Abstract

A regional atmosphere–ocean coupled model is developed, based on the Pennsylvania State University–National Center for Atmospheric Research Mesoscale Model in conjunction with the Princeton Ocean Model, to investigate atmosphere–ocean coupled processes that might occur over the Yellow and East China Sea shelves in winter. To examine how the coupled processes actually work in the ocean, sea surface temperatures (SSTs) computed in both coupled and uncoupled models are compared with SSTs synthesized from multiple satellite observations. The results indicate that the coupled model significantly improves the negative SST bias in shallow waters around the Chinese coast produced by the uncoupled model. Cool and dry northerly winds from the Asian landmass reduce SST in these shallow waters through intensive upward heat loss. Thereafter, the horizontal gradient of sea level pressure (SLP) around the Chinese coast moderates because the land–ocean heat contrast weakens owing to the reduced SST in the coastal waters. As a result, the wind speed weakens, in line with the moderated horizontal SLP gradient. Moreover, northerly winds can reduce the transport of cool and dry air from the Asian landmass. Hence, upward heat flux around the coastal waters is reduced because of the weakening of the northerly winds and the decreased cool and dry air. This negative feedback thereby suppresses excessive SST cooling along the Chinese coast during winter.

1. Introduction

It is frequently found that surface winds are positively correlated with both sea surface temperature (SST) and its gradient, partly because of vertical momentum mixing within the atmospheric boundary layer (Wallace et al. 1989) and partly because of the adjustment of sea level pressure (SLP) above sharp oceanic fronts (Lindzen and Nigam 1987). Recent advances in satellite observations have revealed how ocean winds are modified above oceanic fronts on horizontal scales much smaller than typical atmospheric processes (e.g., Chelton et al. 2001; Nonaka and Xie 2003; Vecchi et al. 2004). For instance, Chelton et al. (2001) developed a quantitative procedure to examine the vertical mixing mechanism of winds modified by oceanic tropical instability waves and showed that the perturbed wind stress curl (divergence) is linearly proportional to the crosswind (downwind) gradient of perturbed SST. Based on analyses of satellite microwave measurements over the North Pacific, Nonaka and Xie (2003) found positive correlation between SST and wind speeds over the intense Kuroshio and Kuroshio Extension. The correlation usually observed between these two properties is negative, especially above weak currents south of the Aleutian low. Furthermore, Vecchi et al. (2004) showed that low (high) wind velocities are observed over cold (warm) SST areas in the western Arabian Sea. As described above, a positive SST–wind correlation, which suggests active influence of the midlatitude ocean on the atmosphere, has been detected by satellite in various regions of the world.

In the Yellow and East China Seas (Fig. 1, hereafter YECS) during winter, oceanic influences on atmospheric processes are likely because, even in midwinter, relatively warm water originating from the Kuroshio spreads over the sea surface and because vertical momentum/heat transfer is enhanced by cool, dry northerly monsoon winds prevailing over these warm seas at this time. For instance, Xie et al. (2002) showed positive correlation between surface winds and SST over the winter YECS, using wind speeds observed by the Quick Scatterometer (QuikSCAT)/Seawinds and SST observed by the Tropical Rain Measuring Mission (TRMM) Microwave Imager (TMI). In addition, they investigated the rapid growth of an extratropical cyclone above the Kuroshio SST front along the shelf break in the East China Sea using a regional atmospheric numerical model. They demonstrated that a sharp SST gradient is favorable for enhancing low-level baroclinicity, which increases the growth rate of extratropical cyclones. A close relationship between SST and atmospheric fields in the YECS is not confined to these areas but also affects the activity of extratropical cyclones above neighboring seas. For example, using a satellite-derived dataset and a regional atmospheric numerical model, Isobe and Beardsley (2007) and Isobe and Kako (2012) showed that the YECS remotely affected development of extratropical cyclones over both the Japan Sea and areas south of Japan. Hence, it is likely that atmospheric processes are relatively sensitive to oceanic processes over the East Asian marginal seas where sharp oceanic fronts and submesoscale eddies prevail.

Fig. 1.

Model domain and bottom topography: (a) boxes depicted by solid lines indicate MM5V3 and POM domains. Letters A to E placed near the dotted line segments indicate locations of inflow, outflow, and sidewall boundaries; see text for details. (b) Area within small box over Korean Peninsula is enlarged to show positions of the KMA buoys (closed dots). Isobaths (m) are shown by solid curves in both panels.

Fig. 1.

Model domain and bottom topography: (a) boxes depicted by solid lines indicate MM5V3 and POM domains. Letters A to E placed near the dotted line segments indicate locations of inflow, outflow, and sidewall boundaries; see text for details. (b) Area within small box over Korean Peninsula is enlarged to show positions of the KMA buoys (closed dots). Isobaths (m) are shown by solid curves in both panels.

In both conventional oceanography and meteorology, it has been assumed that SST is passively influenced by the atmosphere via surface heat flux. In addition, the surface momentum flux, which drives ocean currents, also alters SST fields by the horizontal advection of heat. In fact, these atmospheric forcings have a significant impact on oceanic conditions over the YECS shelves because most of the area is shallower than 100 m (Fig. 1). Xie et al. (2002) demonstrated that isotherms appear to parallel isobaths over YECS shelves in winter since shallow areas become much cooler because of smaller heat content relative to deeper areas. In addition, Hsueh (1988) and Isobe (2008) showed that strong northerly (southward) wind events over the Yellow Sea are followed by northward counterwind ocean currents 1–2 days later. Thereafter, this northward ocean current alters the SST field in the Yellow Sea, as shown by Lie et al. (2001).

As mentioned above, the relatively warm YECS beneath cool and dry air might have an active role in altering the wind field during winter, whereas the SST field must be altered passively by strong northerly winter monsoon winds prevailing over the seas. These two processes are potentially interactive (two way), and it is on this relationship that we shall focus. Is atmosphere–ocean interaction possible in shelf seas? Different from the tropics where well-known interaction such as the Bjerknes feedback exists on the ENSO time scale, the difficulty in activating two-way processes in midlatitude shelf seas stems from the considerably different spatiotemporal scales between oceanic and atmospheric processes. The most energetic signal in the SST field in shallow shelf seas is seasonal (regional), whereas subweekly (synoptic) and monsoonal (planetary) signals are more energetic in the wind field.

The most reliable method to discern two-way oceanic and atmospheric processes on various time scales is probably a regional atmosphere–ocean coupled model, corroborated by observed datasets. In the present study, both coupled and uncoupled models are constructed to compare their results with synthesized satellite-derived SSTs and to evaluate how effective the coupled model is in reproducing observed SST variability in the YECS. The present conclusion is of negative feedback between monsoon winds and SST in coastal waters during winter.

2. Model setup and observed data

a. The regional atmospheric circulation model

The Pennsylvania State University–National Center for Atmospheric Research (NCAR) Mesoscale Model (MM5V3) (Grell et al. 1994) is adopted for the atmospheric circulation model, which constitutes the atmospheric part of the coupled model. Figure 1a shows the model domain used in both atmospheric and oceanic circulation models. The MM5V3 has a horizontal resolution of 8 km and 23 sigma levels in the vertical. The National Centers for Environmental Prediction Final Operational Global Data (http://rda.ucar.edu/datasets/ds083.2/) are used for the initial and lateral boundary conditions (see section 2c for computational period). Multiscale Ultrahigh Resolution Sea Surface Temperature (MURSST) (Table 1) data from the Jet Propulsion Laboratory website (http://mur.jpl.nasa.gov/index.php) are used for surface boundary conditions in regions outside the oceanic circulation model domain over which SST is computed (see section 2c for details). Land surface temperature is predicted using a five-layer soil model based on the vertical diffusion equation [ISOIL = 1 in the MM5 user guide, Dudhia et al. (2005)]. In addition, the Grell cumulus parameterization (ICUPA = 3), Medium-Range Forecast (MRF) planetary boundary layer (IBLTYP = 5), cloud and rainwater (IMPHYS = 4), and cloud radiation (IFRAD = 2) schemes are also included in the model. Furthermore, the radiative condition is chosen for the upper boundary (IFUPR = 1) to absorb momentum propagating upward. The above choices are optimal for modeling atmospheric processes influenced by oceanic fronts and eddies over the East Asian marginal seas. They were used in Yamamoto and Hirose (2007), who successfully reproduced development of a cyclone in the Japan Sea.

Table 1.

Specifications of three satellite-synthesized products. Satellites are listed as the AVHRR (1), MODIS (2), MODIS-Aqua (3), Advanced Microwave Scanning Radiometer–Earth Observing System (4), WindSat microwave radiometer (5), and TMI (6).

Specifications of three satellite-synthesized products. Satellites are listed as the AVHRR (1), MODIS (2), MODIS-Aqua (3), Advanced Microwave Scanning Radiometer–Earth Observing System (4), WindSat microwave radiometer (5), and TMI (6).
Specifications of three satellite-synthesized products. Satellites are listed as the AVHRR (1), MODIS (2), MODIS-Aqua (3), Advanced Microwave Scanning Radiometer–Earth Observing System (4), WindSat microwave radiometer (5), and TMI (6).

b. The regional ocean circulation model

The Princeton Ocean Model (POM) (Mellor 2003) is adopted for the ocean circulation model (Fig. 1a), which constitutes the oceanic part of the coupled model constructed. The POM uses 1/12° grid boxes in both latitude and longitude and 16 sigma levels vertically. The model has four open boundaries, at which the Kuroshio inflow (section A in Fig. 1a) and outflow (section B), the Taiwan warm current (C), and Tsushima Current (D) are given. Daily cross sections of temperature, salinity, and current velocities computed using the Data Assimilation Research of the East Asian Marine System (DREAMS) (Hirose 2011) are prescribed at each open boundary and are linearly interpolated at each time step. However, the southeast model boundary (E) is regarded as an artificial sidewall across which ocean currents are prohibited because water exchange across this boundary is unlikely to affect the shelf circulation (Chang and Isobe 2003). The POM initial condition is also set by the DREAMS at the beginning of calculation. In addition, sea surface elevation and tidal currents computed using the four major harmonic constants (M2, S2, O1, and K1) (Matsumoto et al. 2000) are prescribed at each boundary (A–E in Fig. 1a). The accuracy of the modeled tidal currents was validated by Isobe et al. (2007). They compared tidal currents calculated using the same ocean circulation model with those observed using an onboard acoustic Doppler current profiler in the East China Sea shelf. The boundary condition at the sea surface is provided by the MM5V3 in a coupling procedure, a detailed description of which is given in the next subsection.

Bottom topography data (ETOPO5) with 5-min resolution are provided by the National Geophysical Data Center and smoothed using a combination of Laplacian and median filters to avoid pressure gradient errors (Haney 1991). Horizontal viscosity and diffusivity are parameterized with the Smagorinsky diffusion formula in the original POM code (Mellor 2003), and vertical viscosity and diffusivity are obtained through the second-order turbulence closure scheme of Mellor and Yamada (1982).

c. Coupling procedures

A flowchart (Fig. 2) summarizes the coupling procedure between the atmospheric and oceanic models. Initially, SST derived from the DREAMS is input to the MM5V3 as the lower boundary condition (1 in Fig. 2). Second, the computation using MM5V3 proceeds for 6 h using SST fixed in time (2 in Fig. 2). Third is the 6-h computation by POM (4 in Fig. 2), using MM5V3-derived surface heat and freshwater/momentum fluxes. These are linearly interpolated at each POM time step, using two sequential MM5V3 results updated once every 6 h (3 in Fig. 2). Thereafter, the next round of MM5V3 starts (6 in Fig. 2), using SST updated from the previous POM (5 in Fig. 2). As mentioned above, atmospheric conditions input to the POM are updated each time step using linear interpolation, whereas the SST input to MM5V3 is updated once every 6 h. Fixing the SST during the 6-h period for each round of MM5V3 computations is justified because, in general, oceanic processes proceed much more slowly than atmospheric ones. In our procedure, SST changes over periods shorter than 6 h were negligibly small (less than 0.1°C on average, shown later in Fig. 5) compared with atmospheric processes, which varied significantly over this period.

Fig. 2.

Flowchart showing coupling procedure between atmospheric and oceanic models. Properties computed by these models are exchanged, in the numerical order within boxes, as described in the text.

Fig. 2.

Flowchart showing coupling procedure between atmospheric and oceanic models. Properties computed by these models are exchanged, in the numerical order within boxes, as described in the text.

Heat, freshwater, and momentum fluxes are computed in line with the procedure below. Downward shortwave and longwave radiation, latent heat flux, and sensible heat flux are computed in MM5V3 using the SST from the POM. Upward shortwave radiation is computed using the prescribed albedo (0.06) at the ocean surface and downward shortwave radiation. Upward longwave radiation is estimated from the fourth power of the POM-derived SST, scaled by the Stefan–Boltzmann constant and emissivity at the ocean surface. Freshwater flux through the sea surface is input to the POM as the difference between precipitation derived from MM5V3 and evaporation computed using latent heat flux and POM-derived SST. Freshwater flux from the Changjiang River mouth, which accounts for 90% of all river discharge into the model domain, is obtained from the same climatological data as those in Chang and Isobe (2003). Surface momentum flux input to the POM is derived from surface winds at 10 m in MM5V3, using a stability-independent drag coefficient following Large and Pond (1981). The above exchange of surface fluxes and SST is conducted at the nearest grid point of both models.

In total, coupling computations were conducted for three periods, from 1 November to 15 March in 2007/08, 2008/09, and 2009/10. These three winter seasons were chosen because three concurrent satellite-derived SST datasets (shown later in section 2e) are available for model validation. Although the computations are carried out for a period of five months, model results are dumped every 6 h from 1 December to the end of February for subsequent analyses to avoid initial disturbances possibly generated at the beginning of computation. The computation in March is required for preparing SST data used in the uncoupled experiment (see next subsection).

d. Uncoupled experiments

To investigate how the coupled model is capable of reproducing oceanic and atmospheric processes across the model domain, we conducted an “uncoupled” experiment in which the aforementioned SST exchange between MM5V3 and POM is disconnected. The monthly averaged values of SST computed using the POM in the coupled experiment for all three years are linearly interpolated at each time step during the MM5V3 computation and are input to the lower boundary of MM5V3 in this experiment. Here, the winter (December through February) average of the interpolated SST in each year has a bias relative to that of the SST in the coupled model because SST does not change linearly with time in that model. This bias might bias the uncoupled model results. Hence, the difference in the winter average between the interpolated SST and that of the coupled model was removed from the interpolated SST input to the uncoupled model. In this uncoupled experiment, MURSST is also given at the lower boundary of MM5V3 over the area outside the POM domain. The use of monthly averaged SSTs derived from the coupled model signifies that the difference between the coupled and uncoupled models is simply an absence of coupled processes triggered by SST fluctuations shorter than one month. Hence, coupled processes longer than the monthly time scale may be included in both coupled and uncoupled experiments. However, these long-term processes are beyond the present scope. In fact, use of the monthly averaged MURSST rather than POM-derived SST, even in the POM domain, was the alternative choice for decoupling. However, it would be difficult to distinguish whether the modeled processes are caused by the decoupling or the difference between MURSST and the SST from use of the coupled model. The computational period is shorter than in the coupled experiment because the monthly averaged SST from the coupled model results is unavailable at either end of the period 1 November through 15 March. Thus, in the uncoupled experiment, the computation from 16 November to the end of February is repeated thrice from 2007 to 2010.

All conditions, except the computational period and SST input to the MM5V3, are the same as those used in the coupled experiment. As in that experiment, the momentum, heat, and freshwater fluxes obtained from MM5V3, for which 6-h results are interpolated at each POM time step, are input to the POM as boundary conditions at the sea surface. Hence, the interpolated values of monthly mean SST in the coupled experiment are used to estimate surface heat flux. In addition, all lateral boundary conditions for the POM are identical to those used in the coupled experiment. It may be a concern that less variability of SST input to the atmospheric model directly leads to less variability of surface heat flux. We checked the difference between standard deviations of coupled and uncoupled surface net heat fluxes over winter (December–February) from 2007 to 2010 (not shown) and confirmed this. Variability in the uncoupled model is comparable to that in the coupled model across the model domain.

e. Satellite SST products and buoy data

To validate the modeled SST, we chose one of three satellite-derived SST products (Table 1) based on syntheses of multiple satellite observations, which have been widely used in the oceanographic community. The accuracy of these three products was compared using meteorological buoy data (shown below) to determine the most appropriate SST product for validation of modeled SST from 2007 to 2010 in each winter (hereafter, December–February, unless otherwise stated).

A brief description regarding the three satellite SST products follows. The first product is the New Generation Sea Surface Temperature (NGSST) (Guan and Kawamura 2004) provided by Tohoku University. Daily averaged SST maps within each 0.05° × 0.05° spatial grid were derived by a three-dimensional optimum interpolation (OI) method. This product is constructed by synthesizing infrared [Advanced Very High Resolution Radiometer (AVHRR), Moderate Resolution Imaging Spectroradiometer (MODIS)-Terra, and MODIS-Aqua] and microwave radiometers (AMSR-E). The second product is the microwave and infrared (MWIR) SST dataset (Gentemann et al. 2004) provided on a Remote Sensing Systems website (http://www.ssmi.com/sst/microwave_oi_sst_data_description.html). This product contains daily averaged SSTs on a 0.09° × 0.09° spatial grid and is obtained from a three-dimensional OI method in the same way as NGSST. WindSat, MODIS-Terra, MODIS-Aqua, AMSR-E, and TMI observations are synthesized to build this SST product. The third product is MURSST (Chin et al. 2010), which provides daily averaged SST maps on each 0.011° × 0.011° spatial grid. This product is constructed using multiresolution analysis, motion-compensated analysis, and composite imagery, the techniques of which were described by Chin et al. (2010). This product uses satellite SST data from AVHRR, MODIS-Terra, MODIS-Aqua, and AMSR-E.

The Korean Meteorological Administration (KMA) has observed SST at six buoy stations along the Korean coast (Fig. 1b). SST is measured at 1.2-m depth at a buoy southwest of Cheju Island and at 20-cm depth at five other buoys. The aforementioned satellite-derived SST products were compared with these in situ SST time series, which are not included in the three satellite SST products. To facilitate this comparison, hourly buoy data were averaged over each day during the three winters from 2007 to 2010.

3. Results

a. Validation of satellite SST products

The daily temperature data at KMA buoys were compared with those of three satellite SST products (Table 2). In this comparison, we used satellite data at grid points closest to the location of each buoy. The biases of all satellite-derived SST products, especially of the NGSST data, show overestimation of in situ SST observed at the KMA buoys. However, the MURSST bias is smallest among those of the other products. Moreover, the rms error (RMSE) and correlation coefficient (Corr) for MURSST are 1.66°C and 0.91, respectively, which are better than those of the NGSST and MWIR products.

Table 2.

Statistical values between the satellite SST products and KMA buoys. Data number is 1005.

Statistical values between the satellite SST products and KMA buoys. Data number is 1005.
Statistical values between the satellite SST products and KMA buoys. Data number is 1005.

Probably, the higher accuracy of MURSST is a result of its greater spatial resolution relative to that of the NGSST and MWIR products. In gridding satellite-derived SSTs with relatively coarse resolution, SST data observed in deep oceans are likely to “contaminate” gridded SSTs over shallow coastal waters where winter SSTs are considerably cooler than in the deep ocean because of the difference in heat content. In addition to the areas near the Korean Peninsula, SSTs of NGSST and MWIR are warmer than those of MURSST in the shallow waters along the Chinese and Japanese coasts (not shown).

The aforementioned accuracy of MURSST lends confidence to its use in validating the modeled SSTs, especially in coastal waters. However, MURSST also has a positive bias (1.27°C), which might cause disagreement between observed and modeled SSTs. Hence, we corrected the MURSST data (T) to be consistent with KMA buoy data using a linear regression formula (T* = −1.964 + 1.052T) because we focus on SSTs in shallow coastal waters (<50 m, see buoy location in Fig. 1).

b. Evaluation of modeled SST

Figure 3 shows SST maps averaged over the winters from 2007 to 2010 using MURSST (Fig. 3a), coupled (Fig. 3b), and uncoupled (Fig. 3c) models. The horizontal SST gradient is superimposed on these maps. Overall, the SST maps simulated by both coupled and uncoupled models are consistent with the satellite-derived SST map. The most remarkable feature is the Kuroshio front along the shelf break at, roughly, the 200-m isobath in the East China Sea. This relatively weak front is also in areas between the 50-m and 100-m isobaths from the Changjiang River mouth to Cheju Island. In addition to these two fronts, a sharp SST gradient appears along the entire coast of the YECS because of the intense surface cooling, which makes the shallow coastal areas much cooler than the offshore waters owing to relatively smaller heat content.

Fig. 3.

Horizontal views of SST (solid curves) and its horizontal gradient (color shading), averaged over the winters (December–February) from 2007 to 2010, for (a) MURSST, (b) coupled, and (c) uncoupled models. Contour interval of SST is 1°C. Values of color shading are indicated by scale (×10−2 °C km−1) below each panel. Also shown by thin broken curves are 50-m, 100-m, and 200-m isobaths.

Fig. 3.

Horizontal views of SST (solid curves) and its horizontal gradient (color shading), averaged over the winters (December–February) from 2007 to 2010, for (a) MURSST, (b) coupled, and (c) uncoupled models. Contour interval of SST is 1°C. Values of color shading are indicated by scale (×10−2 °C km−1) below each panel. Also shown by thin broken curves are 50-m, 100-m, and 200-m isobaths.

RMSEs (Figs. 4a,b) and correlation coefficients (Figs. 4d,e) were calculated between daily MURSST and modeled SST over the 2007–10 winters. Overall, the RMSEs (correlation coefficients) between MUR and modeled SSTs are large (small) around the areas of the sharp SST gradient mentioned above, regardless whether the model is coupled or uncoupled. Large and complex SST fluctuations are likely around frontal regions, where SST changes abruptly in space. This is reasonable because both coupled and uncoupled models cannot fully reproduce these fluctuations.

Fig. 4.

(top left) RMSE and (bottom left) correlation coefficients computed between daily MUR and modeled SSTs over winters (December–February) from 2007 to 2010. (a),(d) The coupled and (b),(e) uncoupled model and (c),(f) differences between coupled and uncoupled models, respectively. Dense stippling in correlation coefficient maps [(d) and (e)] is used for lower values to designate areas with inaccurate modeling. In (f), the difference in correlation coefficients is plotted only when it exceeds significant values suggested by 95% confidence level. Box in (f) indicates area over which the SST is averaged to compare between the two models and MURSST in Fig. 5.

Fig. 4.

(top left) RMSE and (bottom left) correlation coefficients computed between daily MUR and modeled SSTs over winters (December–February) from 2007 to 2010. (a),(d) The coupled and (b),(e) uncoupled model and (c),(f) differences between coupled and uncoupled models, respectively. Dense stippling in correlation coefficient maps [(d) and (e)] is used for lower values to designate areas with inaccurate modeling. In (f), the difference in correlation coefficients is plotted only when it exceeds significant values suggested by 95% confidence level. Box in (f) indicates area over which the SST is averaged to compare between the two models and MURSST in Fig. 5.

Of particular interest is that differences of both RMSE and correlation coefficient are large between the coupled and uncoupled models, especially in coastal waters (Figs. 4c,f). Overall, the RMSEs (correlation coefficients) decrease (increase) significantly in the coupled model, suggesting that atmosphere–ocean coupled processes certainly improve model accuracy in coastal waters. Furthermore, areas of enhanced accuracy are more remarkable along the Chinese coast than along the coasts of the Japanese islands, Korean Peninsula, and Taiwan Island. For example, Fig. 5 shows temporal variation of daily SST averaged over the region (32°–34°N, 121°–122°E; box in Fig. 4f) near the Chinese coast during the winters from 2009 to 2010. The SST computed by the coupled model is consistent with MURSST, whereas that in the uncoupled model has a large negative bias. A sea ice model is not included in the present configuration. If we incorporated such a model into the ocean model, the negative bias could not exceed the freezing point. The area where SST in the uncoupled model has a large negative bias is not only within the aforementioned small box but also in coastal waters where the SST accuracy was considerably improved in the coupled model (Figs. 4c,f) (temporal variations not shown). These results suggest that an atmosphere–ocean coupled process is active in the YECS coastal waters, which effectively suppresses the excess cooling produced by the uncoupled model. It is this suppressive mechanism that the present study tends to uncover.

Fig. 5.

Temporal variations of daily SST across the region (32°–34°N, 121°–122°E; Fig. 4f) during winter 2009/10. As shown in the upper right-hand corner, different line types are used for temperatures derived from MURSST, coupled model, and uncoupled model. The abscissa indicates days elapsed from 1 December.

Fig. 5.

Temporal variations of daily SST across the region (32°–34°N, 121°–122°E; Fig. 4f) during winter 2009/10. As shown in the upper right-hand corner, different line types are used for temperatures derived from MURSST, coupled model, and uncoupled model. The abscissa indicates days elapsed from 1 December.

4. Feedback process

a. A negative feedback process in coastal waters

SST in the YECS is greatly affected by atmosphere–ocean heat exchange (e.g., Hirose et al. 1999). In particular, latent and sensible heat losses are enhanced over the winter YECS because the northerly monsoonal wind advects cold and dry air from the Asian landmass. Therefore, it is reasonable to start by comparing surface winds reproduced in the coupled and uncoupled models to discover why excessive cooling is suppressed in the coupled model. Figure 6a shows the average field of surface (10-m height) winds from the coupled model during the 2007 to 2010 winters, their difference between the coupled and uncoupled models (Fig. 6b), the surface air temperature difference between those models (Fig. 6c), and the surface turbulent heat flux difference between those models (Fig. 6d). Parallel to the Asian coastline, the winter monsoon prevails (Fig. 6a). An atmosphere–ocean coupled process diminishes the wind speed by several percent in the direction from the ocean to the landmass (Fig. 6b). Here, it would be desirable to compare observed and modeled winds. Unfortunately, over the coastal waters where there are significant differences of winds between the coupled and uncoupled models, we have no wind data. This is because microwave satellite data are usually contaminated by land in these areas. Areas with the most weakened winds are restricted to near the Chinese coast; this therefore suggests that the coupled process occurs in this restricted area. An increase of surface air temperature is shown by the coupled model along the Chinese coast (Fig. 6c). The surface air specific humidity also increases there (not shown). This is probably attributable to a reduction of the cold and dry air from the Asian landmass because of weakening of the northerly winds. The areas of weakened surface winds and increased surface air temperature are consistent with those in which SST accuracy is considerably improved by the coupled model (Figs. 4c,f). This means that the wind speed reduction and increases of surface air temperature and specific humidity in the coupled model are closely related to the suppression of excess cooling along the Chinese coast. In fact, surface heat loss is reduced by about 8 W m−2 along that coast using the coupled model (Fig. 6d). However, in the offshore area (24°–28°N, 121°–125°E), the surface heat flux difference is inconsistent with that of SST (Figs. 4c,f). This is likely because the surface heat flux difference is insufficient to cause the SST difference, as the ocean in that area is deeper than that of the coastal area. Unlike the areas along the Chinese coast, decreasing wind speeds in the coupled model are not remarkable along the coasts of the Japanese islands, Korean Peninsula, and Taiwan Island.

Fig. 6.

(a) Averaged field of surface wind computed by coupled model over winters (December–February) from 2007 to 2010; (b) its difference between coupled and uncoupled model; (c) surface air temperature difference between coupled and uncoupled model; and (d) surface heat flux difference between coupled and uncoupled model. Values from coupled model minus those from uncoupled model are depicted in (b),(c), and (d). Negative values indicated that upward (ocean to air) heat flux from coupled model was smaller than that in uncoupled model. Modeled results only over POM domain are depicted. Vectors indicate wind, the scales of which are shown in upper right-hand corner of each panel. Contours with stippling indicate wind speed or heat flux, the scale of which is shown at the bottom of each panel. Contour intervals are (a) 0.5 m s−1, (b) 0.05 m s−1, (c) 0.025°C, and (d) 2 W m−2.

Fig. 6.

(a) Averaged field of surface wind computed by coupled model over winters (December–February) from 2007 to 2010; (b) its difference between coupled and uncoupled model; (c) surface air temperature difference between coupled and uncoupled model; and (d) surface heat flux difference between coupled and uncoupled model. Values from coupled model minus those from uncoupled model are depicted in (b),(c), and (d). Negative values indicated that upward (ocean to air) heat flux from coupled model was smaller than that in uncoupled model. Modeled results only over POM domain are depicted. Vectors indicate wind, the scales of which are shown in upper right-hand corner of each panel. Contours with stippling indicate wind speed or heat flux, the scale of which is shown at the bottom of each panel. Contour intervals are (a) 0.5 m s−1, (b) 0.05 m s−1, (c) 0.025°C, and (d) 2 W m−2.

In general, surface wind speeds are associated with the horizontal gradient of SLP, so we compared those gradients in the coupled and uncoupled models over the 2007 to 2010 winters (Fig. 7). Apart from areas with high mountains (>200 m, thick lines in Fig. 7) that can complicate the SLP gradient, the gradient in the coupled model is more moderate than that produced by the uncoupled model within a narrow band along the Chinese coast. This area of moderate gradient is consistent with the area of weakened wind speed (Fig. 6b). Therefore, the adjustment between winds and SLP is altered within the aforesaid narrow band. Similar to the surface wind field (Fig. 6b), weakening of the horizontal SLP gradient in the coupled model is not remarkable over the coasts of the Japanese islands, Korean Peninsula, and Taiwan Island. This is probably because, in general, that gradient (and hence, wind speed) across a coastline is strongly influenced by a difference in heat content between the landmass and ocean and because small landmasses such as peninsulas and islands surrounded by warm oceans cannot generate a sharp difference in heat content.

Fig. 7.

Difference (×10−2 hPa km−1) between horizontal SLP gradients in coupled and uncoupled models. Values are averaged over winters from 2007 to 2010; differences over the POM domain only are depicted. To remove “noisy” data, spatial smoothing using a box filter of horizontal scale 30 km was performed once. Values for terrain higher than 200 m (surrounded by bold curve) are omitted because of overcrowding.

Fig. 7.

Difference (×10−2 hPa km−1) between horizontal SLP gradients in coupled and uncoupled models. Values are averaged over winters from 2007 to 2010; differences over the POM domain only are depicted. To remove “noisy” data, spatial smoothing using a box filter of horizontal scale 30 km was performed once. Values for terrain higher than 200 m (surrounded by bold curve) are omitted because of overcrowding.

Why the horizontal SLP gradient (and hence wind speeds) in the coupled model should be lower than those in the uncoupled model along the Chinese coast needs to be addressed. In general, SLP in winter above warm oceans is lower than that above cool landmasses at midlatitudes because of air temperature differences between oceans and landmasses. Hence, the difference in the horizontal SLP gradient is likely caused by a difference in SST variation between the two models. To examine the relationships among SLP, SST, and wind speeds their differences between the coupled and uncoupled models are calculated across the ocean (32°–34°N, 121°–122°E; see box in Fig. 7) near the Chinese coast during winter 2008/09 (Fig. 8a). The SSTs used in the uncoupled model are monthly averages interpolated at each time step. Temporal variations of SLP, SST, and wind speed differences are dominant on a subweekly time scale. This is probably associated with the passage of extratropical cyclones. A positive difference in SLP variation, except for the last two weeks, is consistent with moderate SLP gradients along the Chinese coast (Fig. 7) caused by relatively high SLP over the ocean. It is apparent that the temporal variation of SST difference is negatively (positively) correlated with that of the SLP (wind speed) difference. However, the SST difference varies in time approximately in phase with the other two differences. Specifically, the variation of SST difference is followed by both SLP and wind speed differences with a delay of 6 h (Fig. 8b). The same relationship of SST, SLP, and wind speeds is evident along the entire Chinese coast from the model (not shown). The above suggests a coupled process of negative feedback such that reduced SST from intense northerly winds over the shallow Chinese coastal waters (Fig. 6a) generates high SLP there (negative correlation in Fig. 8). This in turn weakens the SLP gradient and wind speed there (positive correlation in Fig. 8). The cold and dry air is moderated by the weakened northerly winds over the coastal waters (Fig. 6c). Hence, upward heat flux is suppressed by those weakened winds, along with increased air temperature and air specific humidity (Fig. 6d). The excess cooling in those waters is thereafter suppressed in the coupled model (Fig. 5).

Fig. 8.

(a) Relationship among SST, SLP, and wind speed averaged over the region (32°–34°N, 121°–122°E; Fig. 7) during winter 2008/09. (a) Indicates temporal variation of the difference in SST, SLP, and wind speed between the coupled and uncoupled models. The meaning of various line types is indicated. The abscissa indicates month/day/year (last two digits). (b) Lag correlations between these time series. Solid (broken) curve shows lag correlation between SST and SLP (SST and wind speed) time series in (a). Positive lags (6 h in the present case) mean that SST varies in time in advance of SLP and wind speed.

Fig. 8.

(a) Relationship among SST, SLP, and wind speed averaged over the region (32°–34°N, 121°–122°E; Fig. 7) during winter 2008/09. (a) Indicates temporal variation of the difference in SST, SLP, and wind speed between the coupled and uncoupled models. The meaning of various line types is indicated. The abscissa indicates month/day/year (last two digits). (b) Lag correlations between these time series. Solid (broken) curve shows lag correlation between SST and SLP (SST and wind speed) time series in (a). Positive lags (6 h in the present case) mean that SST varies in time in advance of SLP and wind speed.

b. Conditions favoring negative feedback

The negative feedback process is triggered by northerly winds, as shown in Fig. 6a. Hence, it is reasonable to believe that this feedback is inactive unless northerly winds prevail. In fact, this feedback does not always occur in the YECS, as shown by the last two weeks in Fig. 8a. The SST during this period was higher than the monthly averaged value used in the uncoupled experiment; thus, SLP is reduced over the ocean. Nevertheless, the difference in wind speed between the coupled and uncoupled models is very small compared with prior periods, suggesting that coupled processes between the atmosphere and ocean are inactive. Figure 9 shows composite maps of SLP and surface winds from the coupled model in positive and negative phases of the SST difference shown in Fig. 8a but for the extended period of three winters during 2007–10. It was found that the SLP increases from the landmass to the ocean during the negative phase. Northwesterly winds prevail over the YECS (Fig. 6a) because of geostrophy and friction within the planetary boundary layer (Fig. 9a). However, the patterns of both SLP and wind are substantially different during the positive phase (Fig. 9b). Of particular interest is the onshore wind direction around the Chinese coast where the negative feedback occurs in the negative phase. SST in coastal waters is unlikely reduced by surface winds from the warm ocean, so the negative feedback cannot be realized during the positive phase. However, the difference in wind speeds between the coupled and uncoupled models in the negative phase is larger than that of wind speeds in the positive phase. Therefore, the negative feedback prevails for winds, SST, and SLP in the coastal waters on the Chinese side of the YECS during winter.

Fig. 9.

Composite maps of SLP (contours) and wind field (vectors) from coupled model in (a) negative and (b) positive phases of SST difference shown in Fig. 8a but for 2007–10 winters. Modeled values are plotted only within the POM domain. Contour interval is 1 hPa. Wind vectors are scaled as shown in the upper right-hand corner of each panel.

Fig. 9.

Composite maps of SLP (contours) and wind field (vectors) from coupled model in (a) negative and (b) positive phases of SST difference shown in Fig. 8a but for 2007–10 winters. Modeled values are plotted only within the POM domain. Contour interval is 1 hPa. Wind vectors are scaled as shown in the upper right-hand corner of each panel.

In summary, we emphasize only two prerequisites for the aforementioned negative feedback process to occur in midlatitude coastal waters during winter. One is that the warm and shallow shelf seas adjacent to a huge landmass, such as the YECS adjoining the Asian landmass, maintain a sharp gradient of heat content (and hence SLP). The other is that offshore winds transport cool and dry air over a warm ocean, thus reducing the SST in coastal waters.

5. Conclusions

The present study has shown that atmosphere–ocean coupled processes can operate over midlatitude shelf seas, as well as over the ocean in the tropics. Figure 10 shows a schematic of the negative feedback process in the shallow coastal waters of the YECS during winter. Northerly monsoonal winds prevail over the YECS in winter because, in general, the horizontal SLP gradient is sharp because of the large landmass–ocean heat contrast (Fig. 10a). Enhanced cool and dry winds from the Asian landmass decrease the SST, especially in shallow coastal waters, because of intense upward heat loss (Fig. 10b). Thereafter, the horizontal SLP gradient along the Chinese coast moderates as the land–ocean heat contrast weakens, owing to the reduced SST in coastal areas (Fig. 10b). As a result, wind speeds weaken in line with the moderated horizontal SLP gradient along the Chinese coast (Fig. 10b). Furthermore, surface air temperature and specific humidity become high because the cool and dry air is moderated by weakening the northerly winds over coastal waters; hence, upward heat flux around these waters is reduced. This negative feedback process can suppress excess cooling along the Chinese coast during winter. Temporal variation of this negative feedback process appears to be dominant on a subweekly time scale (Fig. 8a). In addition, it seems likely that this subweekly negative feedback contributes to diminishing the winter-average SST.

Fig. 10.

Schematic view of THE atmosphere–ocean coupled process over coastal waters of YECS in winter. The difference in SLP before and after the negative feedback process is expressed by the smaller size of the right-hand red circle around “L”; similarly, the change (decrease) in wind speed in the coupled process is expressed by the smaller vector size.

Fig. 10.

Schematic view of THE atmosphere–ocean coupled process over coastal waters of YECS in winter. The difference in SLP before and after the negative feedback process is expressed by the smaller size of the right-hand red circle around “L”; similarly, the change (decrease) in wind speed in the coupled process is expressed by the smaller vector size.

Atmospheric processes are not critical when considering the aforementioned negative feedback because winds and air temperature change by only several percent (Figs. 6a–c). Nevertheless, in oceanic processes, this atmosphere–ocean coupling is important. This is because the excess cooling in shallow coastal waters during winter (greater than 3°C in the present case, Fig. 5) will not be improved in ocean circulation models unless we adopt numerical models that incorporate two-way processes with the atmosphere. Therefore, atmosphere–ocean coupled models are required in shelf sea research, whenever the focus is on oceanic processes in coastal waters adjacent to a large landmass that are exposed to strong offshore winds during winter.

Acknowledgments

This work was partly supported by the MEXT through Grant-in-Aid for Scientific Research (22106002). Four reviewers and the editor provided valuable comments to improve this paper significantly. The authors thank Dr. Chang Pil-Hun for providing the buoy data from KMA.

REFERENCES

REFERENCES
Chang
,
P.-H.
, and
A.
Isobe
,
2003
:
A numerical study on the Changjiang diluted water in the East and Yellow China Seas
.
J. Geophys. Res.
,
108
, 3299,
doi:10.1029/2002JC001749
.
Chelton
,
D. B.
, and
Coauthors
,
2001
:
Observations of coupling between surface wind stress and sea surface temperature in the eastern tropical Pacific
.
J. Climate
,
14
,
1479
1498
.
Chin
,
T. M.
,
J.
Vazquez
,
E. M.
Armstrong
, and
A. J.
Mariano
,
2010
: Algorithm theoretical basis document: Multi-scale, motion-compensated analysis of sea surface temperature, version 1.1. NASA Measures Algorithm Theoretical Basis Doc., 17 pp.
Dudhia
,
J.
,
D.
Gill
,
K.
Manning
,
W.
Wang
, and
C.
Bruyere
, cited
2005
: PSU/NCAR mesoscale modeling system tutorial class notes and users’ guide (MM5 modeling system version 3). [Available online at http://www.mmm.ucar.edu/mm5/documents/tutorial-v3-notes.html.]
Gentemann
,
C. L.
,
G. A.
Wick
,
J.
Cummings
, and
E.
Bayler
,
2004
: Multi-sensor improved sea surface temperature (MISST) for GODAE. Preprints, 13th Conf. on Satellite Meteorology and Oceanography, Amer. Meteor. Soc., Norfolk, VA, 4.2. [Available online at https://ams.confex.com/ams/13SATMET/webprogram/Paper80505.html.]
Grell
,
G. A.
,
J.
Dudhia
, and
D. R.
Stauffer
,
1994
: A description of the fifth-generation Penn State/NCAR Mesoscale Model (MM5). NCAR Tech. Note NCAR/TN-398+STR, 117 pp.
Guan
,
L.
, and
H.
Kawamura
,
2004
:
Merging satellite infrared and microwave SSTs: Methodology and evaluation of the new SST
.
J. Oceanogr.
,
60
,
905
912
.
Haney
,
R. L.
,
1991
:
On the pressure gradient force over the steep topography in sigma coordinate ocean models
.
J. Phys. Oceanogr.
,
21
,
610
619
.
Hirose
,
N.
,
2011
:
Inverse estimation of empirical parameters used in a regional ocean circulation model
.
J. Oceanogr.
,
67
,
323
336
.
Hirose
,
N.
,
H.-C.
Lee
, and
J.-H.
Yoon
,
1999
:
Surface heat flux in the East China Sea and the Yellow Sea
.
J. Phys. Oceanogr.
,
29
,
401
417
.
Hsueh
,
Y.
,
1988
:
Recent current observations in the eastern Yellow Sea
.
J. Geophys. Res.
,
93
(
C6
),
6875
6884
.
Isobe
,
A.
,
2008
:
Recent advances in ocean-circulation research on the Yellow and East China Sea shelves
.
J. Oceanogr.
,
64
,
569
584
.
Isobe
,
A.
, and
R. C.
Beardsley
,
2007
:
Atmosphere and marginal-sea interaction leading to an interannual variation in cold-air outbreak activity over the Japan Sea
.
J. Climate
,
20
,
5707
5714
.
Isobe
,
A.
, and
S.
Kako
,
2012
:
A role of the Yellow and East China Seas in the development of extratropical cyclones in winter
.
J. Climate
,
25
,
8328
8340
.
Isobe
,
A.
,
T.
Kuramitsu
,
H.
Nozaki
, and
P.-H.
Chang
,
2007
:
Reliability of ADCP data detided with a numerical model on the East China Sea shelf
.
J. Oceanogr.
,
63
,
135
141
.
Large
,
W. G.
, and
S.
Pond
,
1981
:
Open ocean momentum flux measurement in moderate to strong winds
.
J. Phys. Oceanogr.
,
11
,
324
336
.
Lie
,
H.-J.
,
C.-H.
Cho
,
J.-H.
Lee
, and
S.
Lee
,
2001
:
Does the Yellow Sea warm current really exist as a persistent mean flow?
J. Geophys. Res.
,
106
(
C10
),
22 199
22 210
.
Lindzen
,
R. S.
, and
S.
Nigam
,
1987
:
On the role of sea surface temperature gradients in forcing low-level winds and convergence in the tropics
.
J. Atmos. Sci.
,
44
,
2418
2436
.
Matsumoto
,
K.
,
T.
Takanezawa
, and
M.
Ooe
,
2000
:
Ocean tide models developed by assimilating TOPEX/POSEIDON altimeter data into hydrodynamical model: A global model and a regional model around Japan
.
J. Oceanogr.
,
56
,
567
581
.
Mellor
,
G. L.
,
2003
: Users guide for a three-dimensional, primitive equation, numerical ocean model (June 2003 version). Program in Atmospheric and Oceanic Science, Princeton University, 53 pp.
Mellor
,
G. L.
, and
T.
Yamada
,
1982
:
Development of a turbulence closure model for geographical fluid problems
.
Rev. Geophys.
,
20
,
851
875
.
Nonaka
,
M.
, and
S.-P.
Xie
,
2003
:
Covariations of sea surface temperature and wind over the Kuroshio and its extension: Evidence for ocean-to-atmosphere feedback
.
J. Climate
,
16
,
1404
1413
.
Vecchi
,
G. A.
,
S.-P.
Xie
, and
A. S.
Fischer
,
2004
:
Ocean–atmosphere covariability in the western Arabian Sea
.
J. Climate
,
17
,
1213
1224
.
Wallace
,
J. M.
,
T. P.
Mitchell
, and
C.
Deser
,
1989
:
The influence of sea surface temperature on surface wind in the eastern equatorial Pacific: Seasonal and interannual variability
.
J. Climate
,
2
,
1492
1499
.
Xie
,
S.-P.
,
J.
Hafner
,
Y.
Tanimoto
,
W. T.
Liu
,
H.
Tokinaga
, and
H.
Xu
,
2002
:
Bathymetric effect on the winter sea surface temperature and climate of the Yellow and East China Seas
.
Geophys. Res. Lett.
,
29
, 2228,
doi:10.1029/2002GL015884
.
Yamamoto
,
M.
, and
N.
Hirose
,
2007
:
Impact of SST reanalyzed using OGCM on weather simulation: A case of a developing cyclone in the Japan Sea area
.
Geophys. Res. Lett.
,
34
, L05808,
doi:10.1029/2006GL028386
.