An Evaluation of a 1/18 o Resolution Regional Ocean Circulation Model of CROCO in the Southern Sunda Shelf

The Coastal and Regional Ocean Community Model (CROCO) was applied to simulate the three dimensional ocean circulation on the southern Sunda Shelf with a horizontal resolution of 1/18 o and 40 vertical layers. This configuration is called as CROCO18-SSS. The initial and open boundary conditions were derived from a global reanalysis product at a resolution of 1/12 o from the Copernicus Marine Environment Monitoring Service (CMEMS). The regional model was driven by atmospheric, tides, and river discharges data from a wide variety of datasets, each with a different resolution. To test the model performance and capabilities of the CROCO18-SSS model outputs including sea level anomaly, current velocity, temperature, salinity, and volume transport, we have compared in the simulation results with other reanalysis product and satellite-derived datasets. The CROCO18-SSS model outputs reproduced the regional ocean circulation well from observations and reanalysis products. For example, the reversal of the monsoonal current in the southern sunda shelf, as well as strong persistent southward Makassar throughflow and their interaction in the southern Makassar Strait, can be seen clearly. The vertical structure of the zonal current component was comparable to that of the reanalysis product reference. The seasonal shift in temperature and salinity, with the water becoming colder and saltier during the northwest monsoon and warmer and fresher during the southeast monsoon, was also accurately reproduced by the CROCO18-SSS model. Additionally, the mean volume transport calculated by the CROCO18-SSS model for the Makassar and Karimata Straits was comparable to that reported in previous studies.


Introduction
Ocean modelling has long been used to study ocean dynamics and water mass stratification (Tranchant et al., 2016); and to analyze oceanographic phenomena, such as fronts (Ismail and Ribbe, 2019), upwelling (Kok et al., 2017), eddies (Horhoruw et al., 2020), barotropic and baroclinic tides (Koropitan and Ikeda, 2008;Nugroho et al., 2018;Katavouta et al., 2022), and ocean current systems (Tozuka et al., 2009;Susanto et al., 2012;Wei et al., 2016). Some benefits of using ocean modelling are as tools to better understand and to evaluate the physical processes and dynamics of ocean phenomena, which could contribute to the enhancement of our knowledge in the field of marine science and technology.
The Coastal and Regional Ocean Community model (CROCO) has been developed to investigate the ocean circulation dynamics. It is a recent hydrodynamic modelling system based on ROMS-AGRIF (Shchepetkin and McWilliams, 2005;Debreu et al., 2012). It adopts surface-and terrain-following curvilinear coordinates (Shchepetkin and McWilliams, 2005). The model is capable to address coastal phenomena and their interactions with the open sea. Previous studies have applied CROCO modelling to study atmospheric wave-ocean coupling (Uchiyama et al., 2010), internal waves, sediments (Blaas et al., 2007), and biogeochemistry (Gutknecht et al., 2013). It has also been applied to study the coastal front (Ismail and Ribbe, 2019), impact of rivers on ocean circulation (Ramadhan et al., 2021), oil spill prediction , and structure of coastal upwelling dynamics (de Mello et al., 2022).
The bathymetry in the southern Sunda shelf is less than 70 m-depth which is dominant in the model domain, including the Karimata Strait, Java Sea, the western Flores Sea, and the southern part of Makassar Strait (Sarr et al. 2019). The seasonal reversal of the Asian-Australian monsoon system (NWM -northwest monsoon and SEM -southeast monsoon) strongly influences the SSS ocean dynamics (Wyrtki, 1961;Hamzah et al., 2020). During the NWM period (December to February), northwesterly winds are fully developed, which is associated with the rainy season. In contrast, during the SEM period (June to August), southeasterly monsoon winds blow with dry air masses from the Australian Continent. Consequently, surface current distribution and oceanographic parameters are strongly controlled by the monsoon system (Apriansyah et al., 2022). The ocean dynamics of SSS waters are also influenced by the Sunda Shelf Throughflow (Apriansyah and Atmadipoera, 2020), Indonesian Throughflow (ITF) from the Makassar Strait (Gordon, 2005;Du and Qu, 2010), complex bathymetry, unique atmospheric characteristics (Yulihastin et al., 2021), tides (Nugroho et al., 2018;Katavouta et al., 2022), and rivers (Ramadhan et al., 2021). The complex bathymetry, atmosphere, ocean dynamics, tides, and river runoff surrounding SSS must be considered as model inputs.
In this study, we have performed a higher resolution CROCO modelling that was applied to simulate ocean circulation in the southern Sunda Shelf (SSS). This study focused on evaluating the model capability and confidence for some model output variables, such as current velocity, sea level anomaly, temperature, salinity, and volume transport, by comparing the model output dataset with the reanalysis products and satellite-derived datasets. Other related papers from this simulation may be discussed elsewhere. Therefore, this study evaluates the performance of the model output from a realistic 3-dimension hydrodynamic model of CROCO and to investigate seasonal variations in ocean circulation and its variability in the Sunda Shelf waters.

Materials and Methods
The regional model configuration of the southern Sunda Shelf water was determined using CROCO version 1.2. The model domain was set up with a spatial resolution of 1/18 o (approximately 5 km) and 40 vertical layers, with much denser vertical layers near the surface. This configuration is referred to as CROCO18-SSS.
This study used vertical S-coordinate stretching parameters of 7 s  = , 1 b  = , and 2 c hm = , which increased vertical resolution at the surface. A minimum depth of 5 m was selected for the study. The specified value cannot be zero, and must be set based on the horizontal resolution of the model grid. The bathymetric model was derived using the topography of the General Bathymetric Chart of the Oceans (GEBCO) gridded dataset for 15 s (Weatherall et al., 2015). The slope parameter was 0.25, which prevented horizontal pressure gradient errors (Ismail and Ribbe, 2019). The bottom friction coefficient was calculated using logarithmic law (bottom roughness = 0.01) (Malauene et al., 2018). Ganju and Sherwood, (2010) provided further details regarding the algorithm that parameterizes bottom roughness in an ocean model.
The momentum term uses 3rd-order upstream biased advection and 4th-order centered advection with harmonic averaging for vertical advection. For vertical mixing, K-profile (KPP) parameterization was used, and for lateral tracer advection, a split and rotated 3rd-order upstream biased lateral tracer scheme was used. Flather open boundary conditions for barotropic velocities and radiative open boundary conditions ).
The hydrodynamic model configuration was applied using the ROMS_TOOLS toolbox (Penven et al., 2008). The model area stretches from 104 °E to 120 °E and 9.5 °S to 0.5 °N (Figure 1.). The open boundaries were set up on the northern, eastern, southern, and western sides of the model domain. The lateral boundary conditions were obtained from the GLORYS12V1 reanalysis product (Lellouche et al., 2018), with a horizontal resolution of 1/12 o and a time span from 2010 to 2020. Table 1 presents the setup of the CROCO18-SSS configuration.
The model is forced by atmospheric parameters and river discharge along its open lateral boundaries by tides and oceanic conditions. Atmospheric forcings were obtained from ERA5 data (Hersbach et al., 2020). ERA5 is the fifth generation of the ECMWF's atmospheric reanalysis of global climate (European Center for Medium-Range Weather Forecast, ECMWF). Surface forcing encompasses precipitation, air pressure, relative humidity, air temperature, shortwave and longwave radiation, and wind stress imposed from the ERA5 reanalysis at a 1/4 o horizontal resolution. Explicit tidal forcing was obtained from the TPXO.9v4 dataset at a resolution of 1/6 o horizontally (Egbert and Erofeeva, 2002). The 10 tidal constituents (K1, O1, Q1, P1, M2, N2, S2, K2, Mm, and Mf) were used as the open boundaries. The daily river discharge from the Global Flood Awareness System (GloFas) at a resolution of 1/10 o (Harrigan et al., 2020) was applied at the river mouths along the coastlines of the study area. River discharge-point sources were mostly available along the coastlines of the model domain. There were 117 river sources, with an average discharge of above 50 m 3 .s -1 . GloFas was chosen because it has been validated by field measurements in a study by Ramadhan et al. (2021). The CROCO18-SSS simulations ran from January 1, 2009, to December 31, 2020, with a 1-year spin-up (2009).

Data comparison
The model outputs of CROCO18-SSS were compared with the observational satellite-derived dataset and reanalysis product dataset to test the model performance and confidence. The observational satellite-derived dataset consisted of sea level anomalies (SLA), temperature, and salinity. Daily SLA data were obtained from the DUACS satellite, with a horizontal resolution of 1/4 o (Taburet et al., 2019). Sea surface temperature data were obtained from the OSTIA satellite system, with a horizontal resolution of 1/20 o (Good et al., 2020). Salinity data were obtained using a multivariate optimal interpolation algorithm that combined SMOS satellite images, in situ salinity measurements, and satellite SST information. The products were developed by COnsiglio Nazionale delle Ricerche (CNR) (Droghei et al., 2016). The data were collected weekly at a horizontal resolution of 1/4 o on January 1, 1993. Multisensory satellite observational datasets were obtained from http://marine.copernicus.eu. Additional datasets from the GLORIS12V1 reanalysis products (SLA, current, salinity, and temperature) were also used for comparison, obtained from https://marine.copernicus.eu, based on the reanalysis products of the model and assimilation of observations with a resolution of 1/12 o . The GLORYS12V1 reanalysis product dataset was validated using an observational dataset (Drévillon et al., 2021).

Data analysis
To evaluate the CROCO18-SSS simulation, several model output variables (temperature, salinity, SLA, currents, and volume transport) were analyzed. The CROCO18-SSS output parameters were compared with GLORYS12V1 reanalysis and satellitederived data. Seasonal and spatial variation of dataset (zonal and meridional currents, temperature, and salinity) were analyzed and contrasted between different monsoon periods (the NWM and SEM).
Climatology analysis or the annual cycle of parameters to investigate the seasonal cycles of the parameters (temperature and salinity) is calculated using Eq.1 below. The climatology method follows Thomson and Emery (2014): where: v denotes meridional velocity (m/s), dA shows the area elements of the PP' and QQ' transects. The VT value was expressed in units of Sv (1 Sv = 10 6 m 3 .s -1 ).
Taylor diagram analysis (Taylor, 2001) was used to evaluate the correlation coefficient, standard deviation, and root mean squared deviation between the CROCO18-SSS dataset, GLORYS12V1 reanalysis dataset, and satellite data such as sea level anomalies, temperature, salinity, and volume transport. The scale value of Taylor diagram ranges between 0 and 1. Model performance improves if the correlation coefficient is close to 1. The RMSD and standard deviation near 0 indicated high accuracy.

Surface current
The CROCO18-SSS surface current simulation matched the GROLYS12V1 reanalysis product ( Figure  2). Surface currents in the SSS are caused by wind stress, which varies seasonally (Kok et al., 2021).
During the NWM (December-February), northwest monsoon winds pushed the Karimata Strait surface currents to the south. This current then turned eastward and was joined by a monsoongenerated surface current in the Java Sea. The eastward surface current splits in the eastern Java Sea, whereas one part continues to the Flores Sea and the remaining current turns northward and joins the southward Makassar throughflow current ( Figure  2a, b.). During the SEM (June-August), the southeast monsoon winds drove the surface currents westward (Figure. 2b,d.). In the southern Makassar Strait, the southward-flowing current divides; one part exits directly into the Indian Ocean through the Lombok Strait, and the rest turns westward and joins the Java Sea surface current. Lombok is one of the three main exit passages of the ITF, besides the Ombai Strait and Timor passages (Atmadipoera et al., 2009). The westward surface current from the Flores Sea extends towards the Java Sea during this period. Owing to the southeast monsoon, this current turns north in the Karimata Strait and joins farther the western boundary current along the Malaysia coast. Monsoon winds generated most of the water entering the Java Sea during the NWM and SEM periods (Fang et al., 2010). In the NWM season, there is a north-south sea level slope between the Natuna Sea and the Java Sea (Fang et al., 2010). This pressure gradient created a horizontal pressure gradient that drove the Sunda Shelf Throughflow (SSTF). The SSTF carries water masses from the South China Sea to Java Sea. During the SEM season, the mass of water in the Java Sea area also comes from ITF-Makassar.
There was a reasonable difference in the current patterns between the CROCO18-SSS and GLORYS12V1 reanalysis models during NWM and SEM. As shown in CROCO18-SSS, there was an eastward flow at 7 o S during SEM. The eastward flow was observed between 4°S and 6 o S in the GLORYS12V1 results. This difference may be related to the differences in the configuration of the model data, bathymetry, model parameter settings, and forcing data. The current pattern across the Java Sea was consistent between the CROCO18-SSS and GLORYS12V1 results during both NWM and SEM periods. Figure 4 compares between the daily sea level anomalies (SLA) of the CROCO18-SSS, GLORYS12V1 reanalysis, and satellite data. It was revealed that the fluctuation of the daily SLA model was mostly similar to the observed data, showing a similar trend and amplitude at all three stations (Figure 4a, c, e.). The CROCO18-SSS -GLORYS12V1 and CROCO18-SSS - In addition, the daily SLA for all stations exhibited large seasonal fluctuations in SLA, associated with the seasonal reversal of monsoon winds over the region. This is in accordance with previous studies (Kok et al., 2021).

Sea surface temperature and salinity
Monsoonal winds are the main forcing that controls physical processes and dynamics over the Sunda Shelf region, causing large seasonal changes in sea surface temperature and salinity as well as seasonal reversal currents in this region. Figure 5 shows the validation results among the CROCO18-SSS data, GLORYS12V1 reanalysis, and the satellitederived dataset of sea surface temperature and salinity at the A2 sampling location (referred to in Figure 1.). The correlation coefficient for temperature was above 0.90. The RMSD estimates are below 0.38 °C. The correlation coefficient for salinity was 0.95, and the comparison between the model salinity CROCO18-SSS and satellite-derived data was 0.76. The RMSD between model-reanalysis dataset and model-satellite dataset are 0.28 psu and 0.50 psu, respectively. The Taylor diagram for sea surface temperature and salinity ( Figure 5.) revealed that the CROCO18-SSS model reproduced the available reanalysis and satellite-derived datasets well.
During the NWM, colder (Figure 6a-c.) and saltier (Figure 7a-c.) water masses flowed southward and joined warmer, fresher water masses in the Java Sea. High precipitation rates during this rainy season significantly contribute to large seasonal salinity variations in the Java Sea (Nuryanto et al., 2016;Yulihastin et al., 2021). Many big rivers discharged into the Java Sea also reduce the salinity (Ramadhan et al., 2021). ). In the Java Sea, eastward monsoonal currents bring cooler and fresher water masses into the southern Makassar Strait. According to Tozuka et al. (2009), the arrival of this fresher water mass creates a north-south horizontal pressure gradient associated with northward surface flow in the upper 50 m and reducing Makassar throughflow in the thermocline layer. This pressure gradient creates a salinity gradient in the southern Makassar Strait.
During the SEM period, colder (Figure 6d-f.) and saltier (Figure 7d-f.) water masses were dominant in the eastern Java Sea and south of Makassar Strait. Water masses from Makassar throughflow as well as from upwelling water in the southern Makassar Strait contribute to variations in temperature and salinity in the region (Atmadipoera and Widyastuti, 2015;Utama et al., 2017). Figures 2,  6, and 7 show that westward surface currents also bring colder and saltier water masses from the Flores Sea to Java Sea. In the Java Sea, less salty water mass derived from large river discharges is still present and mixed with saltier water mass from the Flores Sea, causing much lower saline water mass.
Westward monsoonal current transports colder and fresher Java Sea water as far as crossing equator line near the Karimata Strait. Figure 8 shows the vertical distribution of water temperature in the Java Sea during the NWM (a, b) and SEM (c, d) periods. During the NWM period, there was no stratification of the temperature distribution from the surface to the bottom. Stratification occurred at the bottom of the water during the SEM season. During the NWM period, the water mass of the SCS entered the Karimata Strait, influencing the temperature of the Java Sea. Despite more frequent precipitation during the NWM period, the water temperature was relatively lower than that during the SEM period. This was caused by colder water masses from the Makassar Strait and Flores Sea entering Java Sea during the SEM period. This cold temperature is primarily caused by upwelling in the southern Makassar Strait (Atmadipoera and Widyastuti, 2015;Utama et al., 2017;Purba and Khan, 2019).     Figure 9 depicts the vertical salinity distribution in Java Sea waters along the 114 o S transect during the NWM (a, b) and SEM (c, d) periods. During the NWM period, there was no stratification of salinity. During the SEM season, the vertical salinity distribution near the bottom of the water demonstrated stratification. During the NWM season, the salinity of the water entering the Java Sea from the South China Sea via the Karimata Strait was lower than that during the SEM season. This high salinity (>32 psu) results from the upwelling in the southern Makassar Strait Purba and Khan, 2019;Utama et al., 2017). In both seasons, mixing freshwater masses from rivers in southern Kalimantan (3 o S) lowered the salinity (~ 30.5 psu) of the northern seas. However, this impact was greater during the NWM season. The NWM period was characterized by abundant precipitation. Although the SEM period is distinguished by low precipitation (Yulihastin et al., 2021), the vertical distribution of salinity is more homogeneous.
The CROCO18-SSS models were colder and saltier than the GLORYS12V1 reanalysis and satellite data in both seasons both spatially and vertically. These differences were caused by external forcing (bathymetry, atmosphere, tides, river discharge, and oceanic boundary conditions) and the configuration of each model. Thus, each grid model interacts differently. The horizontal and vertical resolutions of the two datasets also caused model grid interactions that increased the difference between their parameters.
The annual cycles of sea surface temperature (Figure 10a.) and salinity (Figure 10b.) derived from the CROCO18-SSS model were relatively consistent with those of the GLORYS12V1 reanalysis and satellite data, indicating that the CROCO18-SSS model can capture seasonal fluctuations in the Karimata Strait (A1), Java Sea (A2), and Makassar Strait (A3). Figure 10a shows that the sea surface temperatures in the Karimata, Java, and Makassar Straits had two peaks in April and September. Seasonal variations in the sea surface salinity (Figure  10b.) in the Java Sea and Makassar Strait peaked in September. This differs from the Karimata Strait, which peaks during April and October. The seasonal cycles of temperature and salinity in the Karimata, Java, and Makassar Straits are related to seasonal water mass movement from adjacent seas (Kok et al., 2021), such as the South China Sea and Flores Sea.
During the NWM and SEM, cold water masses from the southern South China Sea (SSCS) and Flores Sea influenced the seasonal temperature cycle in the Karimata Strait and Java Sea. The SSCS began advecting cold water in December. Colder water was returned to the Java Sea. This cold-water mass meets the warmer Makassar strait water. This movement ended in March, causing the Karimata Strait and Java Sea temperatures to drop, while the Makassar Strait temperatures rose. In addition to the SSCS, the Flores Sea and Makassar Strait currents cool the Karimata Strait and Java Sea. Currents from the Flores Sea enter the Java Sea in May, bringing cold water masses that remain until October. The direction of the Flores Sea current changed in November, increasing the temperature. During SEM, saltwater masses from the Flores Sea and Makassar Strait influenced the seasonal salinity cycle in the Karimata and Java Seas. This movement made June to October salty. River outflow leads to low salinity along the coast, especially in the vicinity of the river mouths. Throughout the season, the salinity was low. This seasonal cycle of sea surface temperature and salinity describes the southern Sunda Shelf water mass.

Volume Transport
This section evaluates water transport through the KST and MST. Water masses follow both paths from the Pacific to the Indian Ocean (Gordon, 2005;Du and Qu, 2010;Gordon et al., 2012;He et al., 2015;Apriansyah and Atmadipoera, 2020). Figure  11 compares the CROCO18-SSS and GLORYS12V1 reanalysis data for the volume transport of the Karimata (KST) and Makassar straits (MKT). The correlation between KST and MKT volume transport was above 0.9. The RMSD values of KST and MKT were 0.1982 and 0.923, respectively. Evaluation of volume transport in the two transects showed that the CROCO18-SSS model output was significant for the GLORYS12V1 reanalysis product.   The calculation of the transport of the two straits reveals the water mass exchange in Karimata (Figure 12.) and Makassar Strait (Figure 13.). Figure  12 shows a comparison of the volume transport in the Karimata Strait. CROCO18-SSS and GLORYS12V1 have almost identical phases and magnitudes. Transport to the south was the highest during the NWM season (December-February). Minimum northward transport occurred during the SEM season (June-August). This is consistent with a previous study (Xu et al., 2021), which stated that water mass movement through the Karimata Strait strengthens southerly during the NWM season and reverses direction towards the north during the SEM period, with annual mean 0.78±0.12 Sv (Figure 12c.). Figure 13 compares the volume of transport in the Makassar Strait. The CROCO18-SSS and GLORYS12V1 datasets have almost the same phases and magnitudes. Transport to the south was highest during the SEM season (June-September). The minimum southward transport occurred during the NWM season (December-February). According to Gordon et al. (2019), as shown in Figure 13c, the southerly transport achieves its maximum condition (0-760 m) during the SEM season, with an average transport of 12.5 Sv. Table 2 shows the annual volume transport through the Karimata (KST) and Makassar (MKT) straits based on CROCO18-SSS and previous studies. On an annual scale, estimates of transport through the Karimata and Makassar Straits are -1.51±1.75 Sv and -9.81±2.76 Sv. A comparison of the study's annual transport with that of previous studies indicated a great variation. The estimated annual volume transports from all previous studies in Table  1 are 0.08 -3.2 Sv and 7.39 -12.21 Sv for KST and MKT. The current study and GLORYS12V1 reanalysis provided a higher estimate for KST and a lower estimate for MKT. This could be due to the differences in external forcings and configurations from the other data models.

Conclusions
The CROCO model successfully simulated the 3-dimesional ocean circulation model on the southern Sunda Shelf, covering the Java Sea, southern Natuna Sea, Makassar Strait, and Flores Sea, with a horizontal resolution of 1/18 o and 40 vertical layers. The CROCO18-SSS results correctly reproduced regional ocean circulation in this region. A comparison of the reanalysis product and satellitederived (SLA, temperature, salinity, and volume transport datasets) showed a good agreement. The simulated sea level anomaly was similar to those of the satellite and reanalysis datasets. The vertical structure of the zonal current component is comparable to that of other data references. In addition, the seasonal evolution of temperature and salinity with warmer and fresher seawater (colder and saltier) during the northwest monsoon (the southeast monsoon) was also reproduced well by the model. The mean volume transport through the Karimata and Makassar straits has a similar phase and magnitude, and the annual mean volume transport is estimated of about -1.51±1.75) Sv and -9.81±2.76) S and are also comparable to that reported in previous studies.