The ocean model used here is the same as that used by Chang et al. (2026a); it is based on the MOM6 (Adcroft et al., 2019) ocean model, coupled with the Sea Ice Simulator version 2 for sea ice dynamics and themodynamics. It solves the hydrostatic primitive equations using the Boussinesq approximation on an Arakawa C-grid with a horizontal resolution of 1/24°. The model domain extends from 99 to 170° E and 5 to 63° N and encompassed the Northwest Pacific Ocean. Bathymetry is based on the GEBCO 2024 dataset merged with the high-resolution Korbathy dataset (Fig. 1a) (Seo, 2008). The model depths range from a minimum of 10 m, which allows for tidal variations without requiring a wetting and drying scheme, such as the one recently implemented in MOM6 (Wang et al., 2024). Time integration is performed using a split-explicit method (Hallberg, 1997; Hallberg and Adcroft, 2009), with a baroclinic time step of 300 s. The barotropic time step is chosen as the largest globally stable integer fraction of the baroclinic time step. A 900 s step is used for thermodynamic calculations. Both vertical coordinate systems used 75 layers, with the finest resolution near the surface (2 m layers down to a depth of 14 m). This vertical resolution is consistent with recent MOM6 applications in both global (Adcroft et al., 2019) and regional configurations (Ross et al., 2023; Seijo-Ellis et al., 2024; Drenkard et al., 2025; Liao et al., 2025). In the ZSTAR configuration, the layer thickness gradually increases with depth, reaching 349.43 m near the maximum model depth of 5000 m. In contrast, the HYBRID configuration uses z* coordinates in the upper ocean and switched to isopycnal coordinates in the deeper stratified interior. The transition depth between these two systems varied with latitude (Adcroft et al., 2019). Each interface in HYBRID is assigned a target density referenced to 2000 dbar, ranging from 1010.00 to 1037.2479 kg m⁻³.

The subgrid-scale parameterizations follow those of Adcroft et al. (2019) and Ross et al. (2023). The planetary boundary layer is represented using the energetics-based scheme of Reichl and Hallberg (2018) with Langmuir turbulence updates (Reichl and Li, 2019). Submesoscale restratification is parameterized by the scheme of Fox-Kemper et al. (2011) using a 1500 m frontal length scale. Horizontal viscosity is calculated using a biharmonic formulation (Griffies and Hallberg, 2000), taking the maximum of Smagorinsky viscosity and a fixed form (u4Δx3), with u4 = 0.01 m s⁻¹ and a Smagorinsky coefficient of 0.015. The shear-driven mixing is parameterized according to the method of Jackson et al. (2008).

Both configurations are forced with daily temperature, salinity, SSH, and velocity data from GLORYS12v1 (Jean-Michel et al., 2021). Tidal forcing is imposed using 10 tidal constituents from TPXO9 v1 (Egbert and Erofeeva, 2002), which were applied at the boundaries and as body forces. The self-attraction and loading effect was parameterized using the scalar approximation (Accad and Pekeris, 1978) with a coefficient of 0.01. The barotropic components use radiation conditions (Flather, 1976), and the baroclinic flows apply an Orlanski scheme (Orlanski, 1976) with nudging (Marchesiello et al., 2001). A reservoir scheme with a length scale of 9 km is used to determine the inflowing temperature and salinity at the open boundaries. Surface forcing is derived from hourly ERA5 reanalysis (Hersbach et al., 2020) using the bulk formula of Large and Yeager (2004). River discharge from GloFAS v3.1 (Alfieri et al., 2020) is mapped to the model grid, following Ross et al. (2023). Discharge was applied as freshwater at the surface temperature, with turbulent kinetic energy added to mix the upper 5 m.

Both configurations are initialized using the GLORYS12v1 temperature and salinity fields on 1 January 1993. A 10-year spin-up (1993–2002) is conducted using time-varying boundary and surface forcing, followed by a 10-year hindcast (2003–2012). For tidal diagnostics, analyses of barotropic and baroclinic tidal energetics are conducted using model output from the final year (2012) of the hindcast. Table 1 summarizes the model configuration and parameterizations.

To assess the performance of the simulated sea surface temperature (SST) under different vertical coordinate systems in the Yellow Sea, we utilize the Operational Sea Surface Temperature and Sea Ice Analysis (OSTIA) product distributed by the Copernicus Marine Environment Monitoring Service (Good et al., 2020; 10.48670/moi-00165). OSTIA provides daily, gap-filled foundation SST fields derived from both satellite and in situ observations, ensuring consistency and accuracy for model validation. The product has a horizontal resolution of 0.054° and has data coverage going back to January 1981, making it suitable for evaluating SST variability in shallow tidally dynamic regions, such as the Yellow Sea. Among the available SST datasets, OSTIA performs particularly well in the Yellow Sea, with Woo and Park (2020) reporting lower errors than Daily Optimum Interpolation Sea Surface Temperature when validated against in situ measurements.

We also use observational temperature data from the Korea Oceanographic Data Center (KODC), maintained by the National Institute of Fisheries Science (NIFS). Since 1961, the NIFS has conducted oceanographic surveys around the Korean Peninsula approximately six times a year (during even-numbered months). Their observational network covers major regions such as the East/Japan Sea, Yellow Sea, and Korea/Tsushima Strait, with in situ temperature profiles reported at standard depths (0, 10, 20, 30, 50, 75, 100, 125, 150, 200, 250, 300, 400, and 500 m). In this study, only observations collected in the Yellow Sea are used to validate the vertical temperature structure of the model. The validation for salinity is excluded, as a previous study by Park (2021) reported significant time-dependent bias errors in KODC salinity records, rendering them unsuitable for model evaluation.

To evaluate the ability of the model to represent both barotropic and baroclinic tidal processes, a model output with at least an hourly temporal resolution is required. Given the substantial data volume associated with high-resolution simulations, only the final year (2012) of the 10-year hindcast is archived at an hourly frequency for key variables, including temperature, salinity, SSH, and three-dimensional velocity. To extract the dominant tidal signals, harmonic analysis was performed on the hourly SSH and velocity fields using the UTide Python package (Codiga, 2011), focusing on the M2 tide, which is the dominant constituent of the Yellow Sea. Based on this output, the diagnostic metrics used in this study include the barotropic energy flux, baroclinic kinetic energy (hereafter referred to as KE), baroclinic energy flux, and barotropic-to-baroclinic conversion rate.

The barotropic tidal energy flux can be expressed by the following equation: 1Fbt=ρ0gUbtη, where ρ0 is the reference seawater density, set to 1025 kg m⁻³ in this study; g is the gravitational acceleration; Ubt is the barotropic velocity; and η is the tidal elevation for M2.

The depth-integrated baroclinic KE is calculated as: 2KEbc=∫-Hηρ0(u′2+v′2)/2dz, where u′ and v′ represent the baroclinic components of the horizontal velocity. The baroclinic velocity is derived from velocity profiles. 3u′z,t=uz,t-1H∫-H0uz,tdz Because MOM6 is fundamentally a layer model and does not explicitly provide the vertical velocity (w), the vertical component of the KE is excluded from this calculation.

As described by Kang and Fringer (2012), the baroclinic energy flux can be computed using the baroclinic velocity and the pressure perturbation, where the perturbation pressure p′ is obtained from perturbation density ρ′ relative to the time-averaged background field. The perturbation density is calculated as follows: 4ρ′=ρx,y,z,t-ρb(x,y,z), where ρx,y,z,t is the time-dependent density and ρb(xyz) is the time-independent background density. The perturbation pressure is calculated as follows: 5p′z,t=psurft+g∫zηρ′(z)dz^,6psurft=-1H+η∫-Hη∫Zηρ′(z,t^)gdz^dz. Subsequently, the depth-integrated baroclinic energy flux is calculated by the following equation. 7Fbc=∫-Hηp′(z)u′(z)dz. To further quantify the energy transfer from surface to internal tides, we compute the barotropic-to-baroclinic energy conversion rate, which represents the local generation of baroclinic (internal) tides via the interaction between the barotropic current and internal pressure gradients. This conversion rate can be expressed as: 8C=W⋅p′9W=-∇Hh+ηUbt where W denotes the barotropic vertical velocity generated as the horizontal barotropic flow Ubt is forced to flow over spatial variations in the bottom topography.

The M2 and K1 tidal amplitudes and phases used for comparison with TPXO9 were obtained from harmonic analysis of hourly sea surface height (SSH) over a one-year period to ensure reliable estimation of the principal tidal constituents. To investigate seasonal variability, February and August were selected as representative winter and summer conditions, respectively. For each month, the diagnostic metrics are calculated using hourly outputs from the first 15 d, corresponding to approximately 29 M2 tidal cycles, ensuring robust characterization of the semi-diurnal tidal response. To quantitatively evaluate the barotropic and baroclinic tidal responses influenced by vertical coordinate systems, spatial averages are computed over four energetic tidal regions in the Yellow Sea, as identified by Liu et al. (2019): the Yangtze River Estuary (121.8–124° E, 29.5–32° N), South Yellow Sea (122–125.8° E, 32–33.5° N), South Korean Coast (125–127° E, 33.6–36.6° N), and North Korean Coast (124–126° E, 37–39° N).

Before comparing the barotropic and baroclinic tide simulations, the model representation of key physical properties, including SST and vertical temperature structure, is examined against satellite and in situ observations to ensure that the background stratification and surface conditions, which are critical factors for tidal dynamics, are reasonably captured. Figure 2 presents the spatial distribution of the mean SST and its biases relative to OSTIA in the Yellow Sea. Both the HYBRID and ZSTAR configurations capture the large-scale SST structure reasonably well and exhibit a high spatial correlation (Corr) with OSTIA (0.95 for HYBRID and ZSTAR). However, HYBRID shows superior accuracy, with a lower root mean square error (RMSE) of 0.58 °C, compared with 0.75 °C for ZSTAR. While both models exhibit a warm bias (< 1.0 °C) in the western Yellow Sea and a cool bias along the southeastern margin, ZSTAR additionally displays an extensive cold bias throughout the southern Yellow Sea.

To compare the representations of the temperature field under weakly and strongly stratified conditions, the validation period is divided into two seasonal groups (Fig. 3). Because the NIFS collects in situ temperature profiles around the Korean Peninsula only during even-numbered months, the validation is conducted using data from February, April, and December (FAD) to represent weak stratification and from June, August, and October (JAO) to represent strong stratification. For each season, the depth-averaged temperature bias and RMSE are computed to evaluate the performance of the HYBRID and ZSTAR configurations. During the weakly stratified season (FAD; Fig. 3a and b), both configurations exhibit a warm bias across the domain, particularly in the southern region. However, ZSTAR displays a stronger warm bias, with temperatures exceeding those of HYBRID by approximately 0.5 °C. In contrast, during the strongly stratified season (JAO, Fig. 3c–d), ZSTAR exhibits a pronounced cold bias exceeding -3.0 °C in the central Yellow Sea, while simultaneously showing a warm bias along the Korean coastal region. HYBRID, however, displays consistently reduced biases in both the offshore and nearshore regions, suggesting a more balanced and realistic vertical temperature distribution. The RMSE distributions (Fig. 3e–h) further support these findings. ZSTAR tends to produce larger errors across the section, especially during JAO, when stratification is strong, with peak RMSEs surpassing 2.5 °C. In contrast, HYBRID maintains relatively low RMSE values across most latitudinal bands and depths. This result highlights the superior ability of HYBRID in reproducing vertical thermal structures under various stratification conditions.

The vertical temperature structure in August, when stratification in the Yellow Sea is strongest, is generally well reproduced by both configurations, although noticeable deviations from the observations remain (Fig. 4). HYBRID tends to align more closely with the observed profile than ZSTAR across much of the water column, particularly within the thermocline (approximately 10–40 m depth), where strong vertical gradients occur (Fig. 4a). Although ZSTAR also captures the overall stratification, it exhibits somewhat larger deviations in this depth range and shows a slightly broader spread in its ensemble representation, implying greater vertical variability than observed.

The corresponding bias profiles (Fig. 4b) further indicate that both configurations show cold biases in the upper layer and warm biases below the thermocline. However, ZSTAR generally shows larger absolute biases, exceeding -4 °C locally in the upper 40 m, whereas HYBRID shows relatively smaller biases throughout the water column. Although the differences between the two configurations are not dramatic, these results suggest that HYBRID provides a somewhat improved representation of the sharp summer stratification and the associated thermal structure in this shelf-sea environment.

The spatial distribution of the depth-averaged buoyancy frequency in February and August indicates notable seasonal differences in stratification between the HYBRID and ZSTAR configurations (Fig. 5). In February, both configurations show low buoyancy frequency across most of the Yellow Sea, reflecting weak stratification associated with strong wintertime vertical mixing (Fig. 5a–b). In contrast, the buoyancy frequency increases substantially in August (Fig. 5d–e), particularly in the central Yellow Sea regions as well as in Bohai Bay and the Yangtze River Estuary, reflecting the development of a seasonal thermocline under surface heating and freshwater-induced stratification. The difference maps (Fig. 5c and f) highlight the systematic variations between the two configurations. In August, although HYBRID shows slightly weaker stratification in some localized regions, it produces higher buoyancy frequency in areas where summer stratification is strongly developed, particularly where freshwater input intensifies stratification, indicating that HYBRID captures the enhanced seasonal stratification more effectively than ZSTAR. This contrast is less pronounced in February, which is consistent with the generally well-mixed state of the winter water column. The stronger summer stratification simulated by HYBRID in the central Yellow Sea and near the Korean Coast is likely more reliable because HYBRID demonstrates improved agreement with the observed temperature profiles and reduced bias in these regions (Figs. 3 and 4). Given that the buoyancy frequency is directly derived from the vertical temperature gradient, the better simulation of the temperature fields by HYBRID suggests that its representation of stratification in these areas is also more accurate. In contrast, no direct observational comparison is conducted for the Yangtze River Estuary region in this study. Therefore, although HYBRID simulates stronger stratification in this area, this result should be interpreted with caution. Nevertheless, the enhanced buoyancy frequency may reflect the improved capability of HYBRID to capture freshwater-induced density gradients near river outflows, a behavior that warrants further investigation using dedicated observational datasets. Collectively, these results suggest that the choice of vertical coordinate system can significantly influence the simulation of stratification in shelf regions, particularly under strong surface forcing and freshwater input.

Barotropic tides, which represent the nearly depth-independent component of tidal motion, are the dominant tidal signals across the shallow Yellow Sea, and play a key role in driving horizontal transport and bottom frictional dissipation. Their accurate simulation is essential for resolving coastal sea level variability, tidal mixing, and energy dissipation processes. We evaluate the performance of the HYBRID and ZSTAR vertical coordinate systems in simulating the semidiurnal M2 and diurnal K1 barotropic tides. In addition, barotropic energy fluxes are diagnosed specifically for the dominant M2 constituent to examine how each configuration captured the generation and propagation of tidal energy across the Yellow Sea Basin, as well as the differences in the magnitude and spatial distribution of energy transport.

The M2 tidal amplitudes and phases in the Yellow Sea from the HYBRID and ZSTAR configurations are compared with those from the TPXO9 tidal model (Fig. 6). Both configurations broadly capture the amphidromic system centered near 124° E, 35° N, as well as the large-scale amplitude distribution. However, HYBRID shows better agreement with TPXO, achieving a lower RMSE (24.19 cm) and higher spatial correlation (0.76) than ZSTAR (RMSE: 32.55 cm, Corr: 0.67), indicating improved tidal performance. In terms of amplitude bias (Fig. 6d–f), HYBRID simulated higher M2 amplitudes than ZSTAR across much of the basin, with the largest differences appearing along the Korean coast. Compared with TPXO, HYBRID exhibits smaller overall errors, whereas ZSTAR tends to underestimate amplitudes near the Korean Peninsula, with errors exceeding -40 cm in some regions.

The diurnal K1 tide exhibited a similar pattern (Fig. 7). Both HYBRID and ZSTAR broadly reproduce the observed amphidromic structure centered near 123.5° E, 34° N, as represented in the TPXO. However, HYBRID demonstrates better agreement in terms of amplitude and phase, achieving a lower RMSE (4.72 cm) than ZSTAR (5.70 cm), indicating improved tidal accuracy. Amplitude difference maps reveal that ZSTAR tended to underestimate K1 amplitudes across much of the basin, with negative biases reaching up to -10 cm in several regions.

Because the M2 tide is the dominant tidal constituent in the Yellow Sea and exhibits the most pronounced differences between configurations, subsequent analyses focused on the M2 component for harmonic decomposition and comparison. The barotropic tidal energy flux vectors and magnitudes associated with the M2 constituent in the Yellow Sea show a similar large-scale propagation pattern in both HYBRID and ZSTAR during February and August (Fig. 8). In both configurations, tidal energy propagates from southeast to northwest and dissipates over the shallow shelf. In February (Fig. 8a–b), when the water column is well mixed, the overall energy pathways are broadly similar between the two configurations. However, HYBRID exhibits systematically stronger flux magnitudes across all four tidal regions (Table 2). Quantitative comparisons show that HYBRID produced a 13.5 % larger flux in the Yangtze River Estuary, 20.2 % larger flux in the South Yellow Sea, 27.6 % larger flux along the South Korean Coast, and 42.3 % larger flux along the North Korean Coast, relative to ZSTAR. In August (Fig. 8d–e), when the water column becomes strongly stratified, both configurations simulated a general weakening of the barotropic energy flux, especially in the central basin, which may be attributed to the enhanced energy conversion from barotropic to baroclinic tides under strong summer stratification (Kang et al., 2002). Nevertheless, HYBRID continues to produce stronger energy flux magnitudes than ZSTAR across all four regions, yielding a 17.2 % higher flux in the Yangtze River Estuary, 23.5 % higher flux in the South Yellow Sea, 24.8 % higher flux along the South Korean Coast, and 51.1 % higher flux along the North Korean Coast. These results further support the hypothesis that the HYBRID configuration preserves barotropic tidal energy more effectively under both weakly and strongly stratified conditions.

Although barotropic tides dominate shallow regions of the Yellow Sea, internal (baroclinic) tides play critical roles in subsurface energy propagation and vertical mixing, particularly during periods of strong stratification. Given the marked seasonal stratification in the Yellow Sea, especially in summer, internal tide generation and propagation are expected to vary significantly with the season and vertical coordinate representation. To assess the performance of the model in simulating baroclinic tides, we analyze diagnostics such as the baroclinic KE and energy flux associated with the M2 constituent.

The spatial distribution of the depth-integrated baroclinic KE for the M2 tidal constituent in the Yellow Sea for February and August is compared between the HYBRID and ZSTAR configurations (Fig. 9). In both seasons, pronounced baroclinic KE maxima consistently appear near the Yangtze River Estuary and in regions characterized by steep topographic gradients. These areas are well-known hotspots for internal tide generation associated with strong barotropic-to-baroclinic energy conversion over rough topography. In addition, the Yellow Sea lies close to the critical latitude (∼ 30° N) of the diurnal K1 tide, where the tidal frequency approaches the local inertial frequency, favoring enhanced internal wave activity (Dong et al., 2019). Consequently, enhanced baroclinic KE is also observed around 30–32° N. Such dynamical conditions can further intensify internal wave responses and contribute to the amplification of M2 internal tides, a tendency that is also evident in Fig. 9.

In February (Fig. 9a–c), the baroclinic KE remains relatively low across most of the basin owing to weak stratification. However, the difference maps (Fig. 9c) reveal that ZSTAR tends to produce a higher baroclinic KE than HYBRID over much of the southern shelf, quantitatively exhibiting approximately 19.5 % and 23.6 % greater KE in the Yangtze River Estuary and South Yellow Sea, respectively (Table 3). Along the South Korean and North Korean coasts, the differences are smaller, with nearly identical values in the latter region. Given the overall weak stratification in winter, the enhanced baroclinic KE in ZSTAR might be attributable to spurious energy conversion or numerical noise rather than physically realistic internal tide processes. In contrast, the August distribution (Fig. 9d–f) shows a marked increase in the baroclinic KE, which is consistent with the development of strong summer stratification and active internal tide generation. During this season, HYBRID simulates a higher baroclinic KE than ZSTAR across all four evaluation regions. HYBRID produces 21.7 % greater KE in the South Yellow Sea and 33.6 % greater KE along the North Korean Coast and exceeded the ZSTAR in the Yangtze River Estuary by 9.2 % and 5.0 % on the South Korean Coast. These regional enhancements align with areas of elevated buoyancy frequency (Fig. 5f), particularly along the Korean coast and estuarine fronts, supporting the interpretation that HYBRID captures the barotropic-to-baroclinic energy conversion processes more effectively under strong stratification conditions.

The baroclinic tidal energy flux vectors associated with the M2 tide in the Yellow Sea during February and August reveal clear seasonal differences (Fig. 10). In February, both configurations simulate relatively weak and spatially limited fluxes, consistent with the low baroclinic KE shown in Fig. 9a–b. Most energy fluxes are localized near steep topographic features, particularly along the southern Korean coast and near the Yangtze River Estuary. HYBRID and ZSTAR exhibited broadly similar patterns, although ZSTAR exhibited slightly more scattered flux vectors. In August (Fig. 10c, d), when stratification is strong, the differences between the two configurations become more pronounced. HYBRID generates stronger and more spatially extensive baroclinic energy fluxes throughout the central Yellow Sea, along the Korean coastal shelf, and near the estuarine front of the Yangtze River. These areas coincide with the enhanced baroclinic KE shown in Fig. 9. Coherent flux directions and widespread coverage indicate more efficient internal tide generation and propagation in HYBRID. In contrast, ZSTAR produces weaker, patchier fluxes with reduced spatial reach, suggesting a limited conversion of barotropic to baroclinic energy.

To explicitly assess how internal tides, manifest in the vertical density structure, we analyze the time evolution of the potential density and surface M2 amplitude over a 24 h period (Fig. 11) at a central location in the Yellow Sea (124° E, 36° N) during summer (8 August). This location is selected because it features a relatively deep bathymetry within the Yellow Sea and is influenced by the YSBCW, which plays a key role in the propagation and trapping of internal tides during the stratified summer.

In the HYBRID simulation (Fig. 11a), clear internal tidal signals emerge throughout the stratified water column. The M2 oscillations are evident at depths of approximately 10–15 and 20–30 m, consistent with the internal gravity wave-induced vertical displacements of the isopycnals. The stratification in this region supports the trapping and vertical propagation of baroclinic wave energy, and the alternating isopycnal deflections exhibit a well-organized structure. Coherent oscillations also appear near the bottom (below 50 m), indicating downward propagation of the internal tide energy and its interaction with the YSBCW layer. These vertical fluctuations are largely in phase with the surface M2 cycle, suggesting efficient barotropic-to-baroclinic energy conversion and vertical energy transfer across the water column. In contrast, the ZSTAR configuration (Fig. 11b) shows significantly weaker vertical displacements of the isopycnals. Although the M2-period oscillations are subtly visible in the upper 30 m, they are less coherent and shallower in penetration than those in HYBRID. The lower portion of the water column remained almost undisturbed, and no significant internal tide signatures are observed near the bottom despite the presence of stratification. This indicates that the internal tidal energy generated at the surface or slope regions does not efficiently propagate downward.

To further clarify the underlying mechanisms responsible for the differences in the baroclinic energy intensity and vertical structure between the two configurations, we examine the barotropic-to-baroclinic energy conversion rate, which directly quantifies the generation of internal tidal energy from barotropic tidal forcing in the presence of stratification and bathymetry. As a key diagnostic tool, this term reveals where and how efficiently the barotropic tidal energy is transformed into baroclinic motion across the model domain. Figure 12 illustrates the spatial distribution of the conversion rates for February and August, highlighting the localized regions of strong energy transfer, particularly along steep topographic gradients and near shelf break zones.

In February, when stratification is weak, both HYBRID and ZSTAR exhibit spatially patchy and generally weak barotropic-to-baroclinic conversion across the Yellow Sea (Fig. 12a–c). Negative or near-zero conversion dominates most of the basin, indicating that wintertime internal-tide generation is highly suppressed under weak stratification. Nevertheless, the Yangtze River Estuary remains a primary hotspot due to strong tidal forcing and steep bathymetry, where both configurations show similarly large conversion magnitudes (143.85 MW in HYBRID vs. 144.84 MW in ZSTAR; Table 4). Outside the estuary, however, regional contrasts emerge: ZSTAR produces noticeably stronger conversion in the South Yellow Sea (+30.5 % relative to HYBRID) and more than triple the conversion along the South Korean Coast (+292.2 %), while HYBRID shows slightly greater conversion near the North Korean Coast. Despite these differences, the basin-wide baroclinic KE remains similarly weak in both configurations (Table 3), underscoring that winter internal-tide energetics are primarily controlled by bathymetric forcing rather than stratification-driven processes. As a result, both configurations convert barotropic tidal energy at broadly comparable levels during winter.

In contrast, in August, strong stratification substantially enhances barotropic-to-baroclinic conversion in both configurations (Fig. 12d–e). The Yangtze River Estuary and South Yellow Sea remain the dominant generation hotspots; however, the performance contrast between HYBRID and ZSTAR becomes far more pronounced under these strongly stratified summer conditions. Compared to HYBRID, ZSTAR exhibits a 15.0 % increase in conversion in the Yangtze River Estuary and nearly double (+94.5 %) the conversion in the South Yellow Sea. Along the South Korean Coast, ZSTAR also shows 61.1 % stronger conversion, while conversion near the North Korean Coast shifts from negative values in HYBRID to a moderately positive level in ZSTAR (Table 4). Despite these substantially enhanced conversion magnitudes, ZSTAR produces weaker baroclinic KE and less coherent internal-tide propagation than HYBRID (Figs. 9–11), indicating that a large fraction of the baroclinic energy generated in ZSTAR is rapidly dissipated or trapped locally, rather than radiated outward as sustained internal tides.

This contrasting behavior can be more clearly interpreted through the baroclinic energy budget. Although ZSTAR generates substantially greater barotropic-to-baroclinic conversion in August, the resulting baroclinic KE and baroclinic energy flux remain weaker than those in HYBRID (Figs. 9–11), demonstrating that only a limited portion of the converted energy is radiated away as coherent internal tides. From the depth-integrated and time-mean baroclinic energy balance, <∇H⋅Fbc>=<C>-<D>, where ∇H⋅Fbc is the divergence of baroclinic energy flux, C represents the barotropic-to-baroclinic conversion, and D denotes the dissipation of baroclinic tidal energy.

The strong local damping of internal tide energy in the Yellow Sea has also been reported in previous studies. For example, Liu et al. (2019) demonstrated that internal-tide dissipation is nearly equivalent to barotropic-to-baroclinic conversion over most of the basin, indicating that only a small fraction of the converted energy can radiate away from the generation regions. Our results exhibit a highly consistent behavior: because the divergence of the baroclinic energy flux remains relatively weak over the shallow and frictionally dominated Yellow Sea, the spatial pattern of dissipation <D>=<C>-<∇H⋅Fbc>, closely follows that of conversion in both configurations. Therefore, we do not present dissipation maps separately, as these would provide redundant information without adding physical insight. Instead, we assess dissipation through the regional baroclinic energy budget, which clearly indicates that when the radiated baroclinic energy flux remains weak, the greater conversion in ZSTAR inevitably appears as a larger dissipation term. This implies that much of the additional baroclinic energy generated in ZSTAR is rapidly removed locally, through enhanced damping or numerical mixing, rather than contributing to propagating internal tides. Consequently, HYBRID facilitates more coherent propagation of internal tides away from the generation regions, while ZSTAR loses a larger fraction of the converted energy locally due to limited radiation efficiency. These contrasting energy pathways form the basis for further investigation of the underlying physical drivers, which are examined in the Discussion section.