The variation of marine meteorology elements and the heat flux at Yongxing Island were investigated in September 2013 using observational data collected by the Yongxing Island Air–Sea Flux Tower. Heat flux was measured through eddy covariance and estimated using the bulk flux method. The traditional thirty-minute average method was employed to analyze the eddy covariance data. A TOGA–CORE 3.0 Algorithm was used to estimate the bulk flux. Capacity of the Advanced Weather Research and Forecasting model to simulate the variations in marine meteorological elements was evaluated using the observational data obtained during the same period. Observations suggested a response by the different parameters that were synchronous to the retreat of the monsoon around 7 September 2013. Surface air masses became cool and dry, and the downward short (long) wave radiation was reduced (increased). The Weather Research and Forecasting model simulated the variation of the atmospheric elements well, as observed by the Air-Sea Flux Tower in September 2013. The model was very accurate in simulating surface wind, radiation, and scalar–humidity. However, it failed to simulate the dramatic variations of surface air temperature, though it successfully reproduced the retreat of the summer monsoon, but weakened during the onset of the typhoon. The sensible flux simulated through the model was between the fluxes estimated through the bulk flux and eddy covariance flux. It also overestimated the latent heat flux at times, particularly after 20 September.

Introduction

The atmosphere and the ocean form a two–way coupled system that interacts by exchanging heat, momentum, and mass at the air–sea interface. Air–sea fluxes are affected by oceanic processes such as waves, eddies, fronts, upwellings and downwellings (Small et al., 2008; Chow and Liu, 2012; Frenger et al., 2013), as well as synoptic processes such as typhoons, cold surges and monsoons, among others. The air–sea interfacial fluxes could alter the structure of the upper ocean and atmospheric boundary layer that significantly influences global climate change and extreme regional weathers (Xie, 2004; Chelton and Xie, 2010; Shi et al., 2014). Therefore, air–sea interfacial fluxes are important parameters in studying the mechanism of air–sea interactions on different scales. The South China Sea (SCS) is an epicontinental marginal sea of the western Pacific Ocean and a Mediterranean–type basin. It is dominated by the Asian monsoon, which blows northeasterly from October to April and southwesterly from May to September; it connects to the Indian Ocean and to the western Pacific Ocean through several oceanic and atmospheric processes that have potential influences on global climate change and regional weather (Qu et al., 2002; Liu et al., 2004; Wang et al., 2006; Liu et al., 2006). The SCS is the source of water vapor advection and several tropical cyclones, which are responsible for the extreme weather events that cause floods, storm tides, and other disasters in southern China and Southeast Asia. Therefore, the continuous and accurate long-term monitoring of the air–sea flux in the SCS could help us to further understand the mechanism of air–sea interaction, update the parameterization plans applied in the atmosphere model, and improve our longer term weather forecast and severe weather warning system. For this purpose, the Yongxing Island Air–Sea Flux Tower (YASFT), which has been in operation since August 2013, was established at the Yongxing Island Marine Research Station (YIMRS) of the South China Sea Institute of Oceanology. As the first deep-sea observation station in China, the YIMRS is located in the deep-water basin of the north-central region of the SCS more than 100 nautical miles from the continent, which helps to minimize the influence of land on the air–sea fluxes.

In this article, we analyzed one month of observation (in September 2013) of the air–sea fluxes and other marine meteorological parameters. In addition, observation results were used to examine the capability of the atmospheric model to simulate the air–sea fluxes in the SCS. A brief introduction to the air–sea flux tower, the air–sea fluxes algorithm, and the model configuration are given. Observation results are analyzed and employed to validate the simulation results and we attempt to provide some explanations for the discrepancy between the results of the observation and simulation.

Methods and materials

Air–Sea Flux Tower

The 20 m high Yongxing Island Air–Sea Flux Tower (YASFT), which measures 1 meter deep, is located 97 meters off Yongxing Island at 16.84 °N, 112.33 °E (Figure 1b). Two sets of observation systems, gradient and eddy covariance, are installed on the tower to measure air–sea flux (Figure 1a). The gradient observation system is designed with a Visalia HMP155A temperature and relative humidity probe. There are Met-One wind speed and direction sensors installed at heights of 5, 10, 15 and 18 m. Two pairs of upward/downward Kipp and Zonen CMP22 shortwave radiation sensors and Kipp and Zonen CGR4 long wave radiation sensors were installed at 10 m. To measure the skin temperature of the sea surface, SI–111 infrared radiometer was installed at 10 m. The data acquisition instruments installed are Gill R3–50 ultrasonic wind anemometers, at 10 m, and a Li–Cor 7500A CO2 and H2O infrared gas analyzer was installed at the top of the tower (at a height of 20 m). Taking advantage of the high sampling frequency of 10 Hz, these two instruments make up the eddy covariance system that directly measures the turbulence perturbation of temperature, water vapor, wind velocity, and carbon dioxide (Figure 1a). This study used the results from these two instruments taken in September 2013.

Figure 1.

(a) Yongxing Island Air–Sea Flux Tower (YASFT) and the major instruments included. The YASFT has been operational since August 2013. (b) Map of the northern South China Sea with bathymetry contours (m). The star indicates the location of Yongxing Island. The area enclosed by the black box is the inner nest domain of the WRF model.

Figure 1.

(a) Yongxing Island Air–Sea Flux Tower (YASFT) and the major instruments included. The YASFT has been operational since August 2013. (b) Map of the northern South China Sea with bathymetry contours (m). The star indicates the location of Yongxing Island. The area enclosed by the black box is the inner nest domain of the WRF model.

Air–sea flux algorithm

Three methods are used to calculate the air–sea heat flux, namely the eddy covariance method, bulk algorithm, and scalar gradient method (Stull, 1988). The eddy covariance method is based on the eddy correlation law, which depends on the direct measurement of the turbulence perturbations of meteorological parameters. The bulk algorithm and the scalar gradient method are based on Monin–Obukhov similarity theory and Prandtl's mixing length hypothesis, respectively.

The eddy covariance and bulk algorithm were used in this article to calculate the air–sea heat flux. Using turbulence perturbations measured through the eddy covariance system, the sensible flux Hs, latent flux Hl can be estimated using the following equations: , , where T and q denote temperature and specific humidity, respectively; the overbar indicates the Reynolds– averaged values; and the primes denote the turbulent perturbations. The traditional thirty minute-average method was applied to analyze the eddy covariance data at a high frequency. ρa refers to atmosphere density; Cp is the specific heat of air, using 1004.67 J kg−1 K−1 in our calculation; and Le is the latent heat of evaporation (2.5 × 106 J kg−1). The sensible flux Hs, latent flux Hl was estimated using the bulk algorithm, which is given by , , where ρa, Le, Cp, T, and q are the same values as above; the subscript s and a denote the parameters measured at sea surface and near-surface atmosphere, respectively; and Ws is the wind speed near the surface. Ch and Ce are the bulk transfer coefficients for the temperature and water vapor, respectively. Many algorithms are employed to estimate bulk transfer coefficients. Since the 1990s, the most frequently used algorithm in the area of air–sea interaction was developed from the Tropical Ocean and Global Atmosphere Coupled Ocean–Atmosphere Response Experiment (TOGA–CORE; Fairall et al., 1996, 2003; Brunke et al., 2003). This algorithm was employed to analyze the observation data from YASFT. In the bulk algorithm, the standard level for estimating air–sea flux is 10 m. In this study, we use the meteorological parameters for both the 10 m of YASFT and WRF model.

Global 1 Sea Surface data

In this study, an extremely high resolution sea surface temperature (SST) was recorded and used to drive our model. Using a multi-scale two-dimensional variational (MS–2DVAR) blending algorithm on a global 0.009 degree grid, the Jet Propulsion Laboratory (JPL), our ocean group, produced a Group for High Resolution Sea Surface Temperature (GHRSST) Level 4 sea surface temperature analysis daily on an operational basis (Chao et al., 2009). This global, nearly 1 km SST (G1SST) analysis utilizes satellite data from both infra-red and microwave sensors, as well as in-situ observational data obtained from drifting and moored buoys.

Model configuration

The Weather Research and Forecasting (WRF) model, combined with the Advanced Research WRF (ARW) dynamic solver (Skamarock et al., 2005), was implemented to reproduce the variation of atmospheric data observed by YASFT in September 2013. In this study, the WRF model has two nested domains, namely, D1 and D2, with horizontal resolutions of 18 km and 6 km, respectively. The outer domain (D1) covers the NSCS, whereas the inner domain (D2) focuses on the surrounding areas of Yongxing Island (black box shown in Figure 1b). Two-way nesting was applied to account for the interactions between the runs of two domains. This study utilized the simulation results for D2. The two domains were both initialized with the operational analysis data of the National Centers for Environmental Prediction (NCEP). The lateral boundary conditions for D1 were updated with NCEP analysis data every 6 h. The daily G1SST data, which were updated every 24 h during the simulation period, were employed as the bottom boundary condition. The initial and boundary conditions were designed to gain realistic simulation results and to maintain consistency in analyzing the observation and simulation results. The simulation was achieved using 43 vertical sigma levels including 25 levels below 1000 m. A fine vertical resolution was specified in the boundary layer for accurate simulation of the vertical turbulent momentum exchange in said layer. An improved Mellor–Yamada–Nakanishi–Niino (MYNN) Level 3 scheme was employed for the simulation of the marine atmospheric boundary layer. In addition, the Noah land surface model (Chen and Dudhia, 2001) was adopted in the simulation. The microphysical parameterizations in-clude explicitly resolved water vapor, cloud, and precipitation processes. Moreover, a modified version of the Kain–Fritsch scheme was used to represent the subgrid–scale effects of convection and shallow clouds. The model was run for an integration period of one month, beginning at 0000 UTC on 1 September 2013 and ending at 0000 UTC on 1 October 2013.

Results

Surface sea temperature

The results of the observation show that in the beginning of September, the SST was above 29°C and gradually decreased to about 28°C after the September 10 (Figure 2a). Typhoon Wutip caused an abrupt cooling of the SST by 2°C by late September. The quick fluctuation of SST at the beginning of September could have been caused by other synoptic disturbances. Compared with the observation data, The G1SST data is higher than the observed SST most of the time. The root mean square difference between the observed SST and G1SST was 0.8°C. The correlation coefficient was about 0.61 (Table 1). The magnitude of the fluctuation of SST caused by the synoptic processes is smaller in G1SST than in the observed results due to the change of water depth at the Yongxing Island. There was a delay of cooling in G1SST data during the typhoon period (Figure 2a). These differences could have had an influence on the simulation of other thermal-related parameters, such as temperature and heat flux by the WRF model.

Figure 2.

(Left) Time series of the observed daily SST (°C, black solid circle) and JPL global 1–km SST (°C, gray triangle); (Right) the observed daily 10 meter air temperature.

Figure 2.

(Left) Time series of the observed daily SST (°C, black solid circle) and JPL global 1–km SST (°C, gray triangle); (Right) the observed daily 10 meter air temperature.

Table 1.

Comparison between model-simulated and observed sea surface pressure (SLP), downward short wave (Rsw) and long wave (Rlw) radiation; temperature (T10), wind speed (WS10) and specific humidity (Q10_scalar) at the 10-m height above the sea surface for September 2013. The difference of G1SST and observed SST is also shown in the table.

ParameterMean (Obs)Mean (WRF)Corr CoefRMSDTime resolution
SST (°C) 28.78 29.28 0.61 0.86 Daily 
T10 (°C) 28.36 28.21 0.28 0.91 Hourly 
SLP (hPa) 998.91 1006.00 0.91 6.78 Hourly 
Q10_scalar (g kg−119.53 19.86 0.49 0.74 Hourly 
WS10 m s−1 5.38 6.34 0.60 3.37 Hourly 
Rsw W m−2 168.17 256.48 0.85 204.10 Hourly 
Rlw W m−2 436.40 427.52 0.44 16.02 Hourly 
ParameterMean (Obs)Mean (WRF)Corr CoefRMSDTime resolution
SST (°C) 28.78 29.28 0.61 0.86 Daily 
T10 (°C) 28.36 28.21 0.28 0.91 Hourly 
SLP (hPa) 998.91 1006.00 0.91 6.78 Hourly 
Q10_scalar (g kg−119.53 19.86 0.49 0.74 Hourly 
WS10 m s−1 5.38 6.34 0.60 3.37 Hourly 
Rsw W m−2 168.17 256.48 0.85 204.10 Hourly 
Rlw W m−2 436.40 427.52 0.44 16.02 Hourly 

Surface air temperature

The surface air temperature was about 29°C, with a small diurnal variation in the beginning of September (Figure 3a). From the 7 to 12 September, the surface air temperature, along with dramatically diurnal variation, dropped by about one degree. The WRF model underestimated the air temperature at the beginning of September, but simulated the approximate temperature for the remaining days (Figure 2b). The model did not simulate the strong diurnal variation of the surface air temperature, as shown in the observation (Figure 3a). The drop in air temperature by the end of September caused by Typhoon Wutip was clearly represented by both the observation and simulation.

Figure 3.

Time series of the hourly observations (black line) at YASFT and the simulation of the WRF model (gray line) for (a) 10-m temperature (°C), (b) surface pressure (hPa) and (c) 10-m specific humidity (g kg−1). The specific humidity observed by the Li-Cor 7500A CO2 and the H2O infrared gas analyzer, derived from the observation using the Visalia HMP155A temperature and relative humidity probe, are shown in (c) and (d), respectively.

Figure 3.

Time series of the hourly observations (black line) at YASFT and the simulation of the WRF model (gray line) for (a) 10-m temperature (°C), (b) surface pressure (hPa) and (c) 10-m specific humidity (g kg−1). The specific humidity observed by the Li-Cor 7500A CO2 and the H2O infrared gas analyzer, derived from the observation using the Visalia HMP155A temperature and relative humidity probe, are shown in (c) and (d), respectively.

Surface pressure

In general, the sea surface pressure gradually declined throughout September (Figure 3b). Typhoon Wutip was marked by an extremely low pressure of 960 hPa (Figure 3b). The small fins on the time series of observed surface pressure indicate the diurnal variation of the surface pressure. Unlike that of the surface air temperature, the magnitude of diurnal variation of the surface pressure did not change abruptly. A high correlation between the observed and simulated values of surface pressure was observed for the entire month (Table 1). The WRF model overestimated the surface pressure, and did not reproduce the observed intensity of Typhoon Wutip. Likewise, the apparent diurnal variation of surface pressure was not reproduced well by the WRF model.

Specific humidity

Two observation time series of specific humidity are presented in Figures 3c and d. The first observation time series is achieved by an infrared gas analyzer with a 10 Hz sampling frequency (Figure 3c, hereafter referred to as infra-humidity), and the other is derived from the 5-min average measurements of the temperature and humidity probe (Figure 3d, hereafter referred to as scalar–humidity).

In September, the mean value of the infra-humidity and scalar–humidity were both about 20 g kg−1. Both infra-humidity and scalar–humidity apparently declined on the 7 September, and gradually increased until the onset of Typhoon Wutip. Another apparent drop in humidity was observed before the passage of Typhoon Wutip. During the observation period, infra-humidity showed more apparent and strong perturbations than scalar–humidity because of the high frequency measurement of the infrared gas analyzer. The WRF model underestimated the humidity observed using the infrared gas analyzer (Figure 3c). The high sampling frequency made the infrared gas analyzer very sensitive to the change in humidity. The correlation between the simulated and scalar humidity is considerably higher than that between the simulation and infra-humidity (Figures 3c and d). This result can be explained by the abrupt perturbations of infra-humidity that were either not observed by scalar–humidity or not simulated by the WRF model.

Surface wind

For most of September, the observed wind speed fluctuated at 5 m s−1. The simulated hourly 10 m wind speed is supported by the results obtained by YASFT in September 2013 (Figure 4a). However, the WRF model overestimated the wind speed by about 1 m s−1 (Figure 4a; Table 1). The vector mean of the simulated wind direction was highly consistent with the observed results as well. The WRF model successfully reproduced the retreat of the southern summer monsoon from the September 7 as indicated by the observation (Figure 4b). After the retreat of the southern summer monsoon the wind speed slowed down, while the wind direction became unstable. The sharp rise and fall of the wind speed and the reversal of wind direction at the end of September demonstrated the footprint of Typhoon Wutip at ASWTF.

Figure 4.

(a) Time series of the hourly 10 m wind speed (m s−1); the wind vectors from the observation and simulation results are shown in (b) and (c), respectively.

Figure 4.

(a) Time series of the hourly 10 m wind speed (m s−1); the wind vectors from the observation and simulation results are shown in (b) and (c), respectively.

Downward radiation

In September, the maximum downward short wave radiation was about 900 W m−2 (Figure 5a), which is consistent with previous observations (Yan et al., 2005). The abrupt decline in short wave radiation could have resulted from the synoptic disturbances in the clouds. The phase of the diurnal cycle of the downward short wave radiation simulated through the WRF model was the same as that of the observation. However, the WRF model overestimated the downward shortwave radiations on these cloudy days. The WRF model appropriately simulated the observed downward long wave radiation that fluctuated from 420 W m−2 to 450 W m−2. An increase in downward long wave radiation appeared to occur after the retreat of the summer monsoon around 4 September (Figure 5b).

Figure 5.

The same as in Figure 3, but for the downward short wave radiation (W m−2) (a) and downward long wave radiation (W m−2) (b).

Figure 5.

The same as in Figure 3, but for the downward short wave radiation (W m−2) (a) and downward long wave radiation (W m−2) (b).

Heat flux

Our YASFT measurements showed that near surface turbulent heat fluxes could be measured using two approaches. First, fluxes can be calculated using the algorithm TOGA—CORE with measurements of parameters at a height of 10 m (hereafter referred to as Bulk-flux). Another method that can be used is measuring the fluxes directly using the eddy covariance system (hereafter referred to as Eddy-flux).

For sensible heat flux, Bulk-flux is generally smaller compared to a directly observed Eddy-flux, with a difference between 10 W m−2 and 20 W m−2 (Figure 6a). Considering the small magnitude of sensible heat flux, we were more concerned about the performance of latent heat flux. As shown in Figure 6b, two types of latent heat flux values, Bulk-flux and Eddy-flux, were more consistent compared to the sensible heat flux. To evaluate the capability of the WRF parameterizations, the model is implemented to export heat fluxes (WRF-flux) during the same period. The sensible WRF-flux was found between the values of Bulk-flux and Eddy-flux. The WRF–flux overestimated the latent heat flux sometimes, particularly after the 20 September. In the future, the observation results could be used to improve the parameterization plans of the model.

Figure 6.

Time series of the sensible heat flux (W m−2) (a) and latent heat flux (W m−2) (b). Three types of heat fluxes were employed: that which is directly measured through the eddy–covariance system (black dashes), that which is calculated using TOGA–CORE 3.0 bulk algorithm (black line), and that which is simulated using the WRF model (gray line).

Figure 6.

Time series of the sensible heat flux (W m−2) (a) and latent heat flux (W m−2) (b). Three types of heat fluxes were employed: that which is directly measured through the eddy–covariance system (black dashes), that which is calculated using TOGA–CORE 3.0 bulk algorithm (black line), and that which is simulated using the WRF model (gray line).

Discussion and conclusions

The purposes of this study, analyzing the observation data of YASFT, are two-fold. First, this study aimed to conduct a case study on the change in the atmospheric environment at Yongxing Island during the retreat month of the East Asian summer monsoon in 2013. The second reason was to evaluate the ability of the WRF model to simulate the change in the marine environment during the same period. The observation made by YASFT provided the characteristics of the air–sea interaction at one fixed station. Moreover, the observation results could be used to improve the parameterizations of the model to obtain a more accurate simulation of air–sea interaction processes in the entire SCS.

The observed wind vectors suggest that the summer monsoon retreated around 7 September (Figure 4b). After the retreat of the summer monsoon, the wind speed slowed down and the wind direction became unstable (Figure 4b). Although significant fluctuations in temperature and humidity were observed (Figures 3a and c), the magnitude of the variations in wind speed and wind direction were small (Figure 4b). In the first few days after the retreat of the summer monsoon, the surface air masses became cool and dry (Figures 3a and c) because of the interrupted supply of warm and moist air from the southern ocean by the southern monsoon. Therefore, the rapid fluctuations of temperature and humidity shown in Figures 3a and c will not slow down until the onset of the stable winter monsoon. The downward short wave radiation declined when the summer monsoon began to retreat (Figure 5a), whereas the long wave radiation increased (Figure 5b). The sensible flux slightly increased for a short time after the retreat of the summer monsoon. Thus, the observation data suggests that the retreat of the monsoon resulted in synchronous responses by different parameters. In general, the WRF model simulated the variation of the atmospheric environment around Yongxing Island in September 2013 well.

The model accurately simulated surface wind, radiation, and scalar-humidity (Table 1), but failed to reproduce the rapid change of surface air temperature. The WRF model successfully reproduced the retreat of the southern summer monsoon but weakened during the onset of the typhoon. The sensible flux simulated by the WRF model was between the fluxes estimated by TOGA–CORE 3.0 bulk algorithm and those observed by the eddy–covariance instrument. At times, the WRF model overestimated the latent heat flux, particularly after the 20 September. The difference between the initial field of the model and in-situ observation could be responsible for the apparent differences between the model and the observation in the beginning of simulation (Figures 3a, c and 5b). The difference between G1SST and observed SST could have negatively influenced the simulation of the heat flux. In addition, updates of the SST were in 24 h intervals, which was more frequent than in the observation. This factor could have contributed to the errors in the estimation of the heat flux and surface air temperature. Better simulation results require more accurate initial field and efficient parameterization plans.

The differences in the heat fluxes obtained through the model and observed by YASFT indicate the need for further studies of the heat flux algorithm. The continuous YASFT measurements not only contribute to the validation and improvement of the heat flux algorithm, but also significantly benefit the study of monsoon and air–sea interactions in SCS. Similarly, future long-term series of meteorological observations would be very helpful in developing numerical models and validating results.

Acknowledgements

We would like to thank Professor Xinyu Guo of Ehime University for providing G1SST data. The numerical simulation is supported by the high-performance computing division of the South China Sea Institute of Oceanology.

Funding

This study was supported by the National Natural Science Foundation of China (Nos. 41106028, 41206011, 41476014 and 41306012), the National Natural Basic Research Program of China (973 Program, No. 2011CB403501) and the Infrastructure Construction Project of Xisha & Nansha Station (KZCX2-EW-Y040) and the CAS/SAFEA International Partnership Program for Creative Research Teams.

References

Brunke, M. A., Fariall, C. W., Zeng, X., Eymard, L., and Curry, J. A.,
2003
.
Which bulk aerodynamic algorithms are least problematic in computing ocean surface turbulent fluxes
.
J. Climate
16
,
619
635
.
Chao, Y., Li, Z., Farrara, J. D., and Huang, P.,
2009
.
Blended sea surface temperatures from multiple satellites and in-situ observations for coastal oceans
.
J. Atmos. Oceanic Technol.
26
,
1435
1446
,
Chelton, D. B., and Xie, S.-P.,
2010
.
Coupled ocean–atmosphere interaction at oceanic mesoscales
.
Oceanogr.
4
,
52
69
.
Chen, F., and Dudhia, J.,
2001
.
Coupling an Advanced Land Surface-Hydrology Model with the Penn State-NCAR MM5 Modeling System Part I: Model Implementation and Sensitivity
.
Mon. Wea. Rev.
,
129
,
569
585
.
Chow, C. H., and Liu, Q.,
2012
.
Eddy effects on sea surface temperature and sea surface wind in the continental slope region of the northern South China Sea
.
Geophys. Res. Lett.
39
,
L02601
, doi:.
Fairall, C. W., Bradley, E. F., Rogers, D. P., Edson, J. B., and Young G. S.,
1996
.
Bulk parameterization of air-sea fluxes for Tropical Ocean-Global Atmosphere Coupled-Ocean Atmosphere Response Experiment
.
J. Geophys. Res.
101
(
C2
),
3747
3764
.
Fairall, C. W., Bradley, E. F., Hare, J. E., Grachev, A. A., and Edson, J. B.,
2003
.
Bulk parameterization of air–sea fluxes: Updates and verification for the COARE algorithm
.
J. Climate
16
,
571
591
.
Frenger, I., Gruber, N., Knutti, R., and Münnich, M.,
2013
.
Imprint of Southern Ocean eddies on winds, clouds and rainfal
l,
Nature Geosci.
6
,
608
612
.
Liu, Q., Jiang, X., Xie, S.-P., and Liu, W. T.,
2004
.
A gap in the Indo-Pacific warm pool over the South China Sea in boreal winter: Seasonal development and interannual variability
.
J. Geophys. Res.
109
,
C07012
, doi:.
Liu, Q., Huang, R. X., Wang D. X., Xie Q., and Huang Q.,
2006
.
Interplay between the Indonesian throughflow and the South China Sea throughflow
.
Chin. Sci. Bull.
51
,
50
58
.
Qu, T.,
2002
.
Evidence for water exchange between the South China Sea and the Pacific Ocean through the Luzon Strait
.
Acta Oceanologica Sinica
,
21
,
175
185
.
Shi, R., Guo, X. Y., Wang, D. X., Zeng, L. L., and Chen, J.,
2014
.
Seasonal variability in coastal fronts and its influence on sea surface wind in the Northern South China Sea
.
Deep-Sea Res.
II
, http://dx.doi.org/
Skamarock, W. C., Klemp, J. B., Dudhia, J., Gill, D. O., Barker, D. M., Wang, W., and Powers, J. G.,
2005
. A description of the Advanced Research WRF Version 2.
NCAR Tech. Note NCAR/TN4681STR
,
NCAR
.
Small, R. J., deSzoeke, S. P., Xie, S.-P., O'Neill L., Seo, H., Song, Q., Cornillon, P., Spall, M., and Minobe, S.,
2008
.
Air-sea interaction over ocean fronts and eddies
.
Dyn. Atmos. Ocea
.
45
,
274
319
.
Stull, R.B.,
1988
. An introduction to boundary layer meteorology.
Kluwer Academic Publishers
,
Dordrecht, The Netherlands
.
Wang, D., Liu Q., Huang, R. X., Du, Y., and Qu, T.,
2006
.
Interannual variability of the South China Sea throughflow inferred from wind data and an ocean data assimilation product
.
Geophys. Res. Lett.
33
,
L14605
, doi:.
Xie, S.-P.,
2004
.
Satellite observations of cool ocean–atmosphere interaction
.
Bull. Am. Meteor. Soc.
,
85
,
195
208
.
Yan, J. Y., Tang, Z. Y., Yao, H. D., Li, J. L., Xiao, Y. G., and Chen, Y. D.,
2005
.
Air-Sea Flux Exchange over the Xisha Sea Area Before and After the Onset of Southwest Monsoon in 2002
.
Chin. J. Geophys
.
48
,
1078
1090
.