HTML
--> --> -->upwelling that result in positive and negative SST anomalies, respectively (Gaube et al., 2015; Sabu et al., 2015). Moreover, eddies under large background temperature gradients might also induce anomalous horizontal SST advection. In this situation, maximal SST anomalies can also be found on eddy edges (Nonaka and Xie, 2003; Chow and Liu, 2012; Hausmann and Czaja, 2012). Eddy-induced SST anomalies (normally <1°C) have remarkable influences on the overlying atmosphere, as revealed in the northeastern Atlantic (Bourras et al., 2004), South Atlantic (Frenger et al., 2013; Byrne et al., 2015), Aghulas Return Current (Messager and Swart, 2016), and Kuroshio Extension (Putrasahan et al., 2013; Ma et al., 2015, 2016; Chen et al., 2017). Two principal mechanisms have been cited to explain the atmospheric responses. The first is the vertical mixing mechanism, which emphasizes how eddy-induced SST anomalies change the stability of the marine atmospheric boundary layer (MABL) (Hayes et al., 1989; Wallace et al., 1989). A warm (cold) eddy makes the MABL unstable (stable), which intensifies (inhibits) turbulent mixing and strengthens (weakens) downward momentum transport, accelerating (decelerating) the surface wind. Under this mechanism, a wind anomaly should be situated directly over the area where the maximal SST anomaly exists, favoring a positive phase relationship between the surface wind and the SST anomaly. The second mechanism is the pressure adjustment mechanism (Lindzen and Nigam, 1987), which ascribes wind anomalies to modifications of sea level pressure (SLP) caused by anomalous heating over mesoscale eddies. An additional pressure gradient force occurs on the edge of the eddy where the maximum SST gradient exists. Consequently, the wind anomaly is 90° out of phase with the SST anomaly.
With different eddy features and against different environmental backgrounds, the atmospheric responses to mesoscale eddies and the corresponding physical mechanisms might differ (Spall, 2007; Small et al., 2008). The South China Sea (SCS), lying within the region (0°-23°N, 99°-121°E), is the largest semi-enclosed marginal sea in the tropics. Under the joint effects of the geographical environment and the local wind stress curl, many high eddy kinetic energy centers are embedded within the SCS, presenting a multi-eddy structure (Fang et al., 2002; Wang et al., 2003; Xiu et al., 2010). More than 40 mesoscale eddies with mean radius of about 150 km occur in the SCS annually and persist for 30-300 days. Isolating mesoscale SST signals from the background, (Chen et al., 2011) revealed an average SST anomaly of about 0.4°C due to anticyclonic eddies.
In summary, the SCS is an eddy-active zone. Previous research in this area has focused mainly on the formation or the physical properties of mesoscale eddies, while studies about their impacts on the atmosphere are rare (Chow and Liu, 2012; Sun et al., 2016). To determine systematically the atmospheric response to mesoscale eddies in the SCS, the objective of this study is to resolve the following issues: First, within the context of the monsoon climate zone, we seek to establish whether the impacts of mesoscale eddies on the atmosphere are significant over the SCS and, if so, to determine the principal mechanism behind the effects of such eddies on the overlying atmosphere.
The remainder of this paper is organized as follows: The data and methods used in the study are presented in section 2. In section 3, the atmospheric response to mesoscale eddies over the SCS is estimated, and the strength of the coupling between atmospheric and SST anomalies is assessed. In section 4, utilizing a regional atmospheric model, the leading mechanism of the atmospheric response to a pair of cold and warm eddies off eastern Vietnam is described. Finally, the study's findings are summarized in section 5.
In addition, four sets of daily satellite observations acquired during 2000-2012 were used:
(1) Optimally interpolated (OI) SST products provided by Remote Sensing Systems (RSS; http://www.remss.com). The 9-km-resolution Microwave-IR (MW_IR) OI SST was employed in this work to eliminate contamination from cloud cover and other errors related to rain, diurnal warming, and sea ice.
(2) Objectively analyzed air-sea fluxes (OA Flux) on a 1° grid, provided by the Woods Hole Oceanographic Institution, which include surface latent heat flux (LHF) and sensible heat flux (SHF), ocean evaporation, and air temperature at a height of 2 m.
(3) Blended vector sea wind data obtained from NOAA's National Centers for Environmental Information. Up to six satellite observations were used to produce the blended sea surface wind (at 10 m above the sea surface) at a spatial resolution of 0.25°.
(4) Altimetry SLA and absolute geostrophic velocity data provided by AVISO on a 0.25° grid (https://www.aviso.altimetry.fr/en/data.html).
Three-day mean precipitation, atmospheric water vapor, and cloud liquid water data from the TMI released by RSS (TMI v7.1) were also used.
For uniformity, the datasets mentioned above were interpolated linearly to a 0.25°× 0.25° horizontal resolution with the same grid points. To filter out high-frequency fluctuations related to synoptic phenomena (such as tropical cyclones and easterly waves), a three-week low-pass temporal (Butterworth) filter was applied to these datasets. Following the temporal filter, a spatial (Loess) filter (Cleveland, 1979) was applied to isolate the eddy's signature from the large-scale field, which is a method that has been demonstrated in previous studies (Small et al., 2008; O'Neill et al., 2010). Considering an average diameter of about 300 km for eddies in the SCS and an imprint twice as large on the atmosphere (Frenger et al., 2013), the Loess filter with a half-width of 4.5° was used to retain signals with spatial scales less than 4.5° latitude × 4.5° longitude. For comparison, half-widths of 3° and 6° were also examined. The results showed little numerical difference and, therefore, it was considered that different half-widths within a small range did not affect the integrity of the results of the present work.
Figure1. (a) Spatial distribution of the correlation between prefiltered SST and surface wind speed. Daily data from 2000 to 2012 were filtered; a three-week low-pass temporal (Butterworth) filter and a spatial (Loess) filter with a half-width of 4.5° were applied to all datasets. Correlations shown are significant at the 99% confidence level, based on the t-test. Contour lines denote correlation coefficients equal to 0.2 and 0.4. (b) Spatial distribution of 4511 eddies corresponding to 613 eddy tracks for the period December 1999 to March 2012. Occurrence frequency of eddies in each 1°× 1° grid is displayed based on color scale and numbers.Prefiltered variables (referred to as anomalies) were used to obtain composite maps. A mean composite map for mesoscale eddies was obtained via three steps. First, when an eddy reached its maximum (minimum) SLA, it was taken as the most energetic time of the eddy. Based on the eddy track dataset, the most energetic time for each individual eddy was identified. Then, the location and radius of the eddy and the corresponding oceanic and atmospheric anomaly fields at that time were identified and selected. Second, the oceanic and atmospheric anomaly fields were scaled relative to the individual eddy radii, and then interpolated onto the eddy-radius normalized coordinate. Since the atmospheric responses on the upwind side and downwind side of the eddy are different, the above anomaly fields were rotated according to the background wind direction, fixing the x-axis parallel to the wind direction. Finally, all qualified records were averaged after grid normalization. A distance of twice the eddy radius (2R) was chosen as the normalized axis length with normalized grid spacing of one-fifth the eddy radius.
For the case study in section 4, the Weather Research and Forecasting (WRF) model (ARW system, version 3.5) was employed because it has sufficient capability for mesoscale simulations (Skamarock et al., 2008). The following primary physics options were deployed: (1) WRF Single-Moment 6-class microphysics scheme (Hong and Lim, 2006); (2) RRTMG shortwave radiation scheme (Iacono et al., 2008); (3) RRTMG longwave radiation scheme (Iacono et al., 2008); (4) Eta similarity surface layer option (Janji?, 2002); (5) Noah Land Surface Model (Chen and Dudhia, 2001); (6) Mellor-Yamada-Janjic planetary boundary layer scheme (Janji?, 1994); and (7) Tiedtke cumulus parameterization scheme (Tiedtke, 1989; Zhang et al., 2011).
We applied the double nested dynamic downscaling method to simulate mesoscale features of the atmospheric response. A domain with a 9-km horizontal resolution nested inside a broader 27-km-resolution parent domain was used. For the mother domain, a one-way nesting approach was used, and the six-hourly NCEP FNL (Final) Operational Global Analysis data on a 1° grid were used as the model's initial and boundary conditions. The daily (MW_IR) OI SST data were interpolated temporally to a 6-h frequency and used as the bottom boundary condition to characterize mesoscale eddies in the model run. Analysis fields were horizontally interpolated to the model grids. For the nested domain, the two-way nesting approach was used to connect the coarse domain and the subdomain. To discern the detail of the boundary layer, meteorological fields were vertically interpolated to 43 unevenly spaced vertical levels, the lower 22 layers all within 1 km.
3.1. General features of mesoscale eddies and related mesoscale air-sea interaction over the SCS
Overall, 613 mesoscale eddies in the SCS were identified from the dataset of global eddy tracks for the period December 1999 to March 2012. More than 50 eddies were tracked per year, revealing an average lifetime, a mean radius, SLA and propagation speed of 6.4 weeks, 151.6 km, 6.1 cm, and 6.8 cm s-1, respectively. In addition, 99.4% of eddies in the SCS last longer than three weeks. Thus, the three-week low-pass temporal filter used in this study did not greatly attenuate the imprints of mesoscale eddies. Compared with mesoscale eddies in the open ocean (Chelton et al., 2011), eddies in the SCS are smaller and weaker; however, they still play a vital role in mesoscale air-sea interaction processes.
Figure2. Mean composite maps of SLA (contours) and SST anomaly (colors) for (a) cyclonic and (b) anticyclonic eddies. Prefiltered satellite-derived data corresponding to 237 cyclonic and 201 anticyclonic eddies in the SCS during 2000 to 2012 were averaged. The scale of both axes denotes the normalized distance (2R for twice the eddy radius) from the eddy center in each direction. The x-axis is aligned along the background wind direction, as indicated by the arrow. The SLA contour intervals are 2.5 cm. The white contour indicates the zero line, and the dashed lines (solid lines) represent negative (positive) values. Anomalies in areas without dots are statistically different from zero at the 95% confidence level, based on the t-test. For quality-control purposes, if more than 30% of the samples were unavailable in one grid, the grid is also dotted.Figure 1a shows the temporal correlation between the prefiltered SST and the surface wind speed in the SCS. As expected, much of the SCS is dominated by significant positive correlation. In Fig. 1b, two zones of vigorous eddies are paired with regions of positive correlation: one extends from southwest of Taiwan to the south of Hainan Island, and the other lies along the east coast of Vietnam. Besides these two zones, the band area from west of Luzon to the central coast of Vietnam (13°-15°N, 110°-119°E) is also an eddy-active area. Corresponding to this eddy-active area is a band of significant positive correlation oriented in a southwest-northeast direction from (13°N, 110°E) to (16°N, 117.5°E). Also noteworthy is that the cold currents along the continental shelf south of Vietnam (4°-9°N, 106°-110°E) might also play an important role in the positive SST-wind relationship during the northeast monsoon (Liu et al., 2005). In general, except the area off the western coast of Luzon Island, where the sea surface winds are dominated by orographic forcing, the similarity between the spatial features in Figs. 1a and b suggests the roles of eddies in affecting the atmosphere over the SCS are significant.
2
3.2. Atmospheric response to mesoscale ocean eddies: composite analysis
To provide further general information about the atmospheric response to mesoscale eddies, composite maps of SST and atmospheric quantities were constructed. To avoid interference from very weak eddies, only eddies with maximum SLA larger than 5 cm were included in the composite analysis. After doing so, 438 eddies (71.5% of the total) during 2000 to 2012 in the SCS were selected and divided into two categories: 237 cyclonic and 201 anticyclonic eddies, because of their opposite effects.3.2.1. Composite of SST anomaly
Figure3. Composite maps of the mean SST anomaly (contours; interval: 0.1°C; white contour, 0°C; dashed lines, negative values; solid lines, positive values) together with the (a, e) SHF anomaly (colors; positive upward), (b, f) LHF anomaly (colors; positive upward), (c, g) sea-air temperature difference anomaly (colors), and (d, h) evaporation rate anomaly (colors): (a-d) averaged cyclonic eddies; (e-h) averaged anticyclonic eddies.The annular negative and positive SLA contours in Figs. 2a and b represent the locations of the composited cyclonic and anticyclonic eddies, respectively. The eddy outlines are displayed clearly in the normalized coordinates within the scope of the eddy radius (R, within
3.2.2. Composite of turbulent heat fluxes
Turbulent heat flux is the principal connection between the ocean and the atmosphere. The composite maps of the mean SHF (positive upward) anomaly show a negative SHF center coincides with the cold eddy (Fig. 3a) and a positive anomaly center is located in the south of the warm eddy (Fig. 3e). LHF anomalies coincide well with the eddy centers, with a negative LHF anomaly over the cold eddy (Fig. 3b) and a positive LHF anomaly over the warm eddy (Fig. 3f). The composite SHF and LHF anomalies indicate the ocean obtains heat from the atmosphere over the cold eddy and loses heat to the atmosphere over the warm eddy. Quantitatively, the maximum LHF anomalies (-5.95 W m-2 for cold eddies and 2.3 W m-2 for warm eddies) are larger than the SHF anomalies (-1.04 W m-2 for cold eddies and 0.54 W m-2 for warm eddies).
Anomalous turbulent heat fluxes triggered by SST anomalies are essential to the physical response of the atmosphere to mesoscale eddies. As the SHF is proportional to the air-sea temperature difference and the LHF is proportional to the evaporation rate, the composite maps of the air-sea temperature difference (Figs. 3c and g) and evaporation rate (Figs. 3d and h) have the same patterns as the heat fluxes. As the air-sea temperature difference can reflect the atmospheric stability, the negative (positive) sea minus air temperature difference anomaly in Fig. 3c (Fig. 3g) indicates the stability is intensified (weakened) over the cold (warm) eddy. Moreover, eddy-induced evaporation rate anomalies (Figs. 3d and h) also represent the changes of atmospheric stability. More stable stratification over the cold eddy decelerates the evaporation rate, and vice versa. In addition, extra heating corresponding to turbulent heat flux anomalies might change the SLP as well. Ultimately, the change in both the SLP and atmospheric stability might affect the near-surface wind speed through the mechanisms mentioned above.
3.2.3. Composite wind speed, divergence and vorticity anomalies
Synchronous variations of SST and sea surface wind are crucial to the depiction of mesoscale ocean forcing. As expected, surface winds decelerate over the cold eddy (Fig. 4a) and accelerate over the warm eddy (Fig. 4b). The negative wind speed anomaly, with a minimum value of -0.19 m s-1, matches well with the cold eddy. Likewise, the positive wind speed anomaly, with a maximum value of 0.11 m s-1, is located over the warm eddy, with little phase shift. The nearly in-phase relationship between the wind and SST anomaly favors the vertical mixing mechanism for both cold and warm eddies. It is worth noting that the phase relationship between the SST and the surface wind speed anomaly is robust, i.e., it is unaffected by eddy strength.
Figure4. Composite maps of the mean surface (a, d) wind speed anomaly, (b, e) divergence anomaly, and (c, f) vorticity anomaly: (a-c) averaged cyclonic eddies; (d-f) averaged anticyclonic eddies.The surface wind anomalies over the mesoscale eddies have further effects on horizontal divergence and vorticity. In Fig. 4b, surface flows converge on the upstream of the cold eddy and diverge on the downstream, forming a dipole pattern. A similar north-south dipole vorticity anomaly pattern is also obvious for the cold eddy (Fig. 4c). To explain these dipole features, it is considered that the background wind blows from left to right; thus, convergent and divergent centers of the surface wind occur on the two sides of the eddy along the wind direction. Meanwhile, anticyclonic and cyclonic wind shear forms to the north and to the south of the eddy, perpendicular to the background wind. From this, it is evident that the in-phase relationship between the wind anomaly and the cold eddy is the fundamental cause of the dipole pattern. Therefore, the dipole pattern in the field of anomalous divergence for the cold eddy also supports the vertical mixing mechanism.
For the warm eddy, a single convergence center is located in the eddy center, with several surrounding centers of divergence (Fig. 4e), representing a monopole pattern. It is more sensible to attribute the monopole convergence core to a depression over the warm eddy, which suggests the pressure adjustment mechanism is most appropriate in this context. Furthermore, the most apparent feature of the composite vorticity anomaly for warm eddies (Fig. 4f) is an anticyclonic area centered on the warm eddy, which corresponds to the divergence area seen in Fig. 4e.
In terms of the composite surface wind speed, divergence and vorticity fields, the vertical mixing mechanism is dominant for cold eddies, while for warm eddies the pressure adjustment mechanism is equally important. One possible explanation for this difference is that the SCS is part of the Indo-Pacific warm pool; warm eddies induce weaker SST anomalies than cold eddies (Fig. 2), subsequently creating less of an air-sea temperature difference, whereby their impact on the vertical mixing is limited.
3.2.4. Composite precipitation, water vapor and cloud liquid water
Figure5. Composite maps of the mean (a, d) rain rate anomaly (colors), (b, e) columnar cloud liquid water anomaly (colors), and (c, f) columnar water vapor anomaly (colors): (a-c) averaged cyclonic eddies; (d-f) averaged anticyclonic eddies.Both anomalous surface heating and modification of the local circulation might change the characteristics of precipitation, atmospheric water vapor, and cloud over mesoscale eddies. In Fig. 5a, the cold eddy is sandwiched between two parallel rain bands with opposite signs and northwest-southeast alignment. The positive rain band is located on the left of the cold eddy, which is coincident with the convergence center upstream of the eddy seen in Fig. 4b. The negative rain band corresponds to the divergence center and anticyclonic area (refer to Figs. 4b and c). Both rain anomaly bands are statistically significant. For the warm eddy (Fig. 5d), positive rain anomalies within the eddy are too small to pass the significance test. The areas of negative precipitation anomaly surrounding the eddy center are the most prominent feature, which are related to the descending branch in Fig. 4e.
The rain rate anomaly related to eddies in the SCS reaches 1-2 × 10-2 mm h-1, which is comparable to the Kuroshio Extension region [1.5 × 10-2 mm h-1 (Ma et al., 2015)] and larger than the Southern Ocean [4 1× 10-3 mm h-1 (Frenger et al., 2013)]. The eddy-induced SST anomaly in both the Kuroshio Extension region and the Southern Ocean is 0.6°C and 0.5°C, respectively, much bigger than the SCS (0.2°C), which might be expected to cause greater precipitation. However, the SCS is located within the monsoon region, where the atmosphere is much moister than the Southern Ocean (south of 30°S). Furthermore, the SCS is near the region of the subtropical high and, therefore, the relatively smaller SST anomalies in the SCS generate comparable precipitation anomalies. In Figs. 5b and e, eddy-induced columnar cloud liquid water anomalies exhibit the same spatial features as the precipitation anomalies, where negative (positive) cloud water anomalies are coincident with areas of less (more) precipitation.
In addition to precipitation and cloud anomalies, the features of columnar water vapor also reflect the influence of mesoscale eddies on the entire atmosphere. For cold eddies, the negative columnar water vapor anomaly occurs over the eddy core (Fig. 5c), whereas for warm eddies several positive anomaly centers are found within and around the warm eddy (Fig. 5f). As is known, both surface heating and convergence transport can modify columnar water vapor. Comparison of the spatial pattern of the columnar water vapor anomaly (Figs. 5c and f) with the divergence (Figs. 4b and e) and heat flux (Figs. 3a, b, e and f) anomalies shows the water vapor matches better with heat flux. Therefore, it is supposed that the thermodynamic process of mesoscale eddies in the SCS is more effective than the dynamic process in changing columnar water vapor.
2
3.3. Quantitative descriptions of the coupling between eddies and atmospheric anomalies
To quantify the atmospheric responses to mesoscale eddies in the SCS statistically, the explained variations of atmospheric anomalies associated with eddies were calculated (Table 1). Cyclonic eddies explain 4.9% of the variance in the SST, while for anticyclonic eddies the value is 3.1%. That means the sea surface cooling induced by cyclonic eddies is more obvious than the surface warming caused by anticyclonic eddies in the SCS. In this case, atmospheric responses to cyclonic eddies are more significant in comparison to anticyclonic eddies. The explained variation of surface wind speed is 6% and 5.3% for cyclonic and anticyclonic eddies, respectively. In addition, 5%-7% of the perturbations of LHF, SHF, and evaporation rate can be explained by mesoscale eddies. However, anomalous precipitation and columnar water vapor represent only 3% and 1% of the natural variability, respectively, which suggests the impact of eddies on the upper-level atmosphere is limited in the SCS.Composite maps only reveal the mean state of the atmospheric response to mesoscale eddies. Therefore, scatterplots of SST anomalies versus atmospheric anomalies were produced to infer the atmospheric responses to varying SST anomalies. Based on the 438 eddy samples averaged for the composite analysis, the maximum SST anomalies within the eddy radius and the corresponding atmospheric anomalies in each sample were dotted and binned thereafter. First, if the intensity of an ocean eddy is measured by the SLA, then Fig. 6a shows stronger eddies are corresponding to the larger SST anomalies (r2 = 0.96) in the SCS. Using the least-squares estimation, the linear relationship between the SLA and SST anomaly is significant; a 17.28 cm increase in the SLA would raise the maximum SST anomaly by 1°C. Similar to previous studies on the Southern Ocean and the Kuroshio Extension regions (Frenger et al., 2013; Ma et al., 2015), obvious linear relationships were found to exist between the SST anomaly and atmospheric anomalies in the SCS. In addition, the slopes of the fitting curves were used to denote the strength of the coupling of the two variables. In agreement with the composite analysis (Figs. 4a and d), the surface wind speed increases with the SST anomaly (r2 = 0.96; Fig. 6b), and the coupling strength is 0.79 m s-1 °C-1, which is stronger than the Southern Ocean and the Kuroshio Extension (～0.4 m s-1 °C-1) (Frenger et al., 2013; Ma et al., 2015). In Fig. 6c, the correlation coefficient between the SST and turbulent heat fluxes anomalies is 0.99; a 1°C SST anomaly might provide additional heating (cooling) of 29.19 W m-2. The temperature difference between the sea surface and the air is also proportional to the SST anomaly (r2 = 0.99; Fig. 6d) and the coupling strength is 3.41°C °C-1, which indicates atmospheric stability changes linearly with SST anomalies over mesoscale eddies.
Through Table 1 and the scatterplots above, results show that the standard deviations of atmospheric anomalies according to mesoscale eddies in the SCS are larger than those in the Southern Ocean (Frenger et al., 2013), in agreement with (Sun et al., 2016). This is because a fair amount of eddies are active along the coastline, where complex topography makes the sizes, shapes and spatial distributions of the atmospheric anomalies tailored for different eddies. In addition, seasonal variations of basin-scale SST (Liu et al., 2005) and the eddy seasonal variations (Du et al., 2016) may also influence the variability of atmospheric responses. It is worth noting that the error of satellite observations used in this study may be comparable to the anomaly related to individual eddies. For example, the mean bias error of the OI SST, OA Flux, blended wind speed and TRMM rain rate is about 0.4°C, 7.4 W m-2, 0.2 m s-1 and 2 mm d-1, respectively (Yu et al., 2008; Scheel et al., 2011; Peng et al., 2013). Despite the potential bias, our results are significant being derived from a large number of eddies.
Figure6. Scatterplots of the eddy-induced SST anomaly (SSTA) versus the (a) SLA and (b-d) atmospheric anomalies including the (b) wind speed, (c) turbulent heat fluxes and (d) sea-minus-air temperature difference. Unbinned data (blue dots) represent the total of 438 eddy samples averaged in the composite analysis. The maximum SST anomalies within the eddy radius and the corresponding atmospheric anomalies in each sample were selected. Binned data (red dots with 0.25°C intervals) with error bars denote the mean value and the 1 standard deviation. The correlation coefficients (r1 and r2) and the slopes (s1 and s2) of the least-squares fit to the unbinned and binned data are shown as blue and red lines. All values are significant at the 99% confidence level.2
4.1. Eddy dipole in 2004 and its effect on the atmosphere
An eddy dipole, comprising a cold eddy north of the coastal jet off the central Vietnam coast and a warm eddy south of the jet, is a common feature in the southwestern SCS during the summer monsoon period (Chen et al., 2010; Hu et al., 2011). The case used here formed on 1 June 2004 and persisted for about 30 days. Two tropical depressions had passed by the eddy area. Influenced by these tropical depressions, eddy-related SSTs and surface wind speeds were obscured during the period 4-11 June. Furthermore, the developments of the two eddies were not synchronous during the lifetime of the eddy dipole. Thus, the period of study here was 16-21 June, when the dipole pattern was vigorous and the influences of additional weather systems were eliminated. As shown in Fig. 7a, the eddy dipole is clearly visualized by two SLA centers of opposite sign. The negative SLA with cyclonic currents denotes the cyclonic mesoscale eddy, and the anticyclonic mesoscale eddy with positive SLA and anticyclonic currents is to its south. Under the background southwesterly monsoon, SST anomalies (about -0.3°C for the cold eddy and about 0.5°C for the warm eddy) related to the mesoscale eddies are discernible in the original SST field (Fig. 7b). It should be noted that the variation between the SST and surface wind speed over the eddy dipole was simultaneous and consecutive.
Figure7. Satellite-observed (a) SLA (colors) and surface geostrophic currents (a) (vectors), and (b) SST (colors) and sea surface wind (vectors), averaged during 16-21 June 2004. The labels "CE" and "AE" represent cyclonic eddy and anticyclonic eddy, respectively.
Figure8. Spatial distributions of eddy-induced (a, c, e) surface wind speed anomalies and (b, d, f) turbulence heat flux anomalies: (a, b) spatially filtered satellite observations; (c, d) spatially filtered RSR data; (e, f) differences between RSR and SSR. The spatial (Loess) filter with a half-width of 4.5° was used in (a-d). SST anomalies of 0.2°C (solid line) and -0.2°C (dashed line) are superposed on each subplot to identify the warm and cold eddies, respectively. All variables were averaged for 16-21 June 2004.
Figure9. Vertical cross sections of horizontal wind speed (colors) and TKE (lines) anomalies from the RSR minus SSR differences along the gray line shown in Fig. 8e. Gray solid lines indicate the range of the warm eddy [(12.9°N, 111.5°E) to (14.0°N, 112.6°E)] and the cold eddy [(14.0°N, 112.6°E) to (15.0°N, 113.7°E)]. Averaging period: 16-21 June 2004.
Figure10. Spatial distribution of simulated (a) friction velocity (colors) and (b) SLP (colors) differences between RSR and SSR.
Figure11. (a) Curves of anomalous (RSR minus SSR) momentum budget terms along the gray line shown in Fig. 8e. Gray dotted lines indicate the ranges of the warm and cold eddies. (b-e) Spatial distribution of vector and scalar differences of momentum budget terms [(b) Coriolis force; (c) advection; (d) pressure gradient force; (e) vertical mixing] between RSR and SSR, at the second ($\sim 12$ m) level, averaged for 16-21 June 2004.
Figure12. Schematic diagram of the impact of mesoscale ocean eddies on the atmosphere for an eddy dipole in the southwestern SCS. Cold (blue) and warm (red) SST is corresponding to the cyclonic (north of 14$^\circ$N) and anticyclonic eddy (south of 14$^\circ$N), respectively. Downward (upward) turbulent heat flux anomalies over the cyclonic (anticyclonic) eddy tend to cool (warm) the MABL. This leads to an increment (reduction) in the stability of the MABL and abatement (intensification) of the vertical mixing over the cold (warm) eddy. Subsequently, the vertical wind shear is increased (decreased) and the near surface wind decelerates (accelerates) over the cold (warm) eddy. In addition, the heating anomaly makes the SLP higher (lower) over the cold (warm) eddy. The extra pressure gradient force is opposite (homodromous) to the background wind on the upwind side and homodromous (opposite) on the downwind side. Meanwhile, the pressure gradient force is almost balanced by the horizontal advection on the edges of mesoscale eddies. Accordingly, the vertical mixing mechanism is dominant for wind anomalies over the eddy dipole; the pressure adjustment mechanism is responsible for wind anomalies on eddy edges.2
4.2. Model experiments
The model ran for seven days for the period 15-21 June 2004. A control run with real daily (MW_IR) OI SSTs (the RSR run) and a sensitivity run with smoothed SSTs (the SSR run) were performed to isolate the eddy signals. In the SSR run, the original SST fields were filtered through a spatial (Loess) filter to retain signals larger than 4.5° latitude × 4.5° longitude, and then a smooth filter (10-point moving average) was adopted in the SST fields, meaning no more mesoscale features existed in the SSR run. The SST fields were updated at each time step for all simulation processes. The simulation results for the nested domain, outputted every 12 h from 16 June, were used in the following analysis.2
4.3. Primary mechanism responsible for atmospheric anomalies
Using the realistic SST field, the atmospheric responses to the eddy dipole were successfully rendered in RSR. Validation against satellite observations (Figs. 8a and b), the simulated mean surface wind anomaly ( 0.5 m s-1) and total turbulence heat flux anomaly ( 25 W m-2) are coincident with the eddy dipole (Figs. 8c and d). In addition to the spatial (Loess) filter used in RSR and the observational data (Figs. 8a-d), the eddy impact could be extracted from the difference between RSR and SSR without applying any filter (Figs. 8e and f). As shown in Figs. 8e and f, the wind speed and turbulence heat flux are altered because of the eddy dipole.Anomalous heat fluxes might change the turbulence kinetic energy (TKE) through buoyancy; therefore, TKE was outputted to characterize changes in turbulence intensity over the mesoscale eddies. The eddy-induced TKE anomaly (RSR minus SSR) is negative (positive) over the cold (warm) eddy, where turbulence is weakened (strengthened) within the boundary layer (Fig. 9). Based on the vertical mixing mechanism, a negative TKE anomaly over the cold eddy might obstruct vertical momentum transport. Therefore, higher momentums are isolated in the upper layer (above 200 m), leading to negative wind speed anomalies near the surface and positive anomalies in the upper layer over the cold eddy (Fig. 9). Conversely, enhanced turbulent mixing over the warm eddy leads to increased downward momentum transport, which accelerates the surface wind (Fig. 9). The vertical fluxes of horizontal momentum (\(-\rho\overline{u'\omega'}\) and \(-\rho\overline{v'\omega'}\)) can be represented by the friction velocity \((u_\ast=[(-\overline{u'\omega'})^2+(-\overline{v'\omega')^2}]^{\frac{1}{4}})\). In Fig. 10a, the spatial distribution of the u* anomaly related to the mesoscale eddies is also presented as a dipole pattern. It is evident that the vertical momentum flux is intensified over the warm eddy and weakened over the cold eddy, reconfirming the importance of the vertical mixing mechanism. However, in Fig. 10b, the SLP over the eddy dipole is also changed, showing a rise of about 0.05 hPa over the cold eddy and a decrease of about 0.03 hPa over the warm eddy, which suggests additional involvement of the pressure adjustment mechanism.
To explain the variation of surface wind over the eddy dipole further, the dominant terms in the atmospheric horizontal momentum equations were investigated. In the WRF model, the horizontal momentum equations can be written as: \begin{eqnarray} \dfrac{\partial U}{\partial t}&=&-(\nabla\cdot {V}_{\rm 3d}u)-\mu\alpha\dfrac{\partial p}{\partial x} -\left(\dfrac{\alpha}{\alpha_{\rm d}}\right)\dfrac{\partial p}{\partial\eta}\dfrac{\partial\varphi}{\partial x}+fV+F_{U,{\rm mix}} \ \ (1)\\ \dfrac{\partial V}{\partial t}&=&-(\nabla\cdot {V}_{\rm 3d}v)-\mu\alpha\dfrac{\partial p}{\partial y} -\left(\dfrac{\alpha}{\alpha_{\rm d}}\right)\dfrac{\partial p}{\partial\eta}\dfrac{\partial\varphi}{\partial y}-fU+F_{V,{\rm mix}}\quad \ \ (2)\end{eqnarray} Variables U and V are defined as U=(μ u)/m and V=(μ v)/m, where u and v are the horizontal wind components in the x and y directions, respectively, m is the map-scale factor, and μ is the mass of dry air per unit area within a model column. Here, V 3d is the three-dimensional coupled vector velocity [V 3d=(U,V,W), W is the coupled vertical component of velocity], η represents the terrain-following pressure vertical coordinate, and FU, mix and FV, mix represent the vertical mixing. The other parameters are commonly used variables: t is time; p is pressure; f is the Coriolis parameter; φ is the geopotential; and α and α d represent the inverse density of full parcel density and dry density, respectively [for further details, see (Skamarock et al., 2008)]. In Eqs. (1) and (2), the terms on the left represent the horizontal momentum tendencies. The first term on the right-hand side is the atmospheric advection term; the second and third terms represent the atmospheric pressure gradient force; the fourth term is the Coriolis force; and the final term is the atmospheric vertical mixing term (the curvature and horizontal diffusion terms are omitted here).
The vector differences of the momentum budget terms between RSR and SSR are shown in Fig. 11. Against the background of the southwesterly monsoon (Fig. 7b), a wind reduction is represented as anomalous northeasterly winds; thus, the northwestward Coriolis force terms located over the cold eddy are reasonable (Fig. 11b). Contrary to the cold eddy, anomalous southeastward Coriolis force terms indicate anomalous southwesterly wind, indicating a wind increase over the warm eddy. Compared with the Coriolis force term, the advection term is much larger in the RSR minus SSR field (Fig. 11c). Horizontal advection is convergent over the cold eddy and divergent over the warm eddy. In contrast to the advection term, the pressure gradient forces term is divergent in the cold-eddy area and convergent in the warm-eddy area (Fig. 11d). This is entirely reasonable because the SLP increases and decreases over the cold and warm eddies, respectively, as noted in respect to Fig. 10b. The additional pressure gradient force is opposite (homodromous) to the background wind on the upwind (downwind) side of the cold (warm) eddy, which finally leads to surface wind deceleration (acceleration). The vertical mixing term in Fig. 11e is almost opposite to the anomalous wind vector and it behaves as a drag term in the MABL. For clearer recognition of the contribution of each term in the momentum equation to the surface wind anomaly, their scalar differences between RSR and SSR are also depicted in each subplot. First, the Coriolis force is proportional to the wind speed and, therefore, the decreasing (increasing) wind speed over the cold (warm) eddy is represented as the negative (positive) Coriolis force in Fig. 11b. Second, the advection term accelerates the surface wind upstream of the cold eddy and decelerates the wind speed on the downstream side (Fig. 11c). Similarly, the effects of the pressure gradient force on the surface wind are located mainly on the edges of the mesoscale eddies (Fig. 11d). The anomalous mixing term in Fig. 11e presents a dipole pattern, which is decreased and increased over the cold and warm eddies, respectively, just as expected based on the vertical mixing mechanism. Finally, the eddy-induced anomalies for each momentum term across the eddy dipole are compared in Fig. 11a. The pressure gradient force term and the advection term are largest near the eddy boundaries and they are balanced with each other. The vertical mixing term is bigger than the pressure gradient force term, and it mainly affects the surface wind just over the eddy center. Thus, although both vertical mixing and SLP anomalies are responsible for the wind anomalies, the mixing term is more effective, especially over the eddy dipole. Synthesizing the above analysis, a schematic map of the atmospheric response to the eddy dipole is presented in Fig. 12.
The physical mechanisms of the atmospheric responses to mesoscale eddies were studied through dynamical diagnosis and modeling. Based on the composite analyses, wind anomalies over cold and warm eddies are almost co-located with the SST anomalies, which in combination with the significant changes in atmospheric stability (sea-air temperature difference) represent the character of the vertical mixing mechanism. In addition, the dipole pattern of the divergence anomaly over cold eddies favors the vertical mixing mechanism as well. In contrast, the monopole pattern corresponding to the anomalous divergence over warm eddies suggests the pressure adjustment mechanism also plays a role in affecting the atmosphere. The model study of an eddy dipole off eastern Vietnam showed that heat flux anomalies induced by mesoscale eddies lead to changes in TKE and friction velocity, indicating weak vertical momentum transport over cold eddies and intense vertical momentum transport over warm eddies. Anomalous momentum fluxes ultimately result in changes in surface wind speed. Diagnosis of the atmospheric horizontal momentum equations suggested that surface wind anomalies are produced by the advection, pressure gradient force, and vertical mixing terms. Among them, the pressure gradient term is almost balanced by the advection term; thus, the vertical mixing term dominates the surface wind anomalies over the eddy dipole.
The present study focused on a systematic exposition of the imprint of mesoscale eddies on the atmosphere over the SCS. Recent studies have shown the surface wind response to mesoscale eddies in the SCS has regional diversity and seasonal variation (Chow and Liu, 2012; Sun et al., 2016). In addition, atmospheric anomalies generated by mesoscale eddies might have feedback loops to the eddies themselves (McGillicuddy, 2015; Seo et al., 2016). Therefore, more detailed studies on the effects of eddies under different atmospheric and oceanic conditions should be undertaken. Combining in-situ observations of drifting buoys or field cruises with a fully coupled atmosphere-ocean model would be helpful in obtaining conclusions that are more realistic and comprehensive.
