Ocean circulation for the Indonesian seas driven by tides and atmospheric forcings: Comparison to observational data



[1] A 1/36° high-resolution nested structure ocean model for Indonesian seas that combines atmospheric and tidal forcings (hereafter referred to as IP3-tide) has been developed based on the modified Princeton Ocean Model (POM). Prior to model application, in 2004, we analyzed the IP3-tide by using observational data, derived data, and a global atlas model. A comparison between IP3-tide with 13 tide gauge points and 11 T/P points resulted in 76–92% certainty. Correlation of temperature from scattered depths and points between the model and XBT data reached 97.5% agreement. The modeled velocity successfully captured the low and high frequency variability shown in INSTANT mooring and TAO/TRITON data. Explicit simulation of tidal processes by regional ocean circulation model improved the representation of circulation in the Indonesian seas. At the tidal frequencies, vertical mixing is increased due to the impact of baroclinic tides and horizontal mixing is enhanced by presence of barotropic tidal motion. Enhanced mixing is responsible for eroding the salinity maximum found in the water masses advected from the Pacific Ocean. On the other hand, seasonal variability changes the vertical density structure of water column, which influences the distribution of internal tidal waves. These results demonstrated the importance of explicit tide simulation by regional ocean circulation model for correct presentation of ocean circulation structure and its variability in the Indonesian seas.

1. Introduction

[2] Indonesia's location has a characteristic influence on global climate through complex sea-atmosphere interactions and influences the ocean circulation system [Macdonald and Wunsch, 1996; Gordon and Fine, 1996; Wright, 2001]. Sea surface temperature (SST) variability in the Indonesian seas is closely related with El Niño-Southern Oscillation (ENSO) and the Indian Ocean Dipole phenomena, and SST can be used as a good proxy for predicting the global climate of ENSO [Nicholls, 1983] and Dipole mode [Saji et al., 1999]. Another characteristic of the Indonesian seas is reflected in the work of Sprintall et al. [2009], indicating ∼15 Sv of mater mass is transported to the Indian Ocean. For this reason, the Indonesia seas are recognized as a “linkage” between the Pacific and the Indian Oceans [Gordon et al., 2003], between the atmosphere and the ocean [Giannini et al., 2007], between the past and the modern oceans [Cane and Molnar, 2001], and likewise between biotic diversity and abiotic systems [Barber et al., 2000].

[3] This study creates the basis for launching the application of regional model for the real-time predictions in the Indonesian seas. We believe that reliable up-to-date ocean information may benefit governments, scientists, the private sector, and the community as a whole. Additionally, this information may improve ocean prediction quality. However, the topographic complexity and large size of the Indonesian seas make this area difficult to be entirely modeled in a high-resolution model. At a minimum, the major factors that should be addressed to properly simulate ocean conditions in this region are wind field [Godfrey, 1996] and tide [Ffield and Gordon, 1996; Koch-Larrouy et al., 2007] and a proper treatment of boundary conditions that respects the mean flow currents from the Pacific to Indian Ocean [Sprintall et al., 2009]. The inclusion of these two forcings may limit an Ocean General Circulation Model (OGCM) made for the area around Indonesia, especially in terms of too-coarse forcing resolution and/or poor quality.

[4] The limitations of wind input/atmospheric forcing data are less complex and more related to technical and satellite sensor technology problems, which are improving with time. In contrast, the drawback for tidal input is that it is quite problematic. An effort to improve the tidal component was conducted by Matsumoto et al. [1995] using TOPEX/Poseidon (T/P) altimetry data and by Tsimplis and Woodworth [1994] using coastal tide gauge data. Both of them found large differences between their results and recorded data in Indonesian seas. This finding was reinforced by a statement from Egbert and Ray [2001], who studied M2 tidal energy dissipation and showed a high degree of disagreement between their solutions in the Indonesia seas. Due to the lack of TOPEX data, the lack of sufficient ocean depth in shallow areas and the complex array of islands make tidal modeling a difficult and problematic task. Many tidal prediction models that are mostly derived from altimetry data are not able to cover the shallow part of the Indonesia seas, despite the fact that half of the Indonesian seas and coastal areas contribute to tidal motions [Mathers and Woodworth, 2001; Ponte and Gaspar, 1999; Vinogradov and Ponte, 2010].

[5] In previous high resolution modeling study, Robertson and Ffield [2008] succeeded in modeling barotropic and baroclinic tides in the Indonesia seas using a high-resolution ROM model and used it to discuss the interaction and transfer of energy among tidal constituents [Robertson, 2011] and tide induced mixing processes at the INSTANT mooring locations [Robertson, 2010]. However, another dominant forcing is discussed separately in the work of Metzger et al. [2010] that analyzed the pathway of ITF by using a global high-resolution HYCOM model driven by atmospheric forcing. Both of these high-resolution studies described the Indonesian seas from different interesting angles by using a single forcing, either tides or atmospheric forcing. Recently, Arbic et al. [2010] added a tidal forcing to the model of Metzger et al. [2010], but their results focused on global ocean scale and did not describe the Indonesian seas in particular. Build upon to these studies, we developed a nested structure for a regional ocean model that is driven by not one but two forcings: tide and atmosphere with horizontal resolution higher than that previously reported. The tide forcing is explicitly incorporated in the model such that the mixing process is not artificially parameterized, making this study different from those of Koch-Larrouy et al. [2007, 2010]. By combining the two main forcings, the ocean conditions in Indonesia can be better represented and analyzed. Moreover, we validate this model by using various available data sets.

[6] In this study, the tidal model is introduced later in the third level of the nesting structure (S. M. Varlamov et al., manuscript in preparation, 2011), whereas the first and second level are fully driven by wind stresses and heat and salt fluxes [Miyazawa et al., 2009]. This method is an adaptive way of handling specific target studies on a regional scale. Yet, modeling techniques for treating boundary conditions in tidal-embedded nested models still remain an interesting challenge that is not found in global tidal and atmospheric models. We discuss this matter in section 2, together with detailed descriptions of the model used and notes on new modifications of barotropic divergence. Observational data utilized for the comparison are outlined in section 3. Next, section 4 provides the results of the comparison, whereas the simulated internal tide in the Celebes and Sulu Seas is discussed in section 5. In the final section, we summarize our findings with a brief discussion.

2. Model Description

[7] The Japan Coastal Ocean Predictability Experiment (JCOPE) code developed by the Japan Agency for Marine-Earth Science and Technology (JAMSTEC) was used in this study. The JCOPE model was built specifically to address the complexity and variability of eddies in the Kuroshio system. This code is based on a three-dimensional primitive equation ocean general circulation model known as the Princeton Ocean Model (POM) with a generalized coordinate of sigma [Mellor et al., 2002]. The JCOPE project succeeded in reproducing the variations of the Kuroshio and mesoscale eddies in the northwestern Pacific Ocean [Kagimoto et al., 2008; Miyazawa et al., 2008, 2005]. Most recently, for downscaling purposes, the JCOPE model has been improved to include tidal forcing in its prediction [Guo et al., 2010]. The present version of the model can better represent finer distributions of temperature and salinity by using both bi-harmonic diffusion and flux-corrected transport advection schemes [Miyazawa et al., 2009]. Moreover, the 4th order pressure gradient scheme for the generalized sigma coordinate model was implemented to improve the estimation of internal pressure gradient [Berntsen and Oey, 2009]. However, model version for the Indonesian seas did not include river discharge input which is quite important in introducing lower salinity input and creating coastal front around the Indonesian seas.

2.1. Model Domain

[8] Here, a 1/36° nested grid ocean model with a three-level nesting structure is used. The outermost model encompasses most of the Indian and Pacific Oceans (hereafter known as the IP1 model) and the next-most outer model (hereafter known as the IP2 model) covers Southeast Asia and the Western Pacific. The outer models are necessary to provide the lateral boundary values of the prognostic variables for the higher horizontal resolution model that covers the Central and Eastern part of the Indonesian seas (8°N–13°S, 109°E–134°E; hereafter known as the IP3 model).

[9] ETOPO2v2 and GEBCO bathymetric data were combined to create topographic data with additional bathymetric corrections in certain areas such as the Java Sea (Sea chart 99, printed II, 2008); the Makassar Strait (Sea chart 121, printed IX, 2005); the Nusa Tenggara and Flores Seas (Sea chart 111, printed XIV, 2008) and the Banda, Seram and Halmahera Seas (Sea chart 363, printed V, 2007). These bathymetric charts (scale 1:1000000) were provided by the Hydro-Oceanographic division of the Indonesian Navy. This was an effort to minimize the shortcoming of ocean models as seen in many previous studies that cite the low accuracy of topography in the Indonesian seas [Robertson and Ffield, 2008; Metzger et al., 2010]. To create a proper land/sea mask, we used GEBCO One-Minute data to precisely sketch small islands that can be depicted using the desired grid resolution. An isobath of less than one meter was used as the coastal line, while the shallowest depth was set at 5 meters and the thinnest vertical sigma layer as 15 cm. Special treatment at the boundaries was used to prevent large volume flux differences from the coarser IP2 and tidal OTIS models. We set the depth of the two outer grids as similar to that of the IP2 model and modified the 3rd–20th grids to match the OTIS bathymetric data. In between the transition bathymetry from IP2-OTIS-IP3 topography, we applied Gaussian smoothing with a factor of 0.25. Through the use of these artificial modifications, we were able to prevent model blow-up at boundaries.

2.2. Initial and Boundary Conditions

[10] The nested models were then spun up for approximately twenty years for IP1, five years for IP2 and nearly one year for IP3 from the initial conditions of no motion using monthly mean temperature and salinity (TS) fields obtained from the 2001 World Ocean Atlas (WOA01). Six-hourly surface forcings were then applied to the model's domain as a real surface forcing beginning in October 1992 until the end of the simulation in June 2010 for IP1 and IP2. In IP3 model, surface fluxes were estimated using six-hourly NCEP/NCAR reanalysis data during spin-up process in year 2003. A slightly different treatment was applied for IP3-tide model, where tide forcing was embedded two months before spin-up period finished. Then, IP3-tide model initially experienced ten months spin-up with atmospheric forcing only and then two months spin-up with atmospheric and tidal forcings.

[11] The monthly mean TS of WOA01 was used for TS boundary conditions in IP1, while the elevation and velocity of the outer grid were set to zero, but we let the barotropic mode fluctuate according to radiation boundary conditions. However, for the other fine models, IP2 and IP3, all prognostic variables except for two turbulent properties in the coarse model were linearly interpolated onto the boundary grid points of the fine model. We saved a daily boundary value at the grid points of the nested models. IP2 and IP3 models were high-resolution models with a horizontal grid spacing of 47 layers in the vertical section.

[12] In the case of tidal forcing for IP3, we treat boundary conditions in a different way (Varlamov et al., manuscript in preparation, 2011). A different frequency between interpolated daily boundary values with tidal forcing becomes a primary concern. Although six-hourly forcings of NCEP/NCAR reanalysis data may contain broad band frequencies, tidal frequencies can only be resolved by improving additional interpolated boundary values with the introduction of tidal forcing along boundaries. The traditional method to introduce barotropic tide is simply by adding tidal elevation at the boundary to force the model. Yet, here we improved the modeled elevation by using smooth Newtonian relaxation.

equation image

[13] In the equation (1), ζ*n+1 is the first guess-modeled sea level at the open boundary as estimated by the continuity equation, and αE is the sea level relaxation coefficient with a value less than a unit. However, merely inputting tidal elevation along the boundary can break the mass conservation at the boundaries and further trigger suspicious vertical velocity, leading to discontinuous TS distribution. The way to minimize this problem is by introducing a correction term of lifted/lowered elevation in the barotropic velocity boundary condition, the so-called velocity adjustment term, as seen in the following equation.

equation image

[14] To restore the volume conservation, the boundary volume inflow could be corrected on the corresponding value needed to balance the boundary cell “relaxation” volume change Δζ . dS, where dS is a modeled cell surface area, and Hb and dlb are the undisturbed depth and cell size along the open boundary. This partial solution could work if data on the tidal velocity transport along the boundaries were available. As such, the mass conservation along boundary would be maintained. The rest of the problem is due to the difference between model and tidal bathymetry. A large topographic difference can cause error in baroclinic distributions. Therefore, it is better to select a tidal model that provides tide elevation, velocity transport and topographic data. This is the reason that we chose the low-resolution global inverse tidal model (OTIS) by Egbert and Erofeeva [2002].

2.3. Forcing and Bulk Formulae

[15] The wind stress and surface heat flux are calculated using the six-hourly NCEP/NCAR reanalysis data [Kalnay et al., 1996] with the bulk formulae mentioned by Kagimoto et al. [2008]. The downward shortwave radiation is kept the same as in the original NCEP/NCAR. Downward penetration of shortwave radiation to the water column, based on the work of Paulson and Simpson [1977], is adopted to improve the accuracy of the SST.

[16] No assimilation technique is used in this model; however, salinity at the water's surface is restored to the monthly mean climatology data using a horizontal high resolution of 1/4 degree (World Ocean Atlas 2001) and a time scale of 30 days. Here, we set 30 days of relaxation time to induce a surface salinity flux from the shallow water to the bottom. This step is required to prevent long-term surface salinity drift due to inadequate freshwater flux. The entire IP1 domain is nudged by a weak relaxation of 150 days, whereas the IP2 boundary is relaxed by 180 days from 100 m to the bottom (not at the surface) using monthly climatological data. No relaxation is applied in the IP3 model.

2.4. Modification on the Contribution of “External” Vertical Velocity

[17] We assume that the barotropic ocean tide satisfies the tidal equations of Laplace. The gradient of elevation in time is equal to the divergence of barotropic (depth) average velocity, or equation image as described in the mass conservation equation below.

equation image

[18] In POM, the contribution of the gradient elevation is found in the calculation of vertical velocity. However, a mixture between the dynamic solution of the baroclinic component and the topographic-related solution of gradient elevation in time does not seem as consistent. As a result, we replaced the time elevation difference ∂ζ/∂t with the divergence of barotropic velocity in equation (2). Therefore, vertical velocity is calculated from the dynamics barotropic and baroclinic divergence.

equation image

[19] Second, we redefined the contribution of snapshot-smoothed elevation to time-mean averaged-external-elevation in the implicit scheme. The reason for this is because in the internal mode, baroclinic velocity is calculated from the time average of barotropic velocity, whereas the contribution of elevation is a snapshot-smoothed value. If one uses the strong smoothing technique, then mass conservation can be distorted. By changing the elevation to a time-average value, the baroclinic mode is determined from the mean of the forward and backward barotropic modes.

equation image

[20] To guarantee the effect of the new modification in the vertical velocity calculation, the forward time-step elevation for the implicit calculation is determined from the leapfrog method using the divergence equation itself.

equation image

2.5. Tidal Components

[21] In the IP3-tide model, we use the time-independent amplitude and phase of velocity transport of the global inverse tidal model (OTIS) developed by Egbert and Erofeeva [2002]. The tidal module is embedded in the IP3 model and forecast at every 2-D time step using astronomical parameters and the solution of the NAO99 tidal model developed by Matsumoto et al. [2000]. For the purpose of study, we select only 4 primary tidal components, namely M2, S2, K1, and O1. These components are found to be the four major components drive tidal forcing in the Indonesian seas. These components represent the four major components that drive tidal forcing in the Indonesian seas [Robertson and Ffield, 2008].

[22] The difference between the tidal and non-tidal models comes not only from the type of forcing but also from the boundary conditions. In the non-tidal model, all prognostic values are directly interpolated into the nested model, and the values specified in the nested model are similar to those in the outer IP2 model. However, in the IP3-tidal model, the tidal elevation and volume transport given at the boundary need to be “compromised” with the boundary value given by the IP2 model. Thus, the boundary values in the IP3-tide model are not identical to those in the IP2 model. The difficulties in setting up the combination of tide and atmospheric forcings are due to the following: (1) setting up the topography close to the boundary, (2) matching barotropic and baroclinic influences from the tide in the outer nested model, (3) selecting a parameter for viscosity and diffusion, (4) choosing a type of viscosity and diffusivity formula and (5) controlling vertical mixing for stable calculations.

3. Observation Data

[23] We use only ocean data in year 2004 that are available online and can be accessed by the public for free (Figure 1). The year 2004 was selected due to the availability of continuous INSTANT velocity data used for model validation and analysis. In this simulation, all components of prognostic variables were compared against observational and derived data, with the exception of vertical velocity. Here, we compared the sea level elevation with tide gauge data from the University of Hawaii Sea Level Center. Velocities are checked using INSTANT and TAO/TRITON data in the western-most Pacific, whereas temperature and salinity are assessed using the following observational data from R/V MIRAI in 2004 (http://www.godac.jamstec.go.jp/cruisedata/mirai/e/index.html), XBT data (http://cawcr.gov.au/bmrc/ocean/JAFOOS/BOM_xbt.html; http://www.nodc.noaa.gov/OC5/SELECT/dbsearch/dbsearch.html) and fixed TS TAO/TRITON data (http://www.pmel.noaa.gov/tao/data_deliv/deliv.html). Here, we would like to thank the organizations that provide their data for research purposes.

Figure 1.

Model domain and observational data. The small square symbol is the tide gauge (Green-TG) and the Topex/Poseidon (Purple-TP) point. The filled box is the point at which the data exist during the model time integration (except Darwin stat., TG13), and the open box indicates the point at which the four main harmonic components are compared. The orange triangle is the TAO-TRITON point, and the red diamond symbolizes the location of the INSTANT mooring data. The Mirai cruise data are located in the Indian Ocean and are symbolized with an open yellow circle. The grey contours show the local bathymetric data superimposed over the GEBCO+ ETOPO2 data.

3.1. University of Hawaii Sea Level Data Center

[24] The time series of the sea level and derived constant harmonic components of tide that are compared with the IP3-tide model are taken from University of Hawaii Sea Level Center (http://ilikai.soest.hawaii.edu/uhslc/datai.html). We took hourly “Research quality” data and processed them without any filtering processes. Both the model and the data are in the GMT time zone format, so we directly compared them after we normalized the data into zero means.

[25] Many of the tide gauge stations for our domain have been provided by BAKOSURTANAL of Indonesia, Dept. Survey/Mapping of Malaysia, and Ocean Surveys Div. of the Philippines via the University of Hawaii Sea Level Center website. However, because we limit validation to the year 2004, only a few were able to be used in this study and thus were directly compared to each other. For the other available data in the IP3 domain that were outside our simulation time period, we derived those data into harmonic components and compared them with the modeled harmonic components. We only compared the tidal harmonic results with data from the tide gauge after the year 1973.

[26] We compared the time series of sea level elevation from 8 stations (TG12; Ambon, TG10; Benoa, TG1; Bintulu, TG7; Davao, TG3; Kinabalu, TG4; Sadakan, TG8; Surabaya and TG5; Tawau) within one month (Figure 1). We evaluated the IP3-tide model to those stations in different time periods (Figure 2). Yet, the distribution of tide gauges was concentrated in the northern part of Kalimantan Island (Serawak and Sabah provinces of Malaysia) and in East Java, when most other parts inside the Indonesian seas are not covered by tide gauges. Therefore, we added the result of Topex/Poseidon harmonic analysis of Robertson and Ffield [2008] in locations where limited number of tide gauges were found. Topex/Poseidon's locations provided sea level information in deep sea, whereas most of tide gauge locations were located in shallow zone area. We also compared IP3-tide elevation results with derived amplitudes and phases using the tide gauge data from 5 stations (TG2, TG6, TG9, TG11, TG13) in Table 1 and the Topex/Poseidon harmonic analysis of Robertson and Ffield [2008] in Table 2. In the tide gauge stations, harmonic constants of amplitudes and phases are derived from randomly-selected one month data.

Figure 2.

Normalized SSH of IP3-tide results and tide gauge data. Only a one-month period is presented for each station. The time period is different for each station. The unit for SSH is set to meters, and r2 is the correlation value.

Table 1. Time-Independent Amplitude and Phase Coefficients of the Amplitude and Phase Coefficients for the Four Main Tidal Components at Selected Tide Gauge Stations
LocationTide ModelsM2S2O1K1
 Tide gauge0.1787.780.09119.590.32141.810.35191.53
TG2; MiriOTIS0.1998.80.08130.50.325157.40.39201.9
4.391N; 113.972EIP3-tide0.2092.50.09131.20.32144.70.37199.55
 Tide gauge0.17359.750.1047.980.25150.740.22201.03
TG6; JojoOTIS0.1973.00.15109.50.26164.60.28200.9
6.067N; 121.0EIP3-tide0.19341.390.1228.10.23135.230.25180.0
 Tide gauge0.4665.260.22111.920.19146.610.30175.94
8.116S; 114.383EIP3-tide0.4569.680.21120.00.18146.90.28175.36
 Tide gauge0.5822.380.4396.050.13144.170.18172.6
TG11; PrigiOTIS0.5836.30.3191.70.13157.40.21167.0
8.283S; 111.733EIP3-tide0.5727.10.3098.20.13144.90.22165.37
 Tide gauge1.85236.390.96297.290.34178.580.59199.96
TG13; DarwinOTIS1.70256.80.79300.10.31192.50.56205.8
12.466S; 130.850EIP3-tide1.91261.50.98335.90.34194.10.56214.5
Table 2. Similar to Table 1, but at Selected Topex/Poseidon (T/P) Crossover Stationsa
LocationTide ModelsM2S2O1K1
 T/P Crossover0.131400.05780.452010.22165
5.96S; 114.8EIP3-tide0.23156.90.0171.10.36206.70.24161.8
 T/P Crossover0.83540.451070.141610.24169
9.75S; 119.05EIP3-tide0.8042.10.45110.120.14145.120.24169.9
 T/P Crossover0.462770.323260.161160.22155
2.15S; 119.05EIP3-tide0.43255.60.33315.40.16116.40.22146.4
 T/P Crossover0.621200.241670.191720.29184
9.75S; 127.55EIP3-tide0.59105.40.23166.580.16158.50.27182.0
 T/P Crossover0.582900.343340.131160.18133
2.15N; 120.467EIP3-tide0.54272.10.32317.50.13106.10.16134.2
 T/P Crossover0.431890.142520.272300.27217
9.76S; 133.21EIP3-tide0.52168.340.17240.50.23222.10.23201.5
 T/P Crossover0.572910.333340.171270.13112
@2.033N; 123.28EIP3-tide0.54273.680.31319.00.1594.90.2138.8
 T/P Crossover0.261600.112490.141550.22170
2.05S; 127.55EIP3-tide0.24144.20.12248.90.14130.160.24155.35
 T/P Crossover0.581390.181990.191740.28187
5.9S; 131.82EIP3-tide0.58125.70.18204.30.17159.30.26182.1
 T/P Crossover0.592900.333340.131170.18135
5.93N; 121.88EIP3-tide0.58272.50.33319.60.11104.70.14130.4
 T/P Crossover0.492880.213180.131750.1987
2.016N; 128.967EIP3-tide0.47269.90.23308.40.1364.20.286.65

3.2. FES Global Tidal Atlas Model

[27] The FES2004 model is assimilated from the tide gauge and satellite (T/P and ERS) sea level analysis. The FES2004 atlas is now the default correction of the distributed T/P altimetry data (http://www.legos.obs-mip.fr/en/share/soa/cgi/getarc/v0.0/index.pl.cgi?contexte=SOA&donnees=maree&produit=modele_fes). The FES2004 tidal chart from LEGOS has a higher resolution than does the OTIS model that we use for our tidal boundary conditions.

[28] The amplitude and phase distributions of the IP3-tide model are compared with those of the interpolated 1/8° FES2004 model, which has higher resolution than does the tidal forcing data from the OTIS (Figure 4). We use bilinear interpolation to interpolate 1/8° FES2004 model data into the 1/36° FES2004 model. Only M2 and K1 harmonic components are used here to evaluate the IP3-tide model.

3.3. INSTANT Data

[29] INSTANT was a pioneer project to measure the mean and variability of the Indonesia Throughflow (ITF). Five nations participated in the INSTANT project: Indonesia, France, the Netherlands, the USA and Australia. A few years ago, the INSTANT velocity data were released for research purposes (http://www.marine.csiro.au/∼cow074/instantdata.htm/). Here, we only use data from four of the five stations with data from 2004 (I1; Makassar, I2; Lombok, I4; Ombai and I3; Timor Strait mooring stations) (Figure 1). The time series of 50 m, 150 m, 350 m and 750 m have been quality controlled.

[30] We used mooring data from INSTANT to validate the zonal and meridional velocity components in our model. Different depth in a small strait between the model and real topography may result in big differences in comparison. Thus, we decided to compare zonal and meridional velocity both directly and indirectly. For the direct comparison, we compared time series data of the zonal and meridional upper (50 m) and deeper (750 m) parts with the IP3-tide results. For the indirect comparison, we compared the tidal ellipse of the main four harmonic components (M2, S2, O1, and K1) at 150 m and 350 m.

3.4. XBT Data

[31] Scattered temperature XBT data of Indonesian seas can be found in the Bureau of Meteorology (BoM), Australia, and in the World Ocean Database Center (WODC). We found the BoM data to be well organized with respect to XBT line, especially PX-2 (zonal line within a few degrees of 8°S) and IX-22/PX-11 (meridional line within a few degrees of 126°E) for the case of Indonesia. However, the WODC has salinity data for some stations in addition to temperature. But, temperature data identified as being of good quality are limited. Thus, we decided to compare the modeled temperature using BoM data instead of WODC data. We used more than 45000 distributed temperature data points for the year 2004.

3.5. TAO-TRITON Data

[32] Unfortunately, only one of two available TAO-TRITON stations in the IP3 domain can be used to validate our model which is located at 2°N; 130°E, and it only has data of the first and end of 2004. In time series, we compare the TS profile with TAO/TRITON data deployed at the southwestern boundary of the Pacific Ocean. The data have 12 layers of depth, from the surface to 750 m. These data have been quality controlled. The near-surface (1.5 m) velocity data are used to validate IP3 surface velocity components.

3.6. Mirai Cruise Data

[33] The Mirai cruise in July 2004 along the south of the Java Sea provided transectional XCTD data of 0.5-degree intervals. The XCTD data provide the temperature and salinity profile. Here, we use their data from 115° to 120°E, resulting in 10 points used in this paper. We averaged the analysis of those points because all had similar types of water mass properties. These 10 points were collected on July 17–18, 2004.

4. Comparison Between Model and Observation/Derived Data

4.1. Tidal Disturbance on Sea Surface

[34] Sea level elevation was compared against the tide gauge stations with observational data from 1972 to recent. Three old stations in Balikpapan, Bajor and Lungsurannaga (all on the east coast of Kalimantan Island) were ignored in our harmonic analysis study. To reduce truncation errors in the Julian calendar, a modified-Julian day calendar is used to analyze tidal harmonic analysis. The application of the modified-Julian calendar is effective for January 1972 and later. For this reason, we did not include the three old stations on the east coast of Kalimantan. In our results, 13 stations were available and were used to evaluate the sea level elevation results. Eight of the 13 stations had data within our simulation period, whereas the rest were out of the IP3-tide simulation time period. Then, the tidal harmonic analysis of main four harmonic components was used to analyze sea level data for other five remaining stations. To verify the elevation's skill in the open sea, we took the T/P harmonic component study results of Robertson and Ffield [2008]. We selected 11 points from their paper and used them as additional elevation comparison points.

[35] To directly depict sea level elevation results, we compare the time series of sea level elevation from the IP3-tide model to the tide gauge data. A different time is selected for each of the eight stations, and a one-month duration is plotted in Figure 2. The results show a 76% correlation for Surabaya, more than 80% for Bintulu and Sadakan and more than 90% for the Kinabalu, Tawau, Davao, Benoa and Ambon stations. However, in most cases, this correlation is not a good metric to justify the results.

[36] Next, the results of the harmonic analysis are presented in table 1 and table 2. In table 1, we compare the amplitude and phase of four tidal components from the observational data, FES2004, OTIS and the IP3-tide model. In Table 2, the T/P crossover harmonic components from the work of Robertson and Ffield [2008] are used. A summary of the discussion from the two tables is given in Figure 3, which explains the RMS difference of amplitude and phase between the IP3 model and the others. As is shown in the figures, the amplitude of M2 performed less well than did the other components (RMS difference = 9 cm), but this shortcoming is understandable because the M2 component is the dominant component in this region. Upon closer examination, the mean total variability of the M2 amplitude is 51 cm, indicating the confidence of the M2 component to be more than 83%. In contrast, O1, K1 and S2 had a good correlation, at 93%, 90% and 89%, respectively. By combining all stations and components, the RMS amplitude difference is 6 cm and 33° for the phase, with regard to results from the barotropic model l; OTIS and FES2004 contribute considerably to the mean RMS differences. By excluding FES2004 and OTIS, the RMS differences are reduced to 4.7 cm for amplitude and to 18.6° for phase. Calculating the total RMS difference for amplitude over the mean total variability of the M2 component implies the confidence of this model to be up to 92%.

Figure 3.

Chart of RMS error between the IP3-tide and the other tidal data/models. Amplitude is in meters (bar), and phase is in degrees (dotted line). A “total” means the average of the M2, S2, O1, and K1 components, and “N” is the number of samples.

[37] Point-by-point comparisons have been provided above. We continue to compare the IP3-tide results with those of the high-resolution global atlas model, FES2004. We take a look at the horizontal distribution of M2 and K1 as the major semidiurnal and diurnal components, respectively. We identified different features of the M2 amplitude in the Celebes Sea upon comparing IP3-tide model with FES2004 (Figure 4b). Internal tide signals appear in the surface as disturbed sea surface height. Differences in phase are also observed in some regions. However, the difference among the two models results in a spatial correlation of 0.96 in amplitude and 0.91 in phase for the M2 component, whereas for the K1 component the spatial correlation is 0.94 and 0.91 for amplitude and phase, respectively. Excluding the harmonic analysis of OTIS and FES2004 results in a better performance for IP3-tide but does not imply that the IP3-tide model outperforms the barotropic component. Instead, this result implies that the IP3-tide model successfully resolves horizontal barotropic distribution. Because FES2004 is a good barotropic tidal model, IP3-tide can perform similarly well and, moreover, baroclinic features are also captured in this model.

Figure 4.

Co-amplitude and co-phase of the M2 tide components between (a and c) FES2004 and (b and d) IP3-tide models. Amplitude is indicated with a shaded color, and the units are meters. Co-phase is indicated with the contoured line, and the units are degrees.

[38] In Figure 4, the semidiurnal tide enters the Indonesian seas from the Pacific Ocean through the Makassar Strait and the Halmahera Sea, while the other M2 component is propagating eastward from the Indian Ocean, enters the Indonesian seas through small straits and the Banda Sea, and then continues westward towards the Java Sea. This M2 tide meets with the M2 tide from the Pacific Ocean to the south of the Makassar Strait and the Halmahera Sea. To the south of the Java Sea, the M2 tide bends counter-clockwise north of Australia and forms a low amplitude at the amphidromic point. While in its diurnal component, K1 could easily pass the Indonesian seas with a type of propagation from the Pacific to the Indian Ocean through the Celebes Sea and the Sulawesi-Halmahera Sea. The diurnal K1 tide, which enters from the Celebes Sea and continues to the Makassar Straits, either propagates into the Java Sea and meets with another K1 component originating from the South China Sea or meets with the K1 tide that disrupts the Sulawesi-Halmahera Sea at the Banda Sea. From the Banda Sea, this K1 tide flows to the Indian Ocean. The propagation of K1 is shown to be simpler than that of the M2 component and, in terms of amplitude, the diurnal amplitude of K1 is smaller than the semidiurnal M2.

4.2. Low and High Frequencies of Horizontal Velocities

[39] For a tidal model, accuracy in sea level with respect to the observational data does not guarantee the accuracy of velocity data. Yet, the accuracy of velocity data cannot be achieved without producing good sea level results. After showing a good result for sea level elevation, the time series of hourly mean IP3-tide velocity components is examined and compared with the INSTANT observations at 50 m and 750 m and surface TAO/TRITON. In Figure 5, the surface velocity of the IP3-tide model is shown to be capable of reproducing low and high variation at 2°N;130°E. There is no observational data from February–July 2004, but the trend of the velocity component is good in January and August–December. Almost throughout the entire year, the zonal velocity Uh=−1.5 m is negative. This condition is different for meridional velocity, which is negative in March and positive in August. This implies that this position is the western boundary of the South Equatorial Current that has its counterclockwise eddy in March and its clockwise eddy in August. Due to the existence of this eddy, a high salinity maximum of 35.4 psu turns back to the Pacific Ocean and a lower salinity maximum (34.6–34.8 psu) is detected in the Seram and Halmahera Seas.

Figure 5.

One-year time series of (top) zonal and (bottom) meridional velocity components at a 1.5 m depth between TAO/TRITON data (grey dots) and the IP3-tide model (black line). Units are m/s.

[40] In Figure 6, at a mixed layer depth of 50 m, the low frequency of the IP3-tide model is seen to reproduce with acceptable agreement the findings from the observational data. Similar processes are also observed from the meridional component velocities. The meridional velocities in Lombok and Ombai show similar signals, indicating that these two straits are receiving same influence from the outside. Strong positive meridional flow during January–April in the Lombok, Ombai and Timor straits was not detected in Makassar. However, this strong positive meridional flow in May–June observed in Lombok and Ombai can be detected in Makassar with a lag of less than one month. This positive meridional flow may be triggered by the Kelvin wave in the south of Java [Sprintall et al., 2000; Drushka et al., 2010]. Low frequencies of both zonal and meridional velocity components are identical in the Lombok, Ombai and Timor straits, which implies the homogeneity of the inflow-outflow pattern in these straits. Furthermore, we see that the flow variations in these straits are governed by different mechanisms than that in the Makassar Strait. Basically, the base meridional velocities in all straits are comparable (Vh=−50 m ∼ −0.5 m/s) in magnitude, and they are mainly flowing to the south. Therefore, at the surface (50 m depth), outflow transport in the ITF outlets (Lombok, Ombai and Timor straits) are not fully dependent on single inflow ITF transport in the Makassar Strait.

Figure 6.

One-year time series of (right) zonal and (left) meridional velocity components at a 50 m depth between INSTANT data (green dot) and the IP3-tide model (red dot). A band pass filter of 7–365 days is applied to the IP3 results and the INSTANT data and shown in the black line and grey dots, respectively.

[41] In Figure 7, we compare the time series of zonal and meridional velocities of the IP3-tide model below the thermocline layer. We can see that high frequencies are playing a more important role at this depth. Only in the Makassar and Ombai straits does the intraseasonal timescale show some fluctuation, but for the Lombok and Timor Straits, the intraseasonal and seasonal variability is small. The accuracy of the IP3-tide model reduces with depth; however, IP3-tide can show tidal signals at a depth of 750 m, as is also found in the raw observational data (green dots). A large discrepancy is observed for the Timor Strait, where the model shows positive meridional flow while observational data show negative meridional flow. Positive meridional flow is also found in other straits, which may be due to model-setting problems such as gentle slope topography, interaction between bottom stress and vertical velocity, and viscosity parameters. Of note, Arbic et al. [2010] stressed the importance of an additional bottom dissipative term for tidal models that has not yet been applied in this model.

Figure 7.

One-year time series of (right) zonal and (left) meridional velocity components at 750 m depth between INSTANT data (green dot) and the IP3-tide model (red dot). A band pass filter of 7–365 days is applied to the IP3 results and the INSTANT data and shown in the black line and grey dots, respectively.

[42] Figure 8 shows a tidal ellipse comparison between tidal frequencies from the INSTANT data and the model results. From a qualitative comparison, we see that the orientation of the tidal ellipse between the INSTANT data and the model has been diverted to some degree. However, the model showed consistency with regard to the changing variability in amplitude and angle from 50 m to 750 m, where the observational data and the model depicted the variability with reasonable similarity. The differences between the model results and the observed semi-major of all components are generally less than 10 cm/s. However, it is not surprising that the model should have such discrepancies from the observational data because the mooring is located in straits that have a strong current and complex topography or sill that is not properly resolved by the model topography. An annual mean of the residual of the tide current will leave a mesoscale phenomenon that is similar to the non-tidal condition, which is largely contaminated by atmospheric and boundary forcings. An exact evaluation of the tidal residual current requires more specific tidal simulation with horizontally uniform density structures that are beyond the scope of this paper.

Figure 8.

Tidal ellipse at different depths (50 m, 150 m, 350 m and 750 m) from four main tidal harmonic components (a) M2, (b) S2, (c) O1, and (d) K1 between INSTANT data and the IP3-tide model.

4.3. Temperature and Salinity Profile Between Tide and Non-tide Models

[43] In 2004, the BoM of Australia had more than 170 XBT stations along the PX-2 and IX-22/PX-11 tracks, which encompass the water column in the IP3-tide domain and result in more than 45000 individual data points that are able to be compared to IP3-tide model results. The fitting of temperature from scattered depths and coordinates is shown in Figure 9. The linear fitting resulted in a correlation of 97.5% between the IP3-tide model and the BoM data. However, linear fitting showed that the IP3-tide model predicted temperatures that were almost one degree lower. Linear fitting also showed that IP3-tide temperature in the lower columns tended to be lower, while in the upper thermocline and intermediate layers, the predicted temperature was relatively higher. Strong seasonal variability in the thermocline results in a lower correlation as compared to that in the upper and bottom layers.

Figure 9.

Correlation of scattered temperature data between the IP3-tide model and observational data from 2004. Different colors are given to different depths. The thick black line is the linear curve-fitting function, and the dotted black line is the ideal result.

[44] To more accurately validate the performance of the model's tracer variables, we compared the model results at the TAO-TRITON point for a one-week time integration at the end of August 2004 and in the Indian Ocean with 10 data points of XCTD data from July, 18th–19th, 2004, from the R/V Mirai. Model validation in the Pacific Ocean yielded better results than those in the Indian Ocean. Despite this result, the model incorporating tide data displayed closer agreement with the observational data than did the model without tide data (Figure 10). Aside from those two oceans, we selected other locations in the Lombok Strait, Halmahera Sea, Banda Sea, and Makassar Strait to analyze the importance of tide inside the Indonesian seas. In Lombok, we used WODC data from March 12, 2004, 12 pm at 9.155°S; 116.022°E, while for Halmahera (2°S–1°N; 128–131°E), Banda Sea (4–7°S; 123–131°E) and Makassar Strait (1–5.5°S; 118–120°E) we used WOA01 annual data (see red boxes in Figure 1). The red boxes in the Banda, Makassar and Halmahera seas are similar to the results of Koch-Larrouy et al. [2007]. Interestingly, for all comparison areas, the IP3-tide model was able to reproduce salinity values closer to those observed.

Figure 10.

(a–f) TS profiles between observational data (purple), the IP-tide model (green) and the IP3-non tide model (red) in selected locations, especially in the Pacific Ocean, the Indian Ocean and the Indonesian seas.

[45] Koch-Larrouy et al. [2007] explained that the water mass transformation in the tide and non-tide model is different, despite the fact that the volume transport is nearly similar. They showed that due to a mixing process, the salinity maximum in the eastern route (Halmahera, Seram and Maluku Seas) and in the western route (Makassar Strait and Flores Sea) is strongly eroded. Their results showed that fresher water from the surface and bottom of the sea may reduce the salinity at the thermocline depth and thus explain why the tidal model closer approximates climatological data. Similar conditions were observed in our tidal model. A good comparison between the IP3-tide model and the observational data is shown in Figure 10. At the entrance routes of the Indonesian seas, the salinity maximum is eroded due to incoming fresher water from the surface and bottom of the water body.

[46] In addition, Figure 11 explains that the salinity maximum in the eastern route is transported from the South Pacific Ocean through the support of strong horizontal mixing around the thermocline. Due to strong tidal horizontal mixing, higher salinity and temperature maximum values inside the Indonesia Seas (Banda and Flores Sea) also diffuse widely. In the Makassar Strait at the lower thermocline level, salinity is higher in the IP3-tide model, but this is not due to an inaccurately high salinity in the Indian Ocean. The TS profile of the southern part of the Lombok strait (Figure 10d) provides evidence that a higher salinity in the Indian Sea is not responsible for the increased salinity in the Makassar Strait. Rather, the high salinity in the Makassar Strait may come from neighboring seas such as the Flores Sea.

Figure 11.

Annual mean of salinity distribution at 150 m, 250 m and 350 m (left) without tide and (right) with tide. Salinity units are psu.

[47] Figure 12 shows that horizontal mixing due to tide can be a cause of high diffusivity of salinity in the thermocline layer. High horizontal mixing is one of the consequences of tide. The apparent difference in horizontal mixing between the tide and non-tide models is one of the factors that explains the high salinity in the IP3-tide model. Horizontal mixing conditions influence segments of water columns, such as the surface, the upper thermocline, and the lower thermocline. Strong daily variability helps to dilute high salinity concentrations via advection. Here, horizontal mixing is defined as a deviation of horizontal velocity from the annual mean of horizontal velocity. The annual mean of horizontal mixing in straits and basins was increased in the IP3-tide model, suggesting the importance of tides in the straits. Strong mixing is also found in the Celebes, Halmahera, Banda and Flores Seas. This horizontal mixing improves the distribution of the salinity maximum inside the Indonesian seas. More active horizontal mixing at the thermocline depth implies active vertical mixing as well. The mixing rate given by the IP3-tide model for some straits was twice that given by the non-tide model. Mixing rate has been identified as one of the most important factors for local dynamics, besides geostrophic current [Susanto et al., 2007].

Figure 12.

Annual mean of horizontal mixing distribution at the surface (0 m), the upper thermocline (150 m), and the lower thermocline (350 m) (left) without tide and (right) with tide. Mixing is calculated from the annual RMS of the horizontal velocity deviation. Hourly (daily) data are used for the tide (non-tide) model. Velocity units m/s are not shown for the surface (below surface) condition. No data exist for below the depth of 150 m (350 m) for the upper (lower) thermocline.

[48] Furthermore, vertical diffusivity coefficients calculated using Mellor and Yamada's [1982] level 2.5 model increased in the IP3-tide model in the case of the Indonesian seas. Yet, at the western entrance route of ITF, the IP3-non-tide model depicted larger vertical diffusivity in the Indonesian seas (not shown). Even though mostly vertical diffusivity is quite small, around 1 cm2/s, in our model, but at some levels value around 0.05 m2/s was found and this value is considerably larger than that previously reported by Ffield and Gordon [1992] and Van Aken et al. [1988]. To properly evaluate the vertical mixing process in IP3-tide model, it requires more careful evaluation of the diffusivity due to the nonlinear tide induced perturbations [Hatayama, 2004]. We believed the evaluation to this issue could potentially estimate effective mixing coefficients needed by non-tidal model for correct parameterization of tidal impact on mean circulation based on tide-resolving model simulation results, but this work is beyond of the scope of the present study.

5. Internal Tide in Celebes and Sulu Seas

[49] There is a number of observed internal wave in the Indonesian seas [Jackson, 2007]. More than 400 internal wave occurrences are detected by MODIS in his study. Particularly with respect to the Celebes and Sulu Seas, a number of images firmly prove the existence of more than one type of internal wave. Beyond satellite images, modeling studies of internal waves in these areas are rare. Inclusion of tidal forcing in the IP3 model allowed us to observe the internal tide in these areas. An indication of internal tide can be detected from temperature or from the vertical velocity profile. Figure 13 depicts the meridional transect line at the 9th layer (∼80 m depth) in 121°E in 2004. One internal tide in the Celebes Sea is generated at 5°N and propagates southeastward diurnally at a speed of approximately ∼2 m/s (Figure 13b). The internal tide creates warmer water by pulling surface heat down to the 9th layer that is dominated by a temperature lower than 24°C. This effect can occur because the vertical velocity in the Celebes Sea signals strong downwelling that brings warmer temperatures to the 9th layer (not shown). Yet, in a different season, such as February, the north-south internal tides are not seen. In February, the surface current is strong, and it increases the mixing process, creating weak density structures. The reverse condition is found in August, when the surface temperature is 1°C higher and the salinity 1–1.5 psu higher than in February. Although these differences are small, individual temperature and salinity profiles depict a stronger density structure in August that is more favorable for the generation of internal tides. In the IP3-non-tide model, the stratification in February and August existed, but not as strong as stratification in the IP3-tide model (not shown). From these results, we can say that tidal forcing is important and cannot be neglected if we want to understand the mixing process in the Indonesian seas.

Figure 13.

Evidence of internal waves in the Celebes Sea. (a) A Hovmöller diagram of temperature along 1.5°–5.5°N at 121°E in 2004. (b) Similar to Figure 13a, but depicts early August 2004, at the time of the spring tide. The dotted lines are the propagation of the north-south internal tide. (c) A TS diagram of the IP3-tide model in February (black dots) and August (gray dots). The black (gray) triangles are WOA01 data from February (August). The units for temperature and salinity are degrees Celsius and psu, respectively.

[50] This high internal tide frequency is a result of tidal dynamic processes in this region and is affected by the density structure in the area. During the monsoon season, strong surface currents that develop in the Sulu and Celebes Seas increase the mixing process. We have seen that internal tide occur alongside the mesoscale eddies in this region.

[51] From these modeling results, we hypothesize that the generation of internal waves in the Celebes Seas is due to a different type of tidal propagation and amplitude between the Sulu and Celebes Seas. The Ridge separating the Sulu and Celebes Seas is the facilitator of internal tide generation on both sides. The M2 tide component in the Sulu Sea is small in amplitude, but it propagates faster than the high amplitude internal tides in the Celebes Sea (Figure 4). As such, the Sulu Sea uses its kinetic energy to force the internal tide to propagate to another sea, whereas the Celebes Sea utilizes its potentiality to trigger the internal tide in a neighboring sea. Therefore, we can see that the internal tide in the Celebes Sea is identical to the physical properties of the Sulu Sea, whereas the internal tide in the Sulu Sea also has similar physical properties to that in the Celebes Sea.

[52] In the Celebes Sea, the internal tide travels faster than its ambient M2 type component. Propagating from a low to high M2 amplitude tide (because of the characteristic M2 propagation) does not prevent the internal tide from disturbing sea water elevation at the surface (see Animation S1). A picture of the internal tides in the Celebes Sea from the surface is shown in Figure 14, which was taken on August 19th, 2004, during neap tide. This internal tide is stable; it can even enter the Makassar Strait and decay when approaching a coastal boundary. The amplitude from the baroclinic component of the T/P altimetry data-depicted disturbances in this area are also seen in Figure 4, with a 1.5°–2.0° wave length and a 10 cm amplitude.

Figure 14.

(a) Snapshot of the normalized surface elevation (m) for August 19th, 2004, 01.00 am GMT in the Celebes and Sulu Seas. At this time, the tide condition was low neap tide. Two arrow lines (ascending and descending) represent the Topex Poseidon track line for subfigure in Figure 14b . (b) Comparison of the amplitude of the baroclinic components between the IP3-tide model and Topex Poseidon altimetric data. Topex Poseidon altimetric data were extracted by Robin Robertson and have been 250 km spatially filtered. The 89 and 91 data refer to the ascending and descending tracks, respectively.

[53] We cannot clearly observe the characteristics of the north-south internal tide in the Sulu Sea because it is located at the northern boundary of our model domain and limited only to the southern part of Sulu Sea. Nonetheless, the decaying mechanism of the internal tide in the Sulu Sea is interesting and is similar to the propagation mechanism of the internal tide in the Celebes Sea. In the Sulu Sea, this internal tide may decay before reaching the coastal boundary. The faster M2 component from the South China Sea prevents a relatively slow and low-amplitude internal tide in the Sulu Sea that moves in the opposite direction. There is no superposition or modulation of frequencies because both Seas are dominated by the M2 type. We observe that the northward propagation of the internal tide in the Sulu Sea is not as frequent as the southward internal tide in the Celebes Sea. During low tide, an internal tide in the Sulu Sea can be seen.

[54] Aside from the north-south internal tide, in Figure 14, we observe at least two additional types of internal tides in the Sulu and Celebes Seas. As mentioned earlier, the internal tide is generated as a result of a big difference in the mode of the M2 phase and amplitude around the island chain separating the Sulu and Celebes Seas that propagates in a north-south direction. Another type of internal tide that is generated by topographic disturbances is found at the western boundary of the Sulu Sea. A similar process happens in the Celebes Sea. Another group of internal tides is observed in Figure 14, moving from east to west. Aside from the internal tide in the Celebes and Sulu Seas, internal tides in other location are seen but are not reported on here. Typically, their generation is due to topography disturbances as also reported by Lek [1938] mentioned in the work by Munk [1997].

6. Summary

[55] We have presented a 1/36° regional high-resolution ocean circulation model that is the first to combine tide and atmospheric forcing on the regional scale for the Indonesian seas. The four main tidal constituents (M2, S2, O1, and K1) and six-hourly NCEP/NCAR real-time reanalysis atmospheric forcing generate complex barotropic and baroclinic fields that do not fully replicate the conditions in the Pacific and Indian Oceans. A simulation was conducted during normal sea conditions in 2003–2004, and the analysis focused only on the year 2004. We checked the model performance with various data sets, including observational data (tide gauges, PX-2 and IX-22/PX-11 XBT lines, INSTANT mooring stations and R/V Mirai cruise data), derived data (tidal chart and T/P altimetry data) and global atlas models (FES2004 and OTIS model) for 2004. The modeling results showed acceptably good agreement for all prognostic variables with various observed data at widespread locations in the entire basin. For instance, elevation reaches an RMS difference of 4.7 cm out of a 51 cm mean total variability for amplitude and 18.7° for phase. Temperature deviates by 0.899°C out of a 4°C total annual variability around the thermocline region. Horizontal velocity has good agreement, especially for tide and intra-seasonal frequencies.

[56] The interaction between low and high frequencies has been discussed. By explicitly inserting barotropic tide into an ocean general circulation model, higher frequencies developed, and baroclinicity formed three dimensionally, causing the generation of internal tides that may disturb the tidal surface elevation. The high frequencies enhance the mixing process of mesoscale phenomena, especially in straits and basins. This process can also enhance the background model variability closer to the observed values. Vertical and horizontal mixing associated with tides is considerably improved and naturally generated without any cosmetic tuning of the model's diffusivity term. They are also responsible for eroding salinity maximum of North Pacific Subtropical Water (NPSW) and South Pacific Subtropical Water (SPSW) advected from the Pacific Ocean [Ilahude and Gordon, 1996; Koch-Larrouy et al., 2007]. Whereas low frequencies influence seasonal density structure, the more stratified density structure is more favorable for the generation of internal tides inside the Indonesian seas. This model successfully incorporated a combination of two forcings and initially performed well in the high-resolution mode of the ocean model for complex arrays of archipelagos in the center of the Indonesian seas. From these results, we can conclude that tidal forcing is important and cannot be neglected if we want to understand the mixing process in the Indonesian seas.

[57] However, this conclusion does not mean that this model is free from bias, and the detailed, complex, high-frequency features for this area needs to be improved. Although in the IP3-tide model the water mass property is improved as compared to that in the non-tide model, we need to improve the large discrepancy in the salinity of the Indian Ocean. Many methods can be used to improve the model quality, such as using higher resolution atmospheric and tidal forcing, but the remaining shear stress problem at the bottom of the water column needs to be resolved. Arbic et al. [2010] stressed the importance of their results from an atmospheric-tidal forcing model regarding this issue, and we found a similar problem. Besides, an annual volume transport at 8.2°S; 114–126°E in the IP3-tide model is −9.7 Sv, which is considerably smaller compared to Sprintall et al. [2009]. Therefore, we need a more descriptive bottom stress term that represents more dissipative drag stress at the deeper column. Another remaining issue is the possibility of a more effective simulation by introducing parallel computation. Nonetheless, the development of a tidal model embedded to a regional model has shown more realistic features inside the Indonesian seas, such as the appearance of internal tides, baroclinic tides and tidal mixing in the water column, that are closer to observational data.


[58] This work was funded by the Global Environment Research Fund (F-082) of the Ministry of the Environment, Japan, and a Grant-in-Aid for Scientific Research (A) (18254003 and 21254002) from JSPS (Japan Society for the Promotion of Science). We benefited considerably from the facilities offered by JAMSTEC's Yokohama office, which supported running all of the processes on their 2 supercomputer machines (SX-8R and SGI Altix4700). We thank the institutions mentioned earlier for supporting and providing data validation for this work. The Indo-Pacific code uses the JCOPE2 code developed by the JCOPE group. The first author was supported by a scholarship for foreign students offered by the Ministry of Education, Culture, Sports, Science and Technology (MEXT), Japan, and the JAMSTEC-University cooperation program. The authors gratefully acknowledge the useful comments, support and suggestions from the weekly JCOPE group meetings, Robin Robertson, Xinyu Guo and the anonymous reviewers. The T/P data were received from Robin Robertson through personal communication.