As shown by the magenta rectangle in Figure 1C , the study area is located to the east of Sishili Bay in the northeastern coastal waters of the Shandong Peninsula. In this region, the topography is basically flat in most offshore areas with water depth being approximately 20.0 m and is sloping approaching the shoreline. The shoreline is roughly zonal. To better understand current variations in the northeastern nearshore waters of the Shandong Peninsula, we used observations obtained from the submarine cable real-time observation system (SCROS) in the Yutai marine ranching area (MR; 121.93°E, 37.47°N, Figure 1B ). The SCROS was located at the sea bottom and was approximately 1.6 km away from the shoreline (Zhai et al., 2020a). The SCROS was equipped with a HydroCAT conductivity-temperature-depth-oxygen recorder and an acoustic Doppler current profiler. The conductivity-temperature-depth-oxygen recorder and acoustic Doppler current profiler were at distances of approximately 20 cm and 60 cm vertically away from the sea bottom. The SCROS continuously measured the bottom layer temperature, salinity, depth, and profiles of current speed and direction. The SCROS also transmitted data in a real-time mode (Zhai et al., 2020b, 2021b). Current profiles had a vertical sampling interval of 1.0 m, and the deepest layer was at approximately 1.0 m vertically away from the sea bottom. The SCROS observations lasted two winters of 2019/2020 (December 1, 2019 to February 29, 2020) and 2020/2021 (December 1, 2020 to February 28, 2021). The time intervals were 10 minutes for the conductivity-temperature-depth-oxygen observations and 1 hour for the acoustic Doppler current profiler observations.

In addition, we also used two kinds of satellite-observed sea surface temperature (SST) data sets to validate the model simulations. One data set was the level 4 high-resolution daily SST produced by the operational SST and sea ice analysis system (OSTIA; Donlon et al., 2012). The OSTIA data set was generated through merging satellite observations and field observations using the optimum interpolation method. It is available from April 2006 to the present and over the global ocean with a horizontal resolution of 0.05°. The other data set was the daily Multi-scale Ultra-high Resolution SST (MURSST; Chin et al., 2017). The MURSST data set was generated through merging field observations and three types of satellite observations using a meshless multi-scale interpolation method (Chin et al., 2017). The satellite observations included infrared SST retrievals at a high resolution of approximately 1 km, radiometer SST retrievals at a medium resolution of approximately 4−8.8 km, and microwave SST retrievals at a nominal resolution of 25 km. The MURSST data set is available from June 2002 to the present and over the global ocean with a horizontal resolution of 0.01°.

In the current study, we adopted the Semi-implicit Cross-scale Hydroscience Integrated System Model (SCHISM; Zhang et al., 2016). SCHISM is an open-source community-supported modeling system designed for seamless simulations of three-dimensional baroclinic circulations across creek-lake-river-estuary-shelf-ocean scales. It uses unstructured mixed triangular/quadrangular grid in the horizontal dimension and Localized Sigma Coordinates with Shaved Cell (Zhang et al., 2015) in the vertical dimension. Utilizing semi-implicit time-stepping, SCHISM can be implemented with remarkably high spatial resolutions and free from the Courant-Friedrichs-Lewy constraints, and thus has high numerical efficiency in coastal applications. SCHISM has been successfully applied in understanding the three-dimensional circulation and its effects on coastal marine ecosystem in the Bohai Sea in summer (Wu et al., 2023b, 2023c).

As shown in Figure 1B , the model domain covered the Bohai Sea and Yellow Sea north of 36°N. The horizontal unstructured grid was generated with the OceanMesh2D package (Roberts et al., 2019) and included 42290 elements and 22337 nodes. Model grids were depth-dependent and had finer resolutions in shallow/coastal waters than in deep/open oceans. The horizontal resolutions generally ranged from 8.0 km in the deep ocean to approximately 2.0 km in coastal oceans. In the study area ( Figure 1C ), model grids were further refined, with horizontal resolutions ranging from 1.0 km in offshore waters to approximately 0.6 km in nearshore waters. The number of vertical layers decreased from 26 in the deep waters to 20 in the shallow waters.

The model topography was interpolated from the 1 arc-minute global relief model of Earth’s surface and corrected by a regional nautical chart data set. The simulation period was from November 2018 to March 2021. The atmospheric forcing variables were derived from the fifth-generation atmospheric reanalysis of the global climate produced by the European Center for Medium-Range Weather Forecasts (ERA5; Hersbach et al., 2020; Wu et al., 2020) and included hourly 10-m wind vectors, surface heat fluxes, and surface freshwater fluxes. ERA5 variables are available from January 1940 to the present with a horizontal resolution of 0.25°. Initial values of water temperature, salinity, and sea surface height (SSH) were interpolated from the global HYbrid Coordinate Ocean Model reanalysis data set. Daily water temperature, salinity, SSH, and current vectors at the open boundary (36°N) were also interpolated from the HYbrid Coordinate Ocean Model reanalysis data set. HYbrid Coordinate Ocean Model variables were available from January 1994 to the present with a horizontal resolution of 0.08°. Eight major tidal components, including S₂, M₂, N₂, K₂, K₁, P₁, O_1, and Q₁ obtained from FES2014 (Lyard et al., 2021), were used in the model.

Because nearshore currents were strong daily variations and sea surface winds were complex synoptic variable in the Bohai Sea and North Yellow Sea in winter, conceptual experiments were adapted to better illustrate dynamic processes of two-layer overturning currents. Because winds and sea level slope of CTWs were primarily impact factors in all events shown in the below section, experiments with idealized winds focused on ocean currents driven by local winds and by the propagation of CTWs in the Bohai Sea and North Yellow Sea were adapted. We used the wind forcing barotropic model according to the previous study (Hsueh and Pang, 1989; Takahashi et al., 1995; Lin and Yang, 2011). Model domain and grid in experiments were the same as those in the simulation aforementioned. Sea water temperature and salinity were constant. Water flux was zero at the open boundary. Tide and river variables were ignored. Atmospheric forcing was only composed of sea surface winds.

Three series of experiments with idealized sea surface winds were considered. Firstly, model domain was divided into four regions ( Figure 2A ). In each experiment, sea surface winds were set to 0 except winds in the selected region. These experiments were conducted to check whether the two-layer overturning currents were driven by local winds in the study area or the propagation of CTWs around the Shandong Peninsula. Then, we changed directions of sea surface winds. These experiments were conducted to check that propagations of CTWs were robust phenomena in the Bohai Sea and North Yellow Sea, no matter where sea surface winds blew. Finally, the evolutions of ocean currents in the study area were explored, including local wind-driven currents, currents driven by the propagation of CTWs, and currents driven by local winds and the propagation of CTWs.

Note that there were strong northerly winds and weak southerly winds in the Bohai Sea and North Yellow Sea in winter. Sea surface winds set up in the control run were as follows:

where d represented one day. u 10 and v 10 were zonal and meridional components of the 10 m wind ( Figure 2B ). We choose three times (t_1/2, t_1, t_3/2), which equally divided the period of southerly winds. Three times (t_5/2, t_3, t_7/2) equally divided the period of northerly winds. Note that sea surface winds were spatially constant in the wind zone. Sea surface wind speeds in Case I were setup as:

where d represented one day and u represented sea surface wind speed ( Figure 2C ). Sea surface wind speed in Case II were set as:

where d represented one day and u represented wind speed ( Figure 2D ). The list of information about three series of experiments were provided in Table 1 .

Performance of the model simulation was validated with observations using three statistical parameters, which were simultaneous linear correlation coefficient (R), mean absolute bias (MAB), and root mean square error (RMSE). For time series or space maps of X and Y, R = X ' Y ' / X ' 2 Y ' 2 , M A B = | X − Y | , and R M S E = ( X − Y ) 2 , where X ' and Y ' were anomalies, and 〈 〉 denoted the time mean. Simultaneous linear correlation coefficients were all significant above the 95% confidence level unless otherwise specified.

Model results were evaluated with the satellite observations, atmospheric reanalysis, and SCROS observation. As shown in Figure 3 , climatological winter mean SST in the model simulation showed high spatial correlations with those in the OSTIA (R=0.90), ERA5 (R=0.91) and MUSST (R=0.89). All four products clearly showed a warm tongue extending northward in the central area and cold waters in the coastal areas in the Yellow Sea in winter ( Figures 3A–C ). In the majority of the study area, SSTs in the model simulation showed high correlation coefficients (>0.9), low mean absolute biases (<1.0°C), and low root mean square errors (<1.0°C) with those in the other three ocean products ( Figures 3D–L ). Relatively large mean absolute biases and root mean square errors mainly occurred in the nearshore areas, possibly due to that both the model simulation and satellite observations contained certain errors in such shallow waters near the land. However, we noted that both mean absolute biases and root mean square errors were much smaller in the northeastern coastal waters of the Shandong Peninsula and Qinhuangdao than in other coastal waters of the Bohai Sea and Yellow Sea. This finding indicated that model-simulated SSTs agreed with satellite-observed SSTs and ERA5-reanalyzed SSTs in both magnitudes and spatiotemporal variations in the study area.

During the observation period, the SCROS stayed at the sea bottom. Therefore, the observed water depth variations were equivalent to SSH variations. In the current study, we calculated SSH anomaly (SSHA) as water depth minus its time mean with the SRCOS observation and as SSH minus its time mean with the model simulation. In the study area, hourly SSHA was dominated by tidal variations (Chen et al., 2019). As shown in Figure 4 , hourly (daily) SSHA in the model simulation had a R=0.77 (0.88), MAB=0.30 m (0.07 m), and RMSE=0.35 m (0.09 m) with those in the SCROS observation. This finding indicated that both hourly and daily SSHAs in the model simulation agreed with those in the SCROS observation.

The model-simulated tidal currents agreed with the SCROS-observed tidal currents. Figures 5A–D further compared the daily currents in the SCROS observation and model simulation. Overall, model-simulated daily currents agreed with observed daily currents in temporal variations, with simultaneous linear correlation coefficients ranging 0.56−0.74 for zonal currents and 0.50−0.76 for meridional currents. Zonal currents mostly had the same sign throughout the water column, but their speeds basically decreased from surface to bottom. On the contrary, meridional currents generally had a two-layer overturning structure, with opposite signs in the upper and lower layers (Hu et al., 2021; Zhai et al., 2021c). Model-simulated meridional currents mostly agreed with observed meridional currents in magnitude throughout the water column except in the surface layer where the model results were much smaller than the observations. The discrepancy between the simulated and observed surface meridional currents was possibly because that the model was unable to well resolve the surface current variations at the SCROS site which was so close to the shoreline ( Figure 1C ).

Figures 5E–H presented the daily time series of surface and bottom currents in the model simulation and SCROS observation. In the surface layer, the simulated and observed zonal (meridional) currents had R=0.74 (0.75), MAB=0.06 m s⁻¹ (0.09 m s⁻¹), and RMSE=0.08 m s⁻¹ (0.12 m s⁻¹). In the bottom layer, the simulated and observed zonal (meridional) currents had R=0.56 (0.75), MAB=0.01 m s⁻¹ (0.01 m s⁻¹), and RMSE=0.02 m s⁻¹ (0.01 m s⁻¹).

Based on the above comparisons, we could conclude that the model simulation was reliable in examining the nearshore circulation dynamics off the northeastern Shandong Peninsula in the Yellow Sea in winter.

As shown in Figure 5 , there were strong daily variations in ocean currents in the Yutai marine ranching area off the northeastern Shandong Peninsula. In the following, days with surface meridional currents in the Yutai marine ranching area being smaller than −0.71 m s⁻¹ were defined as Period I, while those with surface meridional currents in the Yutai marine ranching area being larger than 0.56 m s⁻¹ were defined as Period II. Figure 6 presented the composite maps of SSH and current vectors during Period I and Period II to examine the three-dimensional structure of overturning currents in the study area. As shown in the figure, ocean currents basically decreased approaching the shoreline but their directions were quite different in the surface and bottom layers in both periods. In Period I, SSH basically increased approaching the shoreline and southeastward currents prevailed in the surface layer in the whole study area ( Figure 6A ). On the contrary, ocean currents shifted from being eastward in offshore areas to being northeastward in nearshore areas in the bottom layer. This finding indicated that the zonal components of ocean currents were mostly eastward in both the surface and bottom layers but the meridional components of ocean currents showed an overturning structure in nearshore regions. To give more details of the overturning current structure, we calculated the composited ocean currents along the meridional section at 121.92°E and showed the results in Figure 6C . As clearly seen from the figure, overturning currents in nearshore waters in Period I were composed of shoreward currents in the upper layer, downwelling near the shore, and seaward currents in the lower layer.

In Period II, however, SSH basically increased away from the shoreline and northeastward currents prevailed in the surface layer in the whole study area ( Figure 6B ). In the bottom layer, ocean currents shifted from being westward in offshore areas to being southwestward in nearshore areas. Clearly, bottom layer currents in Period II were nearly opposite to those in Period I. This finding indicated that ocean currents experienced more significant vertical changes in Period II than in Period I. Not only the meridional components but also the zonal components of ocean currents had opposite directions in the surface and bottom layers in Period II, which were different from the vertical structure of ocean currents in Period I. As clearly seen from Figure 6D , overturning currents in nearshore waters in Period II were composed of seaward currents in the upper layer, upwelling near the shore and shoreward currents in the lower layer.

Figure 7 presented the composite vertical profiles of ocean currents at 121.92°E, 37.65°N in the offshore area and at 121.92°E, 37.5°N in the nearshore area off the northeastern Shandong Peninsula during Period I and Period II. Overall, current speeds shared the same vertical structure in the whole study area in both periods. With increasing depth, current speeds first decreased from being largest at the sea surface to a local minimum in the middle layer, and then increased to a secondary maximum near the sea bottom. In contrast, current directions showed opposite vertical structures in the two periods. With increasing depth, current directions rotated counterclockwise in Period I but rotated clockwise in Period II. The rotation of current directions with increasing depth became more significant approaching the shoreline. The above findings indicated that ocean currents in different layers in the two periods were controlled by different mechanisms.

In winter, nearshore waters of the Yellow Sea and Bohai Sea are vertically mixed throughout the water column (Zhai et al., 2021c; Liu et al., 2023). As a result, ocean currents in the nearshore waters are primarily driven by sea surface wind and pressure gradient, which results from the horizontal variations in SSH and water density. In order to check the impacts of sea surface wind and horizontal pressure gradient on ocean currents, we firstly calculated the zonal ( u g e o ) and meridional ( v g e o ) geostrophic currents at the depth of h as u g e o = − g f ∂ η ∂ y − g f ρ 0 ∫ − h η ∂ ρ ∂ y d z and v g e o = g f ∂ η ∂ x + g f ρ 0 ∫ − h η ∂ ρ ∂ x d z , where positive x, y and z were eastward, northward and upward, respectively. f was the Coriolis parameter, g = 9.8 m s⁻² was gravity acceleration, η was SSH, ρ 0 = 1025 kg m⁻³ was the reference density of seawater and ρ was water density. The two terms on the right side of each equation denoted geostrophic currents caused by horizontal variations in SSH and water density, respectively. Then we calculated simultaneous linear correlation coefficients of hourly-averaged ocean currents with winds and with geostrophic currents in the zonal and meridional directions in Period I and Period II. A 25-hourly running mean filter was applied to hourly averaged data to remove tidal signals. Correlation coefficients between simulated currents and geostrophic currents caused by horizontal variations in water density were low and neglected. Figure 8 showed the simultaneous linear correlation coefficients in the surface and bottom layers. Note that geostrophic currents in Figure 8 were caused by horizontal variations in SSH. Correlations of simulated currents with geostrophic currents caused by horizontal variations in water density were weak and not shown.

In Period I in the surface layer, zonal currents in the study area were strongly positively correlated with zonal geostrophic currents, with meridional currents in the surface layer positively correlated with meridional winds. In the bottom layer, zonal currents in the study area were strongly positively correlated with zonal geostrophic currents. Meridional currents in the offshore region of the study area in the bottom layer were positively correlated with meridional geostrophic currents. Meridional currents in the nearshore region of the study area in the bottom layer were negatively correlated with meridional winds. The zonal ocean currents in nearly whole study area were strongly correlated with sea level slope, while the meridional currents in the nearshore region of the study area were strongly correlated with meridional winds. This finding suggested sea level slope in the study area was not primarily caused by winds.

In Period II in the surface layer, zonal currents in the study area were strongly positively correlated with zonal geostrophic currents, with meridional currents in the surface layer positively correlated with meridional winds. In the bottom layer, zonal currents in the study area were strongly positively correlated with zonal geostrophic currents. Meridional currents in the offshore region of the study area were positively correlated with meridional geostrophic currents. Meridional currents in the nearshore region of the study area were negatively correlated with meridional winds. Distributions of simultaneous linear correlation coefficients in Period II were similar to those in Period I. However, positive correlation coefficients between zonal ocean currents and zonal geostrophic currents were significantly stronger in Period I than those in Period II. This finding suggested sea level slope might cause aforementioned distinct vertical alterations of horizontal currents in the two periods.

Based on the above results, we could conclude that zonal ocean currents in the surface and bottom layers of the study area and bottom meridional currents in the offshore region of the study area were primarily correlated with sea level slope in the two periods. Surface meridional currents in the study area and bottom meridional currents in the nearshore region of the study area were primarily correlated with meridional winds.

As shown in Figure 8 , strong correlation coefficients suggested that sea level slope and winds were the robust impact factors in all two-layer ocean current events in the two periods. To gain a straight understanding of the currents driven by sea level slope and winds, it was necessary to investigate momentum budgets during a single event in Period I (Event I) and Period II (Event II). The zonal and meridional momentum balance equations with corresponding abbreviations were shown as follows:

where u and v were the zonal and meridional velocities and positive x, y, and z were eastward, northward, and upward, respectively. ρ was the water pressure, p was the water density, and A h and A z were the horizontal and vertical viscosity coefficients, respectively. The abbreviation ACCEL denoted the local acceleration term, PGF was the horizontal pressure gradient force, ADV was the three-dimensional transport term, COR was the Coriolis force, HVIS and VVIS were horizontal and vertical viscosities, respectively. 〈 〉 x and 〈 〉 y denoted the zonal and the meridional component, respectively. HVIS term was ignored because it was small. A 25-hourly running mean filter was applied to hourly averaged data to remove tidal signals. After removing tidal signals from the data, we calculated all the above terms.

Figure 9A displayed sea surface winds, surface currents, and bottom currents in the Yutai marine ranching area in Event I. On 5 January 2020, sea surface winds blew westward. Surface currents were westward and bottom currents were southward. Subsequently, both surface winds and surface currents rotated counterclockwise. On 7 January, there were strong northwesterly winds, surface southeastward currents, and bottom northeastward currents. After that, both surface winds and ocean currents weakened.

Strong surface southward and bottom northward currents occurred on 7 January 2020 in Event I, showing two-layer ocean currents that were similar to composite results in Period I ( Figure 9A ). Before that on 5 January, surface westward currents were driven by easterly winds (negative surface zonal VVIS terms in Figure 9C ). Then winds shifted counterclockwise and drove surface southwestward Ekman currents, which were at the right angle to the wind by surface northwestward Coriolis forces ( Figure 9C ; Ekman, 1905). At that time, westward sea level slope increased rapidly, which resulted in the rapid strengthening of eastward PGF ( Figures 9C, E ). On 7 January, the strong eastward PGF turned westward currents to the eastward in the surface and bottom layers, which were at the left angle of northerly winds ( Figures 9C, E ). As winds shifted southeastward, both surface eastward VVIS and PGF drove eastward currents, indicating that Ekman transport was not the main impact of westward sea level slope ( Figures 9C, E ). Meanwhile, southward sea level slope increased ( Figure 9G ). Note that northward PGF caused by southward sea level slope was stronger than surface southward VVIS caused by northerly winds and surface southward COR caused by eastward currents, so that northward PGF slowed down surface southward currents ( Figure 9G ). The strong northward PGF also drove bottom northward currents ( Figure 9I ). On 8 January 2020, sea level slope turned southeastward and slowed down eastward currents ( Figures 9C, E ). Meanwhile, southward sea level slope decreased rapidly so that surface southward currents increased by northerly wind and bottom northward currents weakened by bottom friction ( Figures 9G, I ). After that, ocean currents decreased as both wind and sea level slope decreased.

Figure 9B displayed sea surface winds, surface currents, and bottom currents in the Yutai marine ranching area in Event II. At 12:00 on 18 January 2021, there were northerly winds, surface southward currents, and bottom northward currents ( Figure 9B ). Then surface winds and surface currents rotated clockwise. Sea surface winds shifted to the southeasterly at 22:00 on 19 January 2021, and surface currents shifted to be northward and bottom currents shifted to be southwestward. Then, both sea surface winds and ocean currents weakened.

Strong surface northward and bottom southwestward currents occurred at 22:00 on 19 January 2021 in Event II, showing two-layer ocean currents that were similar to composite results in Period II ( Figure 9B ). Before that at 12:00 on 18 January, there were surface southward VVIS caused by northerly winds and northward PGF in the surface and bottom layers caused by southward sea level slope, with PGF stronger than VVIS ( Figures 9H, J ). From 12:00 on 18 January to 00:00 on 19 January, winds shifted clockwise to northeasterly and drove surface southeastward Ekman currents, which were at the right angle to the wind by surface northwestward Coriolis forces ( Figure 9D ). At that time, southward sea level slope decreased rapidly ( Figures 9H, J ). Bottom southward friction decreased bottom northward currents so that bottom eastward COR weakened. Bottom westward friction drove bottom westward currents with bottom eastward COR decreasing and caused bottom northward COR. Then sea level slope turned northward. Bottom southward PGF and bottom southward friction turned bottom northward currents to southward, which caused bottom westward COR. This suggested that bottom Ekman layer was regulated by bottom COR and bottom frictions when sea level slope changed ( Figures 9F, J ). From 00:00 on 19 January to 22:00 on 19 January 2021, surface southeasterly winds drove northwestward Ekman currents ( Figure 9D ). At that time, northward sea level slope increased. The strong sea level slope caused strong southward PGF, which prevented the development of surface northward currents and strengthened bottom northward currents ( Figures 9H, J ). After 22:00 on 19 January 2021, southeasterly winds weakened rapidly ( Figures 9D, H ). Strong southward PGF slowed down surface northward currents ( Figure 9H ). Bottom friction weakened bottom southward currents as northward sea level slope decreased ( Figure 9J ).

Momentum terms demonstrated that ocean currents were dominated by winds and horizontal pressure gradient in both Event I and Event II and proved that horizontal pressure gradient caused by sea level slope in the study area was not primarily caused by winds. Therefore, ocean currents and sea level were displayed to find the factors that drove sea level slope in Event I and Event II. Figures 10A–E displayed surface winds, depth-averaged currents, and sea level characteristics in Bohai Sea and North Yellow Sea in Event I. On 5 January 2020, easterly winds drove water converging in the western part of the Bohai Sea ( Figure 10A ). From 6 January to 8 January as winds turned counterclockwise, high sea level moved eastward along northern coastline of the Shandong Peninsula ( Figures 10B–D ). At this time, depth-averaged currents were southeastward, almost parallel to sea level contours. Sea level decreased with contours parallel to coastline as the offshore distance increased, exhibiting the propagation of CTWs (Wang et al., 2020; Li and Huang, 2019). After 9 January, high sea level moved to the South Yellow Sea, there were northward currents in the center of the North Yellow Sea resulting in sea level rising ( Figure 10E ).

Figures 10F–J displayed surface and bottom currents and SSH in the study area ( Figure 1C ) in Event I. Sea surface winds in the study area were almost consistent with contemporary sea surface winds in the North Yellow Sea and not shown. On 5 January 2020, there were surface northwestward currents and bottom southwestward currents with southeasterly winds ( Figure 10F ). From 6 January to 7 January, CTWs of high sea level propagated to the study area and caused westward sea level slope ( Figures 10G, H ). Westward sea level slope caused eastward PGF and turned westward currents to eastward currents in the surface and bottom layers. The propagation of CTWs raised sea level in the nearshore region of the study area rapidly, which caused southward sea level slope and strong northward PGF ( Figures 10H, I ). As westward currents turned eastward, both currents and sea level were generally in agreement with composite results. After 9 January, there were westward geostrophic currents caused by northward sea level slope when sea level rose in the center of the North Yellow Sea ( Figure 10J ).

Figures 10K–O displayed surface winds, depth-averaged currents, and SSH in the Bohai Sea and North Yellow Sea in Event II. There were CTWs of high sea level along the northern coastline of the Shandong Peninsula and northerly winds in the North Yellow Sea at 12:00 on 18 January ( Figure 10K ). From 12:00 on 18 January to 00:00 on 19 January, high sea level in the south part of North Yellow Sea moved eastward when winds in the North Yellow Sea rotated clockwise ( Figures 10K, L ). From 19 to 20 January, southeasterly winds in the Bohai Sea and North Yellow Sea drove water diverging in the south part of the Bohai Sea and North Yellow Sea, resulting in CTWs with low sea level along the northern coastline of the Shandong Peninsula ( Figures 10M, N ). Then CTWs with low sea level moved eastward when winds weakened, followed by CTWs with high sea level ( Figure 10O ).

CTWs with high sea level caused southward sea level slope in the study area at 12:00 on 18 January ( Figure 10K ). Then as high sea level moved eastward, southward sea level slope decreased, resulting in the decreasing of northward PGF ( Figure 10P ). At 19 January, CTWs with low sea level were generated by southeasterly winds in the Bohai Sea and North Yellow Sea ( Figures 10L, Q ). From 19 to 20 January, CTWs passed through the study area and caused southward sea level slope in the study area, resulting in strong northward PGF ( Figures 10R, S ). Meanwhile, surface and bottom currents were consistent with the composite results. After low sea level moved eastward, the high sea level gradually extended to the study area ( Figure 10T ).

Because nearshore currents were strong daily variations and sea surface winds were complex synoptic variable in the Bohai Sea and North Yellow Sea in winter, conceptual experiments with idealized winds focused on ocean currents driven by local winds and by the propagation of CTWs were adapted to better illustrate dynamic processes of two-layer overturning currents. Figure 11 displayed results of the control run. Low (high) sea level appeared and moved eastward along coastline of northeastern Shandong Peninsula with southerly (northerly) winds developing and relaxing, which was consistent with CTWs aforementioned. Additionally, there were surface northward currents and bottom southwestward currents at t_1/2 and t₁. There were surface southeastward currents and bottom northeastward currents at t₃ and t_7/2. The general characteristics of two-layer overturning currents in the study area and CTWs in the Bohai Sea and North Yellow Sea obtained from idealized model were almost consistent with those in realistic model presented in the previous sections.

Figure 12 displayed results of the region experiments. In Expa1 (Expa5) where sea surface winds in R1 were available, sea surface winds drove surface southward (northward) currents, resulting in nearshore water convergence (divergence) and sea level rising (dropping). Then, southward (northward) sea level slope drove bottom northward (southward) currents. The northerly (southerly) winds caused sea level in the southwestern area of R1 rising (dropping). There were geostrophic currents rotating clockwise (counterclockwise) around high (low) sea level at the left boundary of R1 and counterclockwise (clockwise) around low (high) water level at the right boundary of R1. In Expa2 (Expa6), the northerly (southerly) winds caused sea level at the south boundary of R2 rising (dropping). High (low) sea level extended eastward along the coastline. Even though there was no wind in R1, the nearshore sea level rose (dropped). The southward (northward) sea level slope drove strong northeastward (southwestward) currents in the bottom Ekman layer, whose horizontal currents turned anticlockwise and weakened from surface layer to bottom layer. Both sea level slope and ocean currents were strong in the nearshore region. In Expa3 (Expa7), the northerly (southerly) winds converged (diverged) the sea level at the south boundary of R3 so that there were weak southwestward (northeastward) currents in R1. In Expa4 (Expa8), currents in the study area were much smaller than those in other experiments, indicating that winds in R4 played little impact on sea water in R1. Ocean currents in R1 were mainly driven by winds in R1 and sea level slope from R2. Additionally, as wind reversed both wind-driven currents and sea level slope driven currents reversed.

Figure 13 showed sea level of CTWs in the Bohai Sea and North Yellow Sea in the experiments where wind directions were changed. Sea level rose (dropped) in the southern part of Bohai Sea and North Yellow Sea when northerly (southerly) winds developed. Then northerly (southerly) winds relaxed. low (high) sea level moved eastward along coastline. Meanwhile, high (low) sea level moved from northern part to western part of Bohai Sea. After that, high (low) sea level moved eastward following nearshore low (high) sea level. When easterly (westerly) winds developed, sea level rose (dropped) in the west area of Bohai Sea and dropped (rose) off the northeastern Shandong Peninsula. As easterly (westerly) winds relaxed, low (high) sea level moved eastward along coastline off the northeastern Shandong Peninsula followed by eastward high (low) sea level. Notably, as winds turned southerly, high sea level along the coastline of northeastern Shandong Peninsula was reconstructed. Then high sea level moved eastward as southerly winds relaxed.

The evolution of wind-driven currents and geostrophic currents caused by the propagation of CTWs were explored. Model results within 121.75−121.95°E and 37.5−37.65°N were averaged and smoothed to obtain current profiles. Model results at the northern and southern boundaries of the aforementioned rectangular area were selected to calculate meridional sea level difference (MSLD). The positive MSLD meant that offshore sea level was higher than nearshore sea level.

Figures 14A–C displayed MSLD and current profiles of Expc1. At the beginning, southerly winds drove surface northeastward Ekman currents. Sea level in the northeastern (southwestern) part of the study area rose (dropped) and MSLD was increased. Meanwhile, bottom currents were southwestward caused by northeastward sea level slope. Then southerly winds weakened, resulting in the decreasing of northeastward sea level slope. The MSLD decreased and bottom southward currents weakened with wind-driven surface northeastward currents gradually extending to bottom layer. Sea level and currents under northerly winds were opposite to those under southerly winds.

Figures 14A, D, E displayed MSLD and current profiles of Expc2. MSLD in the study area increased when CTWs generated by southerly winds and strong westward currents occurred. Then positive MSLD and westward currents slowly decreased, even though southerly winds decreased and reserved. Subsequently, when northerly winds were strong, MSLD decreased rapidly and turned local westward currents to eastward. When the northerly winds weakened, high sea level moved eastward.

As shown in Figures 14A, F, G , MSLD and ocean currents in the study area were formed by those driven by winds in the study area and by the propagation of CTWs in the Bohai Sea and North Yellow Sea. At the beginning, southerly winds developed and diverged nearshore sea water. MSLD in Expc1 was stronger than that in Expc2 and currents in Expc3 were similar with those in Expc1, indicating that ocean currents in R1 were mainly driven by local winds. Then southerly winds were strong. MSLD in Expc2 was stronger than those in Expc1. During this period, currents driven by sea level slope of CTWs reversed surface wind-driven eastward currents but strengthened bottom southeastward currents. Horizontal currents rotated counterclockwise with increasing depth. Even though southerly winds relaxed, both MSLD and currents driven by CTWs slightly decreased. As sea surface winds reversed to northerly, southwestward currents driven by CTWs strengthened the surface wind-driven southwestward currents, weakened the bottom northward currents, and caused the bottom westward currents. Meanwhile, current directions rotated counterclockwise with increasing depth in the surface layer. Then strong northerly winds caused nearshore sea level to rise. Southward sea level slope of CTWs turned westward currents to eastward in the whole water column and strengthened bottom northward currents. At that time, current directions rotated counterclockwise with increasing depth.