean observations through the use of data assimila- culation model (MITgcm) with a 1° ϫ 1° global grid and tion. the adjoint method assimilating a multivariate global Many assimilation systems in the tropical Pacific ocean dataset, as described in detail by Stammer et al. were based on techniques such as optimal interpolation (2002b, 2004). In the current experiments, we used the or the so-called three-dimensional variational methods (e.g., Behringer et al. 1998; Giese and Carton 1999). ECCO solution from iteration 69, which is nearly iden- These methods provide the analysis of the system state tical to that reported in Köhl et al. (2007). The ECCO using the observations at a given time without enforcing winds are given twice per day, and the heat and fresh- smoothness and dynamical consistency (i.e., conserva- water fluxes are daily, the same as NCEP. The assimi- tion properties) in the time dimension. An example of lation system significantly improved the agreement of ϫ a more advanced assimilation scheme is implemented the 1° 1° grid global model to SSH and SST obser- by Bennett et al. (2000), who used the representer vations, but the skill of the ECCO forcing for nested method to assimilate TAO data into an intermediate higher-resolution model runs remains unclear and will coupled ocean–atmosphere model. Bonekamp et al. be tested here. This question is of interest to the ocean (2001) used the adjoint method to adjust the model community at large since it is part of the general ques- wind stress over the tropical Pacific while constraining tion of how beneficial it is to optimize low-resolution the model to temperature data profiles in the frame- models and then use the adjusted control parameters work of an “identical twin” experiment. Recently, (forcing and boundary conditions in our case) for Weaver et al. (2003) successfully tested the incremental higher-resolution process models. four-dimensional variational data assimilation (4DVAR) The paper is organized as follows. Section 2 describes method (Courtier et al. 1994) using a primitive equa- the model configuration adopted for our experiments. tion model to assimilate temperature profiles by adjust- Section 3 describes the evaluation of the model with ing initial conditions. respect to the different datasets and presents sensi- The present study is part of the Consortium for Es- tivity studies of the model to the forcing fields and the timating the Circulation and Climate of the Ocean horizontal resolution. Section 4 discusses the momen- (ECCO) project (Stammer et al. 2002a). It is intended tum balance of the equatorial circulation as simulated to develop a nested eddy-permitting regional adjoint by the model. Concluding remarks are given in sec- assimilation procedure for the tropical Pacific Ocean. tion 5.

Unauthenticated | Downloaded 09/30/21 07:11 PM UTC JANUARY 2008 H O T E I T E T A L . 133

TABLE 1. Experiments for horizontal resolution sensitivity study and for forcing sensitivity study. “NCEPF” means original NCEP forcing, “ECCOF” means optimized forcing from the ECCO estimation, “NCEPW” means NCEP winds and ECCO heat and salinity fluxes, and “ECCOM” means NCEP forcing plus mean ECCO adjustments. In all cases, the model runs on a 1/3° ϫ 1/3° grid. ECCOF and NCEPF will be used to assess the quality of the adjustments produced by the ECCO optimization procedure relative to the NCEP forcing fields. ECCOF and NCEPW only differ in wind stress, to assess the quality of the wind forcing. ECCOM was performed to test the skills of the mean ECCO adjustments. Finally, the comparison between NCEPF and NCEPW, which are forced with different heat (HF) and freshwater (E Ϫ P) fluxes but identical wind stress, will be used to assess the effects of these forcings. LP stands for Laplacian operator and BH is for biharmonic operator.

Expt Resolution ⌬t Viscosity Diffusivity Wind HF/E Ϫ P 1DEG 1° 60 m LP, 104 m2 sϪ1 LP, 102 m2 sϪ1 ECCO ECCO ECCOF 1/3° 30 m BH, 1011 m4 sϪ1 LP, 5 ϫ 102 m2 sϪ1 ECCO ECCO 1SIXTH 1/6° 5 m BH, 5 ϫ 1010 m4 sϪ1 LP, 2.5 ϫ 102 m2 sϪ1 ECCO ECCO ECCOM 1/3° 30 m BH, 1011 m4 sϪ1 LP, 5 ϫ 102 m2 sϪ1 NCEP ϩ ECCO NCEP ϩ ECCO NCEPF 1/3° 30 m BH, 1011 m4 sϪ1 LP, 5 ϫ 102 m2 sϪ1 NCEP NCEP NCEPW 1/3° 30 m BH, 1011 m4 sϪ1 LP, 5 ϫ 102 m2 sϪ1 NCEP ECCO

2. The model monthly climatology (Levitus and Boyer 1994; Levitus et al. 1994) with a 30-day time scale. a. Model description b. Open-boundary conditions The numerical model (MITgcm) is described by Mar- shall et al. (1997). It is based on the primitive (Navier– To account for large-scale circulation changes, the Stokes) equations on a sphere under the Boussinesq model is nested into the ECCO global 1° optimization approximation operated in a hydrostatic mode with an (Köhl et al. 2007), which provides an estimate of the implicit free surface. No-slip conditions are imposed at complete ocean state for the period 1992 through 2002. the lateral boundaries while the friction condition is Open boundaries (OBs) for the regional model are pre- quadratic at the bottom with a drag coefficient equal to scribed at 26°S and 26°N, as well as at four straits in the 0.002 mϪ1. The subgrid-scale physics is approximated Indonesian throughflow (ITF), using monthly ECCO by a tracer diffusive operator of second order in the fields. The Lombok Strait and Sumba Strait are south- vertical and by the K-profile parameterization (KPP) in ern open boundaries, while the Ombai Strait and the the surface mixed layer (Large et al. 1994). The hori- Timor Passage are western open boundaries. The OBs zontal diffusive and viscous operators are either of sec- are specified as described by Zhang and Marotzke ond or fourth order, depending on the experiment (see (1999) using monthly mean results of the ECCO global Table 1). The mixing parameters are adjusted as the estimate. cube of the horizontal grid spacing to account for me- The normal velocity fields across the open bound- ridional varying horizontal grid spacing. The model do- aries have been further adjusted (after interpolation to main covers the tropical Pacific basin from 26°Sto26°N the higher-resolution domains) (i) to impose the same and from 104°Eto68°W. The vertical resolution is 10 m transport at 26°N and 26°S as in the global ECCO from the surface to 300 m in depth, with spacing gradu- model, and (ii) to bring the time-mean volume trans- ally increasing below 300 m for a total of 39 layers. The ports across all open boundaries into balance. The ad- maximum bottom depth is 6000 m and the bathymetry justments at each boundary are added as a barotropic is extracted from the global topography prepared by velocity uniformly distributed over all grid points. At Smith and Sandwell (1997). the northern boundary, the net, time-mean, meridional The study covers a 9-yr period from 16 January 1992 transport over the entire period is negligible and fluc- to 31 December 2000, which includes a strong El Niño tuations in monthly averages are smaller than 1 ϫ 106 event. The model is initialized with January potential m3 sϪ1. This is a result of a closed Bering Strait in the temperature and salinity fields provided by the ECCO global ECCO model and the use of a virtual salt flux for global optimization procedure on a 1° ϫ 1° grid (Köhl evaporation minus precipitation (E Ϫ P). The north- et al. 2007), with the velocity fields adjusted over a ward transport entering the model at 26°S leaves the 2-week period. NCEP and ECCO forcing fields were model again through the ITF. The 11.4-Sv mean trans- interpolated to match the horizontal resolution of the port for 1992–2000 is partitioned among the different model grid and were applied with the same temporal straits as follows: 3 ϫ 106 m3 sϪ1 in the Timor Passage, resolution. In both cases, a relaxation term is included 5.8 ϫ 106 m3 sϪ1 in the Ombai Strait, 0.3 ϫ 106 m3 sϪ1 in the surface layer tracer equations, relaxing surface in the Sumba Strait, and 2.3 ϫ 106 m3 sϪ1 in the Lom- temperature and salinity fields toward a Levitus bok Strait. The transport through the Sumba Strait was

Unauthenticated | Downloaded 09/30/21 07:11 PM UTC 134 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 25 neglected. The adjustment to the barotropic velocities In addition various combinations of both forcing fields at the boundaries is not tuned to keep the temperature were used, such as NCEP winds combined with ECCO transports across the boundaries identical to the ECCO heat and freshwater fluxes. Also used were NCEP forc- values. At the northern boundary, where the net vol- ing fields with the time mean of the ECCO adjustments ume transport is zero, the time-mean meridional heat added. The experiments ECCOF and NCEPF (see flux over the entire period is 0.56 PW, compared to Table 1) will be used to assess the quality of the adjust- 0.6 PW before the barotropic adjustment, and 0.46 PW ments produced by the ECCO optimization procedure in the ECCO mean. The volume transport does not with regard to the NCEP forcing fields. The experiment vanish at the other boundaries so the heat fluxes cannot NCEPW differs from ECCOF only in its wind stress be calculated, but the net heat flux out of the domain forcing and is intended to assess the quality of the wind combining the southern boundary and the ITF is about forcing. ECCOM was performed to test the skills of the 0.6 PW. mean ECCO adjustments. Finally, the comparison between the experiments NCEPF and NCEPW, which are 3. Open-loop sensitivity studies forced with different heat and freshwater fluxes but identical wind stress, will be used to assess the effects of a. Experiments the thermohaline forcing. To study the sensitivity of the model’s state to details The ECCO adjustments increase the (1992–2000 in the forcing fields and to the model horizontal reso- time) mean easterly wind stress in the tropical Pacific lution, several 9-yr integrations were performed (Table while making much smaller changes in the meridional 1). These experiments help to determine the best first- wind component (Fig. 1). This agrees with Josey et al. guess state for future assimilation experiments. At the (2002), who reported that NCEP wind stress in the same time they are intended to quantify the minimum tropics was weak. There are also strong north/south model resolution required for acceptable physical accu- gradients in the ECCO zonal wind adjustments be- racy. One important question is why some global tween 5° and 10°N, increasing the wind stress curl in the OGCMs have a bias in SST in the cold tongue region of region of the North Equatorial Counter Current the eastern tropical Pacific (Sun et al. 2003). A specific (NECC). The ECCO adjustments increase the north- goal, therefore, is to identify to what extent forcing ward wind stress north of the equator and east of changes can control model biases in the tropical Pacific 240°E. There is also an increase in the southward wind (i.e., identify whether model biases could result from stress just south of the equator in the eastern Pacific, incorrect forcing) and to see if ECCO forcing can rea- increasing the divergence at the equator. The standard sonably replicate the cold tongue. deviations of the NCEP and ECCO wind stress are For a test of model sensitivity to horizontal resolu- displayed in Figs. 1c,d. For the ECCO winds, the wind tion, the model was run with three different spatial stress variability is significantly higher in the tropical grids: 1° ϫ 1°, 1/3° ϫ 1/3°, and 1/6° ϫ 1/6° (vertical Pacific than provided by NCEP. resolution was held constant at 39 levels). For this com- Stammer et al. (2004) investigated the ECCO heat parison, only the time step and the subgrid-scale hori- and wind stress adjustments in comparison to the zontal mixing of both heat and momentum were NCEP forcing fields. As shown in their study, the changed; all other parameters were held fixed. In par- ECCO heat flux into the ocean is significantly de- ticular, the model was driven by the ECCO forcing creased near the eastern edge of the domain. The fresh- during all model resolution sensitivity experiments. The water flux shows increased evaporation (by about 10%) run of the regional model at 1° ϫ 1° resolution with south of the equator and in the warm pool, while pre- ECCO forcing was intended to test the open boundary cipitation is increased in the central Pacific north of the conditions (the solution should and does match the equator. global 1° ϫ 1° solution except for the effects of the b. Model evaluation sponge layer at the boundary). It can also evaluate the effect of enhanced vertical resolution on the skill of In the following we will evaluate each model run the model (39 levels in the regional model versus 23 against ocean observations to determine the model skill levels in the global model). Since the global ECCO in simulating the ocean as a function of resolution and optimization was constrained by SSH and SST, the forcing. Observations from the tropical Pacific Ocean 1DEG model run can be expected to compare best to are available from the TOGA-TAO moorings these observations in the analysis. (McPhaden et al. 1998) and from several additional Forcing fields were varied for 1/3° model resolution, datasets. Those include surface drifter velocity data ranging from original NCEP forcing to ECCO forcing. (Lagerloef et al. 1999), and satellite observations of

Unauthenticated | Downloaded 09/30/21 07:11 PM UTC JANUARY 2008 H O T E I T E T A L . 135

FIG. 1. Mean of the adjustments imposed by the ECCO optimization procedure on the NCEP wind stress (differences between ECCO and NCEP mean wind stress) for the (a) zonal and (b) meridional components. Standard deviation of the (c) NCEP and (d) ECCO wind stress magnitude. Units are in NmϪ2.

SSH from the Ocean Topography Experiment basinwide slope is best reproduced in runs using the (TOPEX)/Poseidon (T/P; Fu et al. 1994) and of SST ECCO forcing and is weakest in the NCEPF and from Reynolds and Smith (1994). The metric of good- NCEPW runs. The difference in modeled slope, stem- ness used throughout is the weighted RMS misfit be- ming from model differences in eastern basin SSH, is in tween the model and the observed mean and time- part due to the increased mean easterly wind stress in varying ocean state. In that sense it is the metric that is the ECCO forcing, as can be seen from the differences usually being minimized during ECCO optimization between ECCOF and NCEPW runs: the mean ECCO runs. However, we also include parameters such as the easterlies within 5° of the equator are 50% stronger basinwide zonal slope in sea level in the comparison than NCEP winds and produce a 30% larger SSH with T/P results. Comparisons will be presented separately for TOPEX/Poseidon SSH data, for Reynolds and Smith (1994) SST fields, for TAO subsurface tem- peratures and velocities, and for surface drifter velocities and ADCP subsurface velocities.

1) TOPEX/POSEIDON A comparison with sea level data is provided here in terms of the cross-basin zonal slope of the sea surface along the equator observed by the time-mean T/P SSH observations minus the most recent Gravity Recovery and Climate Experiment (GRACE) geoid, and in terms of the RMS misfit between observed and simulated SSH variability. Although details in the simulated and observed time-mean SSH structure on the equator differ substantially at the western side of the section, the cross-basin slope in SSH appears well reproduced in FIG. 2. Mean SSH on the equator from T/P data and the most model runs (Fig. 2). However, we note that the different model runs (see Table 1 for list of runs).

Unauthenticated | Downloaded 09/30/21 07:11 PM UTC 136 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 25

for the NCEPF run, which suggests that the variability of the ECCO winds is responsible for much of the loss in skill in SSH horizontal structure. The small differences between NCEPW and NCEPF show the relatively small effect of the heat and freshwater fluxes on the horizontal SSH structure. For reference, the observed RMS SSH calculated for 1-yr sliding windows (not shown) varies interannually, starting at about 15 cm in 1993 and increasing to about 16–17 cm at the end of 2000, reaching a maximum of about 18 cm during the 1997–98 El Niño event. Using those numbers, the RMS misfits in Fig. 3 could be con- verted into a fractional RMS by dividing by the reference RMS, and would range from a minimum of about 0.15 for the 1DEG run around 1995–96 to about 0.5 for the 1SIXTH run in 1993 and late 1997. FIG. 3. RMS difference with the spatial mean removed between The RMS misfits (in space) between the time-mean T/P anomalies and the different model runs (see Table 1 for list of runs). Units: cm. SSH from the model runs and T/P data over the whole domain show a similar tendency, in that the 1DEG run height difference (Fig. 2) along the equator between is best, but the other runs do not differ significantly (4.4 160° and 240°E. The difference between ECCOM and cm for the 1DEG run, 6.5 cm from the NCEPW run, 6.7 NCEPW is due to both the addition of the ECCO mean cm from the 1SIXTH run, 6.9 cm for the NCEPF run, to the NCEP winds and to the difference between the 7.05 cm from the ECCOF run, and 7.3 cm for the ECCO and NCEP heat and freshwater flux variability. ECCOM run) because part of the misfit is due to errors That the adjusted heat and freshwater fluxes play some in the time-mean SSH estimate (i.e., due to geoid er- role in setting the SSH slope along the equator is docu- rors). mented by the difference between the NCEPF and Finally, we compare the observed point-by-point NCEPW runs. Finally, the difference between ECCOF standard deviation for the monthly mean SSH with and ECCOM shows that the effect of the ECCO vari- similar fields obtained from experiments 1DEG, ability in all forcing components also plays a role in NCEPF, and ECCOF, respectively. SSH anomalies shaping the mean equatorial SSH. were calculated relative to the 1993–2000 mean SSH for In contrast to the time-mean cross-basin SSH slope, observations and model simulations alike. Because the which appears less sensitive to the horizontal model results from the runs NCEPF, ECCOF, and 1° ϫ 1° resolution, the time-varying SSH signal is significantly roughly (leaving out 1SIXTH) define the envelope of affected by the model’s horizontal resolution as well as all results in Fig. 3, we will restrict all further compari- by details of the forcing fields, especially the wind stress sons to those runs. (Fig. 3). Time and spatial means were removed from There are four main “hot spots” for SSH variability the SSH before the calculation, but the seasonal cycle (Fig. 4): on the equator in the cold tongue/Niño-3 re- was not removed. Again the 1° model run (1DEG) gion, north of the equator in the eastern Pacific where shows the best RMS agreement with observed variabil- wind jets through orographic gaps such as the Tehua- ity (about 4.5-cm RMS misfit, or about 20% of the ntepec trigger mesoscale eddies (Chelton et al. 2001; signal) throughout the period, followed by experiments Kessler 2002), near the east coast of Mindanao where NCEPF and NCEPW. In contrast, model runs with 1/3° the NECC begins and the North Equatorial Current and 1/6° resolution forced by ECCO fields lead to the (NEC) splits, and off the east coast of New Guinea in largest discrepancies relative to T/P monthly mean SSH the region of horizontal shear between the South Equa- observations. This indicates that the skill of the 1° torial Current (SEC) and the South Equatorial Counter model does not carry over to higher resolution because Current (SECC; Qiu and Chen 2004). The run 1DEG the ECCO forcing fields estimated with lower resolu- reproduces these hot spots reasonably well, albeit with tion partly compensate the higher viscosity and slug- somewhat weak variability. The variability of the SSH gishness of the 1° model. The ECCO fields therefore obtained from the NCEP forcing is generally weaker make the higher-resolution runs too energetic, espe- than observed, especially in the eastern Pacific. The cially in the western part of the basin (see below). The dominance of the wind stress as a source of variability misfit for the ECCOM run is only slightly larger than of SSH is demonstrated by the similarity of the NCEPF

Unauthenticated | Downloaded 09/30/21 07:11 PM UTC JANUARY 2008 H O T E I T E T A L . 137

FIG. 4. Standard deviation of monthly averaged SSH fields from (a) T/P data, (b) run with NCEP forcing, (c) 1/3° run with ECCO forcing, (d) 1° ϫ 1° run, (e) ECCOM forcing, and (f) NCEPW. and NCEPW runs, which differ only in the buoyancy 1DEG run does best, showing that the optimized forc- forcing. The importance of the time-dependent winds ing retains good skill in the regional model even after a can be assessed by comparing ECCOF and ECCOM. change to higher vertical resolution. The ECCO forcing The largest differences reside in the eastern Pacific, fields have mixed results when applied to higher- where the Tehuantepec eddies are nearly absent from resolution models, generally giving better agreement in the ECCOM run and the cold tongue variability is simi- the eastern Pacific and worse agreement in the west. lar to NCEPF. This suggests that enhanced variability Wind stress has the largest effect on SSH variability, of the ECCO zonal and meridional wind stress in the but heat and freshwater fluxes have a significant effect eastern Pacific is responsible for the increased cold on the time-mean SSH structure, suggesting that errors tongue variability and for triggering the eddies in the in heat fluxes and freshwater fluxes could also be a eastern Pacific. In that region all model runs show a significant source of model biases in the tropical Pacific. dynamic height field that resembles that shown by Of course, incorrect physics might also contribute to Kessler (2002), with the ECCOF run having the best model biases. agreement in the strength and location of the Costa 2) REYNOLDS SST DATA Rica Dome and the anticyclone to the northwest (not shown). Excess SSH variability in the western Pacific, Typically SST fields are used to evaluate the perfor- present in the ECCOF and the 1SIXTH run (the SSH mance of models in specific regions of the tropical Pa- variability of the 1SIXTH run is comparable but about cific (e.g., Niño-1–Niño-4). We will use the SST fields as 15% stronger than what is found in ECCOF), is re- a measure of model skill over the entire model domain duced in the ECCOM, suggesting that the ECCO wind so as to compare with the earlier SSH comparisons. For stress variability in the western Pacific is too strong or later reference, the observed RMS of the monthly that variability is propagating westward. mean SST anomalies varies between 0.8° and 2°C To summarize the results of the SSH comparison, the across the model domain. The RMS differences be-

Unauthenticated | Downloaded 09/30/21 07:11 PM UTC

Fig 4 live 4/C 138 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 25

Reynolds time-mean SST field both in amplitude and spatial structure. These runs (ECCOF, ECCOM, 1DEG, and 1SIXTH) all have roughly similar structures in the eastern Pacific. In contrast, the cold tongue in the NCEP runs is over 2° too warm and too narrow in latitude. The NCEPW run (not shown) is of intermediate quality, showing a cold tongue that is significantly cooler than NCEPF but still warmer than ECCOF or the observations. This means that both wind stress and ECCO heat flux forcing anomalies affect the cold tongue in this model. ECCOM forcing (ECCO mean corrections added to NCEP) shows a slightly better (colder) cold tongue than ECCOF. We speculate that since the enhanced wind stress variability in ECCO forces increased eddy energy, these eddies may tend to mix the cold tongue away (Swenson and Hansen 1999). FIG. 5. RMS differences (°C) between monthly means of the In summary it appears that not just the ECCO wind different model runs and Reynolds SST with temporal means removed. stress but also the ECCO heat flux are essential ingre- dients for this success and that the time-varying forcing tween the model and Reynolds and Smith (1994) is required to reproduce the time-mean SST field. monthly mean SST fields with the temporal means re- There are two obvious hypotheses for explaining the moved (Fig. 5) reveal that the SST fields from different improvement of the representation of the cold tongue runs appear more similar as compared to the previous using ECCO forcing. Either (i) the NCEP forcing is SSH analysis. This is partly due to the restoring of SST correct and the model has errors that cause a bias in the toward the Levitus monthly mean SST climatology. SST (and other fields) or (ii) the model is correct and Over the cold tongue this can lead to an extra heat the ECCO adjustments to the NCEP forcing result in forcing on the order of 20% of the atmospheric heat better agreement with the observations because they flux signal there. In spite of the restoring, the model are more correct. As mentioned above, Stammer et al. runs still differ significantly in SST. Later runs without (2004) supported (ii) by arguing for the plausibility of the restoring term verify that it does not qualitatively the ECCO forcing adjustments. Further data are re- change the differences between runs. quired to distinguish between the hypotheses. For ex- Overall, the RMS model data difference is less than ample, if the forcing fields were made less uncertain 50% of the observed RMS SST variability. As before, using new satellite products, then any model data misfit the best results are obtained from the 1DEG run, di- resulting from these fields would have to be explained rectly followed by the higher-resolution ECCO runs. as model error or as initial and boundary condition This indicates that the optimized ECCO heat fluxes effects, which would need to be ruled out by observa- retain skill in the higher-resolution runs. In contrast, tions. the NCEPF and ECCOM runs produce the worst 3) TAO SUBSURFACE TEMPERATURE DATA agreement with the data, followed by the NCEPW run. The ECCOM run is significantly worse than the ECCOF The Tropical Atmosphere and Ocean buoy array run at several points in the record (1998 and 1999, in consists of about 70 Autonomous Temperature Line particular), underlining the importance of the time vari- Acquisition System (ATLAS) wind and thermistor ability in the ECCO forcing for controlling the SST. chain moorings and current meter moorings in the The SST and SSH (Fig. 4) model data misfits from all tropical Pacific between 8°S and 8°N. Data from the runs show a strong temporal structure, highest during TAO array allow a detailed comparison of temperature the first two years and the 1997/98 ENSO event. This and velocity with the model results along the equator at misfit is significantly higher in NCEP results and less the surface and subsurface. A time–longitude plot of pronounced in ECCO results. The initial misfit is most the SST anomalies along the equator from selected runs likely due to spinup adjustment in all model runs. compared to the TAO data shows that the 1DEG Important features in the observed time-mean SST model agrees best with the TAO SST data, as expected field are the cold tongue in the east and the warm pool from previous comparisons (rhs of Fig. 6). However, in the west (Fig. 6). The runs with ECCO forcing lead the 1DEG run simulates El Niño and La Niña events to the best agreement with the cold tongue in the that are slightly stronger in SST than observed in the

Unauthenticated | Downloaded 09/30/21 07:11 PM UTC JANUARY 2008 H O T E I T E T A L . 139

FIG. 6. (left) Mean of the SST from (a) Reynolds data, (b) run with NCEP forcing, (c) run ECCO forcing, and (d) 1° ϫ 1° run. (right) Longitude–time plots of the SST along the equator from (e) TAO data, (f) run with NCEP forcing, (g) run with ECCO forcing, and (h) 1° ϫ 1° run. east but somewhat weaker than observed in the west. In significant correlations at a 95% confidence level exist the eddy-permitting model, the ECCO winds produce between the observed and the simulated temperature ENSO events with approximately the right strength in signal at almost all TAO locations (insignificant corre- the east but with significantly less strength in the west lations are marked in the figure). Model performance is than produced by NCEP winds, which gives SST values good at 5- and 100-m depth on the equator, but poor at closer to the observations in the west. The 1SIXTH run 250 m, perhaps because of the smaller signal at depth (not shown) has similar (but about 15% stronger) vari- (RMS of temperature at 100 m is above 2°C compared ability than the ECCOF run. to about 0.5°C or less at 250 m). The curves are com- A more quantitative measure of the similarity be- plicated, but the summary of the comparisons off the tween model and data comes from correlating the equator is that the 1DEG is usually among the best, model temperature with the TAO temperature time except for 8°N,5minthewestern Pacific where series at various longitudes, latitudes, and depths (Fig. NCEPF and NCEPW are clearly the best. This suggests 7). Statistical tests (based on the probability of getting that, even for the 1DEG model, ECCO winds are not a correlation as large as the observed value by random good in the west, as we have seen before. Because EC- chance, when the true correlation is zero) revealed that COM does poorly in the west compared to NCEPF, the

Unauthenticated | Downloaded 09/30/21 07:11 PM UTC

Fig 6 live 4/C 140 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 25

FIG. 7. Correlations between TAO and model temperature from the different model runs (see Table 1 for list of runs). Correlations at (left) 5-m, (middle) 100-m, and (right) 250-m depths for (top) 8°N, (middle) the equator, and (bottom) 8°S. Insignificant correlations at the 95% confidence level are marked with symbols of the same colors. mean ECCO winds are adversely affecting the compari- a 3D wind-driven circulation pattern that rises toward son in the west. Plots of cross-equatorial temperature the ocean surface as it travels from west to east along and salinity sections at 170°W (not shown) show no the equator. clear distinctions between the different model runs. The standard World Ocean Circulation Experiment (WOCE)-TOGA drifters with drogues at 15-m depth 4) CURRENT OBSERVATIONS (Yu et al. 1995) provide a good in situ mean surface The strong currents of the tropical Pacific Ocean are velocity dataset for model data comparisons (Fig. 8). well studied and provide a benchmark for model simu- Time-mean drifter velocities were computed as aver- lations. The strongest currents in the region are the ages over 5-day intervals in 1/2° latitude by 2° longitude NECC, Equatorial Undercurrent (EUC), NEC, and bins from the entire time of the drifter record. Overall SEC, which are mainly wind driven (Halpern et al. the velocity structure observed by the drifters is quali- 1995; Blanke and Raynaud 1997; Pedlosky 1996). The tatively simulated by all the model runs. Once again, EUC is associated with the thermocline and halocline the ECCOF and ECCOM currents are too strong in the structure on the equator (Pedlosky 1996) and is part of western tropics, where the NCEPF and NCEPW runs

Unauthenticated | Downloaded 09/30/21 07:11 PM UTC

Fig 7 live 4/C JANUARY 2008 H O T E I T E T A L . 141

FIG. 8. Mean zonal velocity at 15-m depth from (a) drifters data, (b) run with NCEP forcing, (c) run with ECCO forcing, and (d) 1° ϫ 1° run. show better agreement with the observed velocity tion, which has even weaker EUC maxima, showing the structures. In contrast, ECCOF and ECCOM are best importance of vertical resolution. The 1SIXTH run has in the eastern tropics, including the banded structure a slightly more compact EUC, but the differences are south of the equator. Meridional velocities (not shown) not large, and the agreement with the data is not better show similar tendencies. The agreement between the than for ECCOF. The maximum current speed in the model and drifter results suggests that a possible bias EUC is about 15% higher for ECCOF, which has to be between them due to the different averaging periods compared to a 30% larger SSH height difference, indi- and the significantly reduced observation count during cating that the dynamics are more complicated than a the model period (Lagerloef et al. 1999) is not domi- simple frictional balance. nant. TAO profiles of RMS zonal and meridional velocity Mean subsurface zonal model currents can be com- variability compared with respective model results pared against shipboard ADCP measurements show that overall the 1DEG runs are too weak in ve- (Johnson et al. 2002) available along the equator (Fig. locity variability (Fig. 10). With a 1/3° resolution, the 9). Consistent with the the sea level slopes (Fig. 2), the ECCO wind stress (case ECCOF) has good agreement NCEPF and NCEPW runs show less west–east ther- at 110°W for both zonal and meridional velocity, while mocline and EUC tilt than are observed but each pro- NCEP winds produce relatively low variability. Some duce an EUC that is too strong west of the date line. In of this variability comes from tropical instability waves contrast, the ECCOF and ECCOM runs show the best (TIWs), which seem to be more realistic in runs with agreement with the observed EUC structures, both in ECCO forcing when comparing frequency spectra (not speed and location. This is also consistent with TAO shown). As can be expected from the previous discus- moored ADCP observations (not shown). For example, sion, NCEP winds lead to more realistic variability in on average, the velocities of the ECCOF EUC reach the western model domain; there the ECCOF and about 1.0 m sϪ1 at 110-m depth at 220°E, in agreement 1SIXTH runs show too much shallow variability in with the Johnson et al. (2002) analysis and the mean zonal and meridional direction, due to the effects of the from the TAO data, whereas the maximum EUC simu- eddies growing on the stronger currents driven by the lated by the NCEP winds is around 0.85 m sϪ1. The ECCO forcing. 1DEG run has a very sluggish EUC, showing the effects Based on the correlation between observed and of the poor horizontal resolution. The 1DEG run has simulated zonal velocities, most model runs are of com- higher vertical resolution than the global ECCO solu- parable quality in simulating the zonal variability at

Unauthenticated | Downloaded 09/30/21 07:11 PM UTC 142 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 25

FIG. 9. Vertical distribution of the mean zonal current along the equator from (a) Johnson ADCP data, (b) run with NCEP forcing, (c) run with ECCO forcing, and (d) 1° ϫ 1° run. Positive values indicate eastward current.

110°W and 147°E (Fig. 11). An exception is the 1DEG in a 9-yr average, the acceleration residuals ax and ay model in the core of the EUC at 110°W, which may be representing the eddy fluxes and frictions are related to the weakness of the EUC in that run. Inter- Ѩ␾ ͒ ͑ ϩ ١ estingly, the 1DEG run has the best skill between 50- ϭϪ ␷ ϩ ax f u · u Ѩ 1 and 200-m depth at 165°E. Note also the small differ- x ences between the NCEPF and NCEPW curves in most and regions of significant correlation. The two runs differ Ѩ␾ ͒ ͑ ١␷ ϩ only by the ECCO heat and freshwater fluxes, which do ϭ ϩ ay fu u · Ѩ , 2 not apparently contribute strongly to the velocity vari- y ability. ECCOF does somewhat worse than NCEPW where ␾ is the dynamic pressure, u is the 3D velocity, for zonal velocity at 165°E, but is better at 147°E, al- and f is the Coriolis parameter. Ͼ though no model has correlation 0.6 there. Zonal (left panels) and meridional (right panels) acceleration terms (m sϪ1) for the 1°, 1/3°, and 1/6° nu- 4. Equatorial momentum balances merical model forced by ECCO winds were evaluated along the section 189°E at a depth of 140 m (i.e., at the The previous section focused on an analysis of the depth of the EUC at that longitude; Fig. 12). This lo- model’s skill in simulating observed mean and variable cation is chosen to be away from complex topography ocean structures in SSH, SST, and velocity. We will and near the center of the basin. The zonal momentum now shift our attention to a question: To what extent budget shows the importance of the zonal pressure gra- does the model converge toward a stable estimate of dient and the nonlinear terms. The meridional budget the time-mean equatorial momentum balances as a shows a nearly geostrophic balance, as expected (Ped- function of model resolution? To this end we use the losky 1996), with the matching slopes of the dp/dy and y2 relationshipץ/2uץoutput from several runs, driven by ECCO forcing, but fu terms suggesting that the ␤u ϭ that differ in their spatial resolution. The deterministic holds, although it is not as good in the 1° model. terms in the momentum equation are calculated from When sufficiently resolved, the zonal-momentum the 9-yr mean velocity and pressure fields. These terms balance on the equator is dominated by a balance be- do not include the Reynolds stress divergences and fric- tween the pressure gradient and the zonal advection of tion terms. Ignoring the time derivative, which is small the zonal gradient of zonal velocity (Uux), reflecting

Unauthenticated | Downloaded 09/30/21 07:11 PM UTC JANUARY 2008 H O T E I T E T A L . 143

FIG. 10. Profiles of the standard deviation for the (top) zonal and (bottom) meridional velocity along the equator at 110°W, 165°E, and 147°E as estimated from the model ADCP measurements and the different model runs (see Table 1 for list of runs).

FIG. 11. Vertical distribution of the correlations between TAO data and model zonal velocity from the different model runs (see Table 1 for list of runs). Insignificant correlations at the 95% confidence level are marked with symbols of the same colors.

Unauthenticated | Downloaded 09/30/21 07:11 PM UTC

Fig 10 and 11 live 4/C 144 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 25

Ϫ FIG. 12. (left) Zonal and (right) meridional acceleration terms (m s 1) at 140-m depth for 1° ϫ 1°, 1/3° ϫ 1/3°, and 1/6° ϫ 1/6° numerical models forced by ECCO forcing. The sections are shown at 189°E. Balances for the (top) 1° resolution, (middle) 1/3° resolution, and (bottom) 1/6° resolution. the increasing zonal velocity downstream in the EUC at difference between the Coriolis term (green line) and that longitude. Between the equator and 3° off the the summed pressure gradient and advective accelera- equator, the meridional advection of the meridional tions (red line). This is due to both viscosity and Reyn- gradient of zonal velocity (Vuy) takes over in the bal- olds stress. The 1/3° resolution model has a smaller ance, showing the acceleration of the water converging residual than the 1/6° model outside the EUC, due to into the undercurrent. The vertical advection of the the lower level of eddy mixing seen in the 1/3° simula- vertical gradient of zonal velocity (Wuz) is relatively tion. small, but shows acceleration of the water moving up- The zonal momentum balance on the equator and ward. The Wuz term shows asymmetry of the interac- especially the role of nonlinear terms in the balance as tion between the EUC and the shallow overturning cell a function of depth have been examined previously by near the equator, and the EUC is stronger north of the Qiao and Weisberg (1997) from moorings on the equa- equator than south at 140-m depth. There is a residual tor at 140°W. Comparisons of 1/3° and 1/6° model runs

Unauthenticated | Downloaded 09/30/21 07:11 PM UTC

Fig 12 live 4/C JANUARY 2008 H O T E I T E T A L . 145 with the Qiao and Weisberg (1997) results (not shown) curl around the NECC in the western Pacific, forcing reveal qualitative agreement, although the momentum an overstrong NECC. We note that the heat flux is flux divergences seen in the model are about one-half important for improving the mean SST of the simu- the size of means seen by Qiao and Weisberg (1997). lations. ECCO heat fluxes produce a more accurate z) below the EUC, cold tongue, and the stronger ECCO wind stress inץ/uץ)We also see a negative lobe in w which is absent in their data. Kessler et al. (2003) ex- the eastern tropical Pacific produces better velocity amined the dynamics of vertically integrated equatorial variability on the equator. Adjusted surface heat and currents using the Gent and Cane (1989) model and freshwater fluxes appear important for removing ap- found zonal momentum flux divergence with a qualita- parent SST biases in the tropical Pacific, but this does tively similar shape to that found here, although the not distinguish between model errors and errors in figures show the flux form of the balance, which must the forcing dataset. be compared to the sum of the individual term esti- • The ECCO forcing fields have been determined so mates shown here (the MITgcm uses the flux form in that a 1° ϫ 1° global model leads to the best agree- the numerics, but the individual terms have been esti- ment with ocean observations. As a result, the forcing mated separately here for discussion). seems to be too strong in some regions for a high- The 1° run shows significantly poorer balances, which resolution model with lower friction and better physi- is expected. The 1/3° run seems to have just sufficient cal representation, because the assimilation strength- resolution to resolve the mean zonal momentum terms, ened the forcing to counter deficiencies in the physics and turbulent stresses (which are the residuals in the of the 1° model. This was particularly evident in the curves) are of similar sizes in each case. Results from northeastern Pacific where the strong ECCO wind the 1/6° run show further changes in the dynamical bal- stress curl may have compensated for an overly ances; however, the changes relative to the 1/3° run are damped 1° model. Nevertheless, agreement was en- much smaller than what can be seen between the 1/3° hanced in many ways by using the forcing optimized run and the 1° run. We conclude from the figure that a fora1° ϫ 1° model as a starting point for a further model resolution of 1/3° should be barely sufficient to optimization with the higher-resolution model. Sen- perform data assimilation in the tropical Pacific. sitivity to boundary conditions was not explored; they were always taken from the ECCO model. • The 1/6° run does not show a significant improve- 5. Discussion ment in the model performance over the 1/3° run, In this paper we evaluated a regional general circu- despite the use of lower viscosity and diffusivity val- lation model for the tropical Pacific Ocean by varying ues. In contrast, the increase in vertical resolution horizontal resolution and surface forcing and compar- from 23 to 39 levels and the associated decrease in ing against several datasets during 1992–2000. The the vertical viscosity have significant effects on the model was run with 1°, 1/3°, and 1/6° grid resolutions tropical currents in the 1° model runs. However, re- and with combinations of forcing fields obtained from cent results from a 1/3° run with 50 layers suggest no NCEP and ECCO products. The latter are the NCEP significant changes compared to the same run but forcings adjusted by a global 1° model optimization with 39 layers. procedure (Stammer et al. 2004). The goal of the study • A 1/3° horizontal resolution seems just sufficient for is to evaluate the model and to determine the model a skillful simulation of the tropical Pacific circulation configuration that can be used in a quantitative data and its momentum balances. With this resolution, the assimilation approach at reasonable computational ECCO wind stress produces a NECC trough that is cost. The study also closely examined the benefit of too deep, resulting in overly large instability driven using optimized global coarse-resolution products by eddy variability in the western tropical Pacific. (ECCO) to force our nested higher-resolution tropical The estimated time variability of the ECCO wind Pacific model (boundary conditions and surface forc- stress curl is important for stimulating eddy variabil- ing). Our main conclusions can be summarized as fol- ity in the Gulf of Tehuantepec, which is nearly absent lows. in the NCEP-forced runs. • Because the 1/3° solution is sensitive to details of the • As seen in many other studies, the current system of diffusivity and viscosity values, it may become impor- the tropical Pacific Ocean is strongly sensitive to the tant to include those parameters in the set of control wind stress forcing. The ECCO forcing appears to parameters in an optimization of the tropical Pacific produce better model outputs than the NCEP forcing circulation. The present study, however, cannot con- in the eastern Pacific, but has too strong a wind stress clusively determine the real effects of (parameter-

Unauthenticated | Downloaded 09/30/21 07:11 PM UTC 146 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 25

ized) mesoscale eddies in large-scale simulations of Durand, F., and T. Delcroix, 2000: On the variability of the tropi- the tropical Pacific Ocean. This problem has been cal Pacific thermal structure during the 1979–96 period, as addressed previously by Stockdale et al. (1993) and deduced from XBT sections. J. Phys. Oceanogr., 30, 3261– 3269. Megann and New (2001). Fu, L., E. Christensen, and C. Yamarone, 1994: TOPEX/ The conclusions suggest that the optimization of the POSEIDON mission overview. J. Geophys. Res., 99, 24 369– 24 381. model parameters, such as the forcing fields and initial Gent, P. R., and M. A. Cane, 1989: A reduced gravity, primitive conditions, could improve the model behavior by mak- equation model of the upper equatorial ocean. J. Comput. ing it more consistent with the available data. Our next Phys., 81, 444–480. step will therefore be to use the adjoint method to as- Giese, B. S., and J. A. Carton, 1999: Interannual and decadal vari- similate all available data in the tropical Pacific Ocean. ability in the tropical and midlatitude Pacific Ocean. J. Cli- mate, 12, 3402–3418. Gould, J., and Coauthors, 2004: Argo profiling floats bring new Acknowledgments. Reanalysis surface forcing fields era of in situ ocean observations. Eos, Trans. Amer. Geophys. from the National Centers for Environmental Predic- Union, 85, 185–190. tion/National Center for Atmospheric Research were Grima, N., A. Bentamy, K. Katsaros, and Y. Quilfen, 1999: Sen- obtained through a computational grant at NCAR. sitivity of an ocean circulation model forced with satellite Computational support from the Center of Modeling wind stress fields. J. Geophys. Res., 104, 7967–7989. Halpern, D., Y. Chao, C.-C. Ma, and C. Mechoso, 1995: Compari- and Prediction at Scripps and the National Center for son of tropical Pacific temperature and current simulations Atmospheric Research is acknowledged. This work was with two vertical mixing schemes embedded in an ocean gen- supported by NOAA through the Consortium for the eral circulation model and reference to observations. J. Geo- Ocean Role in Climate and by the Consortium for Es- phys. Res., 100, 2515–2522. timating the Circulation and Climate of the Ocean Hoteit, I., B. Cornuelle, A. Köhl, and D. Stammer, 2005: Treating strong adjoint sensitivities in tropical eddy-permitting varia- funded by the National Oceanographic Partnership tional data assimilation. Quart. J. Roy. Meteor. Soc., 131, Program (ONR ECCO Grants N00014-99-1-1049). 3659–3682. Johnson, G., B. Sloyan, W. Kessler, and K. McTaggart, 2002: Di- REFERENCES rect measurements of upper ocean currents and water properties across the tropical Pacific Ocean during the 1990’s. Auad, G., A. Miller, J. Roads, and D. Cayan, 2001: Pacific Ocean Prog. Oceanogr., 52, 31–61. wind stress and surface heat flux anomalies from NCEP re- Josey, S., E. Kent, and P. Taylor, 2002: Wind stress forcing of the analysis and observations: Cross-statistics and ocean model ocean in the SOC climatology: Comparisons with the NCEP– responses. J. Geophys. Res., 106, 22 249–22 265. NCAR, ECMWF, UWM/COADS, and Hellerman and Behringer, D. W., M. Ji, and A. Leetmaa, 1998: An improved Rosenstein datasets. J. Phys. Oceanogr., 32, 1993–2019. coupled model for ENSO prediction and implications for Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Re- ocean initialization. Part I: The ocean data assimilation sys- analysis Project. Bull. Amer. Meteor. Soc., 77, 437–471. tem. Mon. Wea. Rev., 126, 1013–1021. Kessler, W. S., 2002: Mean three-dimensional circulation in the Bennett, A. F., B. S. Chua, D. E. Harrison, and M. J. McPhaden, northeast tropical Pacific. J. Phys. Oceanogr., 32, 2457–2471. 2000: Generalized inversion of Tropical Atmosphere–Ocean ——, G. C. Johnson, and D. W. Moore, 2003: Sverdrup and non- (TAO) data using a coupled model of the tropical Pacific. J. linear dynamics of the Pacific equatorial currents. J. Phys. Climate, 13, 2770–2785. Oceanogr., 33, 994–1008. Bjerknes, J., 1966: A possible response of the atmospheric Hadley Köhl, A., D. Stammer, and B. Cornuelle, 2007: Interannual to circulation to equatorial anomalies of ocean temperature. decadal changes in the ECCO global synthesis. J. Phys. Tellus, 18, 820–829. Oceanogr., 37, 313–337. Blanke, B., and S. Raynaud, 1997: Kinematics of the Pacific equa- Lagerloef, G., G. Mitchum, R. Lukas, and P. Niiler, 1999: Tropical torial undercurrent: An Eulerian and Lagrangian approach Pacific near-surface currents estimated from altimeter, wind, from GCM results. J. Phys. Oceanogr., 27, 1038–1053. and drifter data. J. Geophys. Res., 104, 23 313–23 326. Bonekamp, H., G. J. van Oldenborgh, and G. Burgers, 2001: Large, W., J. McWilliams, and S. Doney, 1994: Oceanic vertical Variational assimilation of TAO and XBT data in the HOPE mixing: A review and a model with a nonlocal boundary layer OGCM, adjusting the surface fluxes in the tropical ocean. J. parameterization. Rev. Geophys., 32, 363–403. Geophys. Res., 106, 16 693–16 709. Levitus, S., and T. P. Boyer, 1994: Temperature. Vol. 4, World Bonjean, F., and G. S. E. Lagerloef, 2002: Diagnostic model and Ocean Atlas 1994, NOAA Atlas NESDIS 4, 117 pp. analysis of the surface currents in the tropical Pacific Ocean. ——, R. Burgett, and T. P. Boyer, 1994: Salinity. Vol. 3, World J. Phys. Oceanogr., 32, 2938–2954. Ocean Atlas 1994, NOAA Atlas NESDIS 3, 99 pp. Chelton, D. B., and Coauthors, 2001: Observations of coupling Marshall, J., A. Adcroft, C. Hill, L. Perelman, and C. Heisey, between surface wind stress and sea surface temperature in 1997: A finite-volume, incompressible Navier Stokes model the eastern tropical Pacific. J. Climate, 14, 1479–1498. for studies of the ocean on parallel computers. J. Geophys. Courtier, P., J.-N. Thepaut, and A. Hollingsworth, 1994: A strat- Res., 102, 5753–5766. egy for operational implementation of 4D-Var, using an in- McCreary, J. P., P. Lu, and Z.-J. Yu, 2002: Dynamics of the Pacific cremental approach. Quart. J. Roy. Meteor. Soc., 120, 1367– subsurface countercurrents. J. Phys. Oceanogr., 32, 2379– 1388. 2404.

Unauthenticated | Downloaded 09/30/21 07:11 PM UTC JANUARY 2008 H O T E I T E T A L . 147

McPhaden, M. J., and Coauthors, 1998: The Tropical Ocean- sch, 2004: Estimating air-sea fluxes of heat, freshwater, and Global Atmosphere (TOGA) observing system: A decade of momentum through global ocean data assimilation. J. Geo- progress. J. Geophys. Res., 103, 14 169–14 260. phys. Res., 109, C05023, doi:10.1029/2003JC002082. Megann, A., and A. New, 2001: The effects of resolution and Stockdale, T. D., D. Anderson, M. Davey, P. Delecleuse, A. Kat- viscosity in an isopycnic-coordinate model of the equatorial tenberg, Y. Kitamura, M. Latif, and T. Yamagata, 1993: In- Pacific. J. Phys. Oceanogr., 31, 1993–2018. tercomparison of tropical ocean GCMs. World Climate Re- Pedlosky, J., 1996: Ocean Circulation Theory. Springer-Verlag, search Programme, WCRP-79, WMO/TD-No. 545, 90 pp. 453 pp. Sun, D.-Z., J. Fasullo, T. Zhang, and A. Roubicek, 2003: On the Putman, W., D. Legler, and J. O’Brien, 2000: Interannual vari- radiative and dynamical feedbacks over the equatorial Pacific ability of synthesized FSU and NCEP–NCAR reanalysis cold tongue. J. Climate, 16, 2425–2432. pseudo-stress products over the Pacific Ocean. J. Climate, 13, Swenson, M. S., and D. V. Hansen, 1999: Tropical Pacific Ocean 3003–3016. mixed layer heat budget: The Pacific cold tongue. J. Phys. Qiao, L., and R. H. Weisberg, 1997: The zonal momentum bal- Oceanogr., 29, 69–81. ance of the equatorial undercurrent in the central Pacific. J. Vialard, J., C. Menkes, J.-P. Boulanger, P. Delecluse, and E. Guil- Phys. Oceanogr., 27, 1094–1119. yardi, 2001: A model study of oceanic mechanisms affecting Qiu, B., and S. Chen, 2004: Seasonal modulations in the eddy field equatorial Pacific sea surface temperature during the 1997–98 of the South Pacific Ocean. J. Phys. Oceanogr., 34, 1515– El Nin˜ o. J. Phys. Oceanogr., 31, 1649–1675. 1527. Weaver, A. T., J. Vialard, and D. L. T. Anderson, 2003: Three- Reynolds, R. W., and T. M. Smith, 1994: Improved global sea sur- and four-dimensional variational assimilation with a general face temperature analysis using optimum interpolation. J. circulation model of the tropical Pacific Ocean. Part I: For- Climate, 7, 929–948. mulation, internal diagnostics, and consistency checks. Mon. Smith, W., and D. Sandwell, 1997: Global seafloor topography Wea. Rev., 131, 1360–1378. from satellite altimetry and ship depth soundings. Science, Yang, S.-K., Y.-T. Hou, A. J. Miller, and K. A. Campana, 1999: 277, 1956–1962. Evaluation of the earth radiation budget in NCEP–NCAR Stammer, D., C. Wunsch, I. Fukumori, and J. Marshall, 2002a: reanalysis with ERBE. J. Climate, 12, 477–492. State estimation improves prospects for ocean research. Eos, Yu, Y., W. Emery, and R. Leben, 1995: Satellite altimeter derived Trans. Amer. Geophys. Union, 83, 289. geostrophic currents in the western tropical Pacific during ——, and Coauthors, 2002b: The global ocean circulation during 1992–1993 and their validation with drifting buoy trajectories. 1992–1997, estimated from ocean observations and a general J. Geophys. Res., 100, 25 069–25 085. circulation model. J. Geophys. Res., 107, 3118, doi:10.1029/ Zhang, K., and J. Marotzke, 1999: The importance of open bound- 2001JC000888. ary estimation for an Indian Ocean GCM data synthesis. J. ——, K. Ueyoshi, A. Köhl, W. G. Large, S. A. Josey, and C. Wun- Mar. Res., 57, 305–334.

Unauthenticated | Downloaded 09/30/21 07:11 PM UTC