1 INTRODUCTION

Mesoscale eddies are ubiquitous in the global oceans and contain most of the ocean's kinetic energy (Macdonald and Wunsch, 1996). They may affect the air-sea interaction by influencing surface wind, latent and sensible heat fluxes, cloud liquid water, water vapor content, and precipitation rate (Ma et al., 2015). In addition, they can effectively transport heat, salt, and other properties of ocean water, and hence redistribute temperature, salinity, chlorophyll, etc. The local redistribution of temperature by eddies has been considered lying vertically as a result of eddy-induced vertical displacement of the isopycnal surfaces (Chelton et al., 2007; Chen et al., 2011; Yang et al., 2013; Zhang et al., 2013). In this process, cyclonic/anticyclonic eddies usually elevate/depress the isopycnal surfaces, and the mixed-layer water is lifted/pushed to a shallower/deeper layer, resulting in a cold/warm core in the interiority of eddies (Chelton et al., 2007, 2011b; Chen et al., 2011; Yang et al., 2013; Zhang et al., 2013). Moreover, studies have also revealed that eddies can lead to a local redistribution of temperature by stirring the background temperature field (Hausmann and Czaja, 2012; Frenger et al., 2015; Gaube et al., 2015; Amores et al., 2017). In this process, eddy rotation can advect the background temperature and lead to an anti-symmetry of temperature on the opposite side of the eddy. The temperature anomaly pattern caused by the former effect is referred to as a monopole pattern, while that caused by the latter effect is referred to as the dipole pattern. As a result, the structure of eddy-induced sea surface temperature anomalies (SSTa) is a superposition of a monopole core due to the elevation/depression of water in the eddy interior and a dipole structure due to the advection of the background SST field. Therefore, it is important to detect both monopole and dipole structures of eddies in order to understand the eddy impact on temperature deviation from the background.

Moreover, in addition to local instantaneous (associated with eddies) temperature displacement, due to their propagation and nonlinear nature (Chelton et al., 2011a), eddies can also lead to irreversible change of background temperature via divergence/convergence of heat fluxes induced by stirring, and/or via trapping and transporting heat, contributing greatly to the large scale heat balance. The monopole and dipole components are responsible for eddy trapping and stirring heat fluxes, respectively. Particularly, the trapping fluxes can be understood as following. During the early stages in the area of eddy generation, cyclonic/anticyclonic eddies induce heat and salt transport vertically due to the convergence/divergence (i.e., downwelling/upwelling effect); then the nonlinearity enables them to trap the water of the origin area at a certain temperature, and carry them to a long distance; finally eddies decay and release the trapped heat in an environment entirely different from which they are trapped (Chelton et al., 2011a). The trapping effect may thus lead to up-gradient fluxes, unlike the stirring effect that is believed responsible for the down-gradient fluxes. The commonly used method to evaluate eddy heat flux is to estimate an Eulerian framework, which ignores the eddy trapping effect (Stammer, 1998; Qiu and Chen, 2005; Souza et al., 2011). The eddy fluxes calculation usually takes the following procedures: firstly, meridional velocity anomalies, as well as the temperature anomalies are determined at grids of study region; secondly, the eddy (temperature) fluxes are calculated as the covariance between the velocity and temperature anomalies.

Previous studies have investigated the mesoscale eddy's influence on local temperature deviations from the background temperature, and accordingly also estimated the potential change to the background temperature via eddy fluxes, in some specific regional areas. For example, Hausmann and Czaja (2012) estimated eddy swirl and drift heat transports in mixed-layer, which is identical to the stir-and-trap effect. Composites that conducted by Gaube et al. (2015) in several regions in the global ocean were segregated by the amplitude of each eddy and revealed that the eddy SSTa pattern converges toward a monopole structure as eddy amplitude increases. Frenger et al. (2015) compared average SSTa structure of mesoscale eddies north of the Antarctic Circumpolar Current (ACC) with those associated with the ACC, divided the SSTa structures into a monopole and a dipole, and finally estimated heat transports for individual eddies in the region associated with stirring and trapping. Amores et al. (2017) investigated mesoscale eddy SST/Sea Surface Salinity (SSS) anomaly patterns in a low-EKE (Eddy Kinetic Energy) and high-SSS area in the North Atlantic subtropical gyre, and drew a conclusion that the sea surface anomalies patterns can be reconstructed by a linear combination of a monopole and a dipole.

However, a question remains unanswered: To what degree are the magnitudes of the two temperature components (monopole and dipole), and hence the eddy heat transports, affected by the intensity of background SST gradient? In this study, we address the above questions by analyzing eddies and their effects in the western North Pacific (WNP). As a hot-spot of mesoscale eddies (Liu et al., 2012; Yang et al., 2013; Ma et al., 2015), the WNP, particularly in the region around the Subtropical Countercurrent (STCC) remains largely unexplored in terms of eddy-induced temperature structures and eddy heat transports. In addition, the STCC region covers the northern part of the Pacific warm pool, with higher SST gradient to the north and lower SST gradient to the south, and the eddies from both northern and southern parts tend to converge at the junction (~21°N) (Liu et al., 2012; Yang et al., 2013), hence is a suitable place for the investigation. We first analyze the eddy characteristics in the WNP, focusing on the convergence of eddies toward 21°N, and then we determine the two components of eddy-induced SST anomalies in two subareas with totally different SST background. We then investigated the relationship between the intensity of the SST gradient and the eddy-induced SSTa structure. Finally, we estimate meridional and zonal eddy heat transport in the two subareas.

2 STUDY REGION, DATA, AND METHOD

The study region is the subtropical WNP (5°-35°N, 120°-180°E), where energetic mesoscale eddies are featured (Fig. 1). The northern half of the study region locates in the northern edge of the western Pacific Ocean warm pool where the SST gradient is more intense. Meanwhile, the southern half is located inside the warm pool where the SST is more homogeneous. Eddy information used in our study is from a global eddy dataset described in details by Chelton et al. (2011b), which consists of eddies with lifetimes of at least 4 weeks and amplitudes of at least 1cm. The eddies are detected and tracked from sea level anomaly fields, spanning the period January 1993 to April 2015 (http://wombat.coas.oregonstate.edu/eddies/). We used eddy properties, including center location, radius, amplitude, and life span of the study region.

The average spatial pattern of eddy-induced SSTa was obtained using weekly SST measurements from 2002 to 2009, which were derived from the Advanced Microwave Scanning Radiometer for Earth Observing System (EOS) (AMSR-E) sensor onboard the EOS Aqua satellite. The spatial resolution of the data is 0.25°. To first order approximation, AMSR-E SST can be representative of the mean temperature of the mixed-layer (Hausmann and Czaja, 2012), whose depth was approximated here by the 2°×2° monthly climatology of de Boyer Montégut et al. (2004).

To characterize the spatial pattern of eddy-induced sea surface height anomaly (denoted SSHA), we used the weekly surface dynamic topography data (SSH data) by AVISO on a 1/3°-Mercator grid (https://www.aviso.altimetry.fr/). The data is a multi-altimeter product merged from at least three altimeter missions in the period of 1993 to 2013.

Following procedures provided by Frenger et al. (2015), SSHa is obtained by subtracting the climatological monthly mean SSH field from the original weekly SSH data. The climatological monthly mean SSH field is calculated based on the original weekly SSH data. For consistency, SSTa is obtained following the same steps.

Composition analysis is applied to find the consistency between eddies and eddy-induced SSHa and SSTa from 2002 to 2009. The procedure to determine the SSTa and SSHa patterns associated with eddies are described below:

Compositions are conducted for eddies in lifetime of at least 5 weeks, and only eddy snapshots that are in the 'stable stage' during their lifetime are selected, and the evolution of the eddies in terms of their mean radius and amplitude along with their (standardized) lifetime was analyzed. A 'stable stage' lasts from 1/5 to 4/5 of the whole eddy lifetime was found, during which both eddy size and amplitude remain nearly stable.

For each eddy snapshot, select the SSTa field that occurred within ±3 days and co-located within three eddy radii of the eddy center.

Rotate the selected SSTa fields in an azimuth determined by the orientation of the large-scale SST gradient. The orientation of the large-scale SST gradient for each eddy snapshot is calculated by spatial-averaging the directions of the background SST gradient over a 4°×4° box centered in the eddy center. This ensures that, after the rotation, the composite eddy is derived on a frame in the same background SST gradient direction. This direction is defined as the "south" here.

Fix the rotated SSTa fields laterally according to the distance from the eddy center, normalized by the individual eddy radius into 0.1×0.1 grid points within a box up to a distance of two eddy standardized radii (R₀) using the IDW (inverse-distance weighted) method. For each grid point, SSTa located within the horizontal range of L=0.1R₀ is set with a weight value of W_i=(L²-d²)/(L²+d²), where L is assigned a value of 0.1 and d is the distance between the rotated SSTa grid and the point to be composited.

Operate the same procedures as above and obtain the SSHa composites.

In addition, we also used the t-test to determine the significance of composite temperature anomaly patterns.

3 RESULT 3.1 Eddy characteristics

Previous studies have found that eddies in this region are concentrated in 21°N (Liu et al., 2012; Yang et al., 2013). We confirm this finding with more eddy characteristics. In particular, we focus on two subareas separated by 21°N line, which are hereafterin referred to as WNP-N (130°-170°E, 21°-28°N) and WNP-S (130°-170°E, 14°-21°N) (dashed magenta boxes in Fig. 1).

Within the study region and period, 2 471 (44.3%) cyclonic and 3 106 (55.7%) anticyclonic long-lived eddies (life span of at least 4 weeks) are identified. Among them 1 411 (1 764) cyclonic (anticyclonic) eddies are generated in WNP-N, more than WNP-S, of which 1 060 (1 342) are cyclonic (anticyclonic) eddies. In addition, WNP-N has the most long-lived eddies (Fig. 2a, b). Regardless of lifespan, it is found that all eddies in the two subareas show a similar evolution pattern. Figure 3 shows the evolution of mean eddy radius and amplitude for both cyclonic and anticyclonic eddies in the two subareas. It clearly shows that the eddies often grow fast from birth to the first fifth of their lifetime, stay stably from the first fifth to the fourth fifth of their lifetime, and then quick decay at the last fifth of their lifetime.

The instantaneous horizontal (zonal/meridional) propagation speed of the continuously moving eddies are determined following Yang et al. (2013). The eddy speed at time t is calculated firstly by subtracting the longitude/latitude of the snapshot at the time (t-tI) from the longitude/latitude of the snapshot at the time (t+tI) and then divided by the two-time intervals (2tI). The method of speed estimation is different from that used by Chelton et al. (2007) from the local least squares fits of the longitudes of eddy centroids as a function of time. The speed calculated by using our method may be slightly slower than that estimated in local least square fits, according to Yang et al. (2013). It is obvious that both CEs and AEs move westward in general (Fig. 4a), the same as argued by previous studies (Chelton et al., 2011b; Yang et al., 2013). The velocity decreases as the latitude gets higher, which coincides with the tendency of the first baroclinic Rossby wave velocity. The mean propagation velocity of eddies average over different latitudes is about 8 cm/s (Fig. 4). Specifically, as shown in Fig. 4a, ignoring the latitudinal change by the west boundary current (higher than 24°N), the average westward speed of cyclonic and anticyclonic eddies are similar at latitudes higher than 18°N, whereas the speed of anticyclonic eddies become much larger than that of cyclonic eddies at lower latitudes.

In addition to the zonal propagation, eddies also have meridional movement. A prominent feature is that cyclonic (anticyclonic) eddies in latitudes higher than 21°N and lower than 21°N tend to move equatorward and poleward, respectively, in the avereage speed of -0.21 (-0.38) cm/s and 0.47 (0.24) cm/s, respectively (Fig. 4b). This is confirmed by the relative propagation trajectories of long-lived eddies that last longer than 12 weeks (Fig. 5). It is found that about 70% cyclonic eddies and 60% anticyclonic eddies move equatorward in WNP-N, whereas, nearly 60% cyclonic eddies and 70% anticyclonic eddies in WNP-S move poleward. This agree with Liu et al. (2012) and Yang et al. (2013); however, the feature of meridional movement in this study is different from the global eddy theoretical and statistical meridional movement direction (Chelton et al., 2007; Dong et al., 2014), indicating that cyclonic and anticyclonic eddies move poleward and equatorward, respectively.

The convergence of eddy movement into 21°N leads to the significant consequence of eddy properties. The most prominent one is shown in terms of eddy amplitude. The highest average amplitude centers around 18°-24°N for both polarities (Fig. 6d-f), and decreases as the latitudes get higher or lower. It is also clearly shown (Fig. 3b) that eddy amplitude is about 10% larger in WNP-N than WNP-S. As the eddy amplitude and eddy intensity (or rotation velocity) shall agree to each other roughly, our finding is supported by Yang et al. (2013) who found the highest EKE at around this latitude.

The latitude-dependent eddy radius also indicates the concentration in 21°N (Fig. 6a-c). It is already shown by the global distribution that the zonally-averaged eddy radius varies with latitudes in general: the larger eddies, the lower latitudes they situated (Chelton et al., 2007). This is confirmed by the results that the average radius of both cyclonic and anticyclonic eddies are about 50% larger in WNP-S than WNP-N (Fig. 3a). However, larger average radius is found at 18°N to 24°N than the surrounding latitudes (Fig. 6a-c). It is the concentration of eddies towards latitudes around 21°N, which results in the gathering there of mature eddies with larger radius and finally brings up a larger average radius.

Moreover, we calculated the total numbers of eddy birth and termination in 1°-wide latitudinal bands during the study period (Fig. 2c). In fact, 20°-22°N has the minimum number of eddy birth and termination, with ~150 anticyclonic eddies and ~175 cyclonic eddies firstly detected around this latitude; however, ~170 anticyclonic eddies and ~185 cyclonic eddies disappeared here, with an increasing rate of ~13% and ~6%, respectively, whereas in latitudes 25°-30°N (for both polarities), 17°-20°N (for cyclonic eddies) and 12°-18°N (for anticyclonic eddies), numbers of eddy termination are smaller than eddy birth. Therefore, it seems that the phenomenon of eddies convergence toward 21°N is associated with and confirmed by more eddy termination than eddy birth around 21°N and opposite trends at higher and lower latitudes.

3.2 Eddy-induced sea surface pattern

In this section, we reconstruct both cyclonic and anticyclonic eddies and the associated SSTa pattern by compositing all the detected sea surface eddies. It shows that the composite eddy-induced SSTa and SSHa are mainly centered within twice the radius of the eddy (Fig. 7), which is coincide with the composite result in Gaube et al. (2015). The average amplitude of cyclonic eddies is about 10 cm in both WNP-N and WNP-S, whereas that of anticyclonic eddies is clearly smaller in WNP-S (< 8 cm) but has a similar amplitude to the cyclonic eddies' in WNP-N (10 cm). The composite eddy in the WNP-N has larger temperature anomalies than in the WNP-S (Fig. 7). In WNP-N, there is a negative SSTa maximum (~ -0.5℃) of cyclonic eddies, whereas a positive SSTa maximum (~0.4℃) in anticyclonic eddies. In WNP-S, there is a negative SSTa maximum (~ -0.4℃) of cyclonic eddies, whereas a positive SSTa maximum (~0.2℃) in anticyclonic eddies.

Another important feature is that the SSTa pattern does not coincide with the SSHa pattern (Fig. 7). It is also shown in a previous study (Hausmann and Czaja, 2012) that there is a systematic shift both in zonal and meridional directions between maximum SSTa and maximum SSHa. For both anticyclonic and cyclonic eddies, the composite SSTa maximum occurs slightly to the west of maximum in SSHa. In addition, over anticyclonic eddies, maximum values are also shifted poleward; whereas over the cyclonic eddies, maximum values barely show meridional shifted. The shift in position is up to half of the average radius. Since the contours of SSHa is aligned with the stream function of the rotation velocity, the discrepancy is crucial for the eddies to induce net heat fluxes (Hausmann and Czaja, 2012), which will be described in the following section.

Figure 7 indicates that both composite eddies have monopole and dipole components in SSTa. The two components are further distinguished and quantified (Figs. 8-10). Following Frenger et al. (2015), the monopole component is calculated by radial averages around the eddy center, and the dipole pattern is regarded as the residual between the total and the monopole. Figure 8 clearly displays the two patterns for both anticyclonic and cyclonic eddies in both subareas. Therefore, the former is a symmetric structure with a maximum in the center and decreases smoothly outward, representing the SSTa pattern induced by upwelling/downwelling. This sort of SSTa could be trapped by eddy and transported to far distance. The latter resembles a dipole structure, which is similar to the eddy-induced chlorophyll anomaly structure shown by Chelton et al. (2011a), representing the eddy swirl distorting the large-scale SST gradient when the SST gradient is north-south oriented to the first order. Figure 9 shows the zonal distribution of the SSTa and SSHa of the two components: the two patterns are pretty distinguishable; in addition, the unaligned SSHa and SSTa distributions are also clearly illustrated.

However, the monopole and dipole components contribute to the total SSTa differently (Fig. 8). In the WNP-N (Fig. 8a, b, e, f), the monopole part of cyclonic and anticyclonic eddies both have an amplitude of about ±0.25℃, while the dipole part of cyclonic and anticyclonic eddies have a magnitude of ~0.2℃. Therefore, in this subarea, both components have similar magnitude. In contrast, in the WNP-S (Fig. 8c, d, g, h), the monopole part has an amplitude of about -0.25℃ for the cyclonic eddies, and has an amplitude of about 0.1℃ for the anticyclonic eddies, while the dipole part of cyclonic eddies has a magnitude of ~0.1℃, and that of the anticyclonic eddies has a magnitude of ~0.05℃. It, therefore, can be concluded that in this subarea the monopole part is larger than the dipole part for mesoscale eddies. For comparison, the dipole part of both cyclones and anti-cyclones in WNP-N has amplitudes about two times of those in WNP-S.

Following Frenger et al. (2015), we calculated the variances of the SSTa (V_total) and those that can be explained by the monopole /dipole contribution of SSTa (V_monopole/V_dipole):

where i denotes the index of SSTa within the composite eddy (two eddy radii), and n is the total numbers.

The variances can indicate how eddies may induce SST variances of the sea surface in the study area. The results are labeled in Fig. 8. In the WNP-N, the variance explained by the dipole part is similar to the monopole part, 51% vs 49% for the cyclonic, and 45% vs 55% for the cyclonic eddies. However, in the WNP-S, the dipole part is weak and contributes much less than the monopole part, 13% vs 87% for the cyclonic eddies; and 33% vs 67% for the anticyclonic eddies. For comparison, the variance explained by the dipole SSTa is much higher in WNP-N than that in WNP-S, especially for the cyclonic eddies (51%/13%); while the variance that is explained by the monopole SSTa is much less in WNP-N than that in WNP-S, especially for the cyclonic eddies (49%/87%).

The differences indicate that the effect of eddy distorting the isotherms (i.e., the dipole component) via eddy circular advection is more intense in WNP-N where large-scale SST gradient is larger and isotherms are denser. It is so large that it can even exceed that caused by eddy upwelling/downwelling (i.e., the monopole component) (51% vs 49%). It also indicates that in the WNP-S, locating in the interior of the warm pool where the background SST gradient is weak, the dipole component is smaller, and the elevation/ depression of isopycnals associate with cyclonic/ anticyclonic eddies (i.e., the monopole component) is the major mechanism of eddy-induced SSTa and SST variances. This will be demonstrated in the following section.

3.3 SSTa varies with large-scale SST gradient

The above results reveal that the dipole pattern of SSTa in an area with a larger background SST gradient differs greatly from the area with more uniform SST. In order to build a relationship between the intensity of the two components and the background large-scale SST gradient, we reclassify eddies in both the two subareas into 10 types in their background large-scale SST gradient (statistics numbers are shown in Table 1), and composite them (Fig. 10). When the gradient value is smaller than 0.2℃/100 km, eddy-induced SSTa almost converges to a monopole structure; however it becomes close to a dipole structure when gradient value exceeds 0.9℃/100 km, even though the monopole SSTa magnitude for cyclonic eddies reaches ~0.5℃. For the magnitude of eddy-induced SSTa, both the monopole and dipole parts increase as the background gradient intensity grows (Fig. 11). In the study region, when background SST gradient intensity grows larger, it usually indicates that the eddies situate farther from the center of the warm pool, where the thermocline becomes shallower. Then the monopole SSTa, which is mainly due to the cyclonic (anticyclonic) eddy's elevation (depression) to the thermocline, will increase with the gradient intensity. However, for the variances that explained by the eddy-induced SSTa, the variance from the monopole part decreases whereas the variance from the dipole part increases as the background gradient intensity grows. It is indicated that the background gradient has a great impact on eddy's SSTa pattern, especially for the part induced by eddy's stirring of isotherms.

Overall, it is confirmable that eddy-induced SSTa structure is deeply affected by the intensity of the background SST gradient field. At a place where background SST field is more uniform, the eddy-induced SSTa converges to a monopole structure, since eddy elevating/depressing isopycnals plays a leading role. However, at a place where SST varies sharply, the dipolarity of eddy-induced SSTa is more evident, since eddy swirl velocity distorting isotherms is a more dominating mechanism.

3.4 Monthly climatology of the eddy-induced SSTa

Now that eddy-induced SSTa structure depends on the intensity of background SST gradient, and that the SST field obviously shows seasonal variation, eddies from both subareas are reclassified by month and re-composited. The magnitudes and variance of both the monopole and dipole SSTa show extinct seasonal modulation (Fig. 12). The seasonal variation patterns of the SSTa magnitude are different between the monopole and dipole parts. For the monopole part, it is clearly small in summer and autumn and large in winter and spring, whereas for the dipole part, it seems highest in spring and smallest in autumn. The seasonal variation is larger for the monopole part and smaller for the dipole part. No significant differences are found between cyclonic and anticyclonic eddies.

It seems that the background SST gradient field has similar seasonal modulation to the seasonal variation of the dipole SSTa, implying the impact of SST gradient on the magnitude of SSTa. On the contrary, the eddy intensity does not show a similar variation pattern to either of the monopole or dipole part, indicating that it has little influence on the variance of SSTa.

The seasonal variation of monopole and dipole SSTa can be explained by the meridional shift of the three-dimensional structure of the Western Pacific Ocean warm pool. In summer and autumn, warm pool expands northwards and the study area possesses relatively uniform SST, consequently, both parts of SSTa weaken in magnitude and have little difference. However, in winter and spring, warm pool retreats and isotherms in the study region intensifies, in consequence, both parts of SSTa enhance in magnitude whereas the magnitude of monopole SSTa is clearly larger than that of dipole one. In view of the monopole SSTa, the warm pool thickens in summer and thins in winter. Thus, the elevation or depression effect of thermocline weakens in summer and intensifies in winter; therefore the variance of monopole SSTa is lowest in summer and highest in winter, whereas the variance of dipole part is completely opposite. Overall, the seasonal modulation of the two components of SSTa also certifies the decomposition of eddy-induced SSTa and their corresponding mechanisms are practical.

3.5 Eddy-induced meridional heat transport

Based on the composite SSHa distribution, the eddy average swirl velocity is computed by using the geostrophic relationship. The distribution of the velocity along the zonal diameter is shown in Fig. 9 (dashed lines). The figure shows that the maximum of dipole SSTa occurs around 1 radius from the eddy center, the swirl velocity occurs nearby, and they both decrease to zero at around two radii from the center. As already mentioned, the discrepancy is crucial for the eddies to induce net heat fluxes (Hausmann and Czaja, 2012), as demonstrated by the equations below.

Following Frenger et al. (2015), eddy trapping heat transport (Q_trap) and stirring heat transport (Q_stir) in the mixed layer are estimated in the two subareas as:

with an average density of ρ₀=1 025 kg/m³, a specific heat capacity of c_p=4 000 J/(kg·K), an assumed mean mixed layer of h_m=100 m. Integration is conducted inside 2 eddy radii (R) from the eddy center, around where the rotation velocity decays to zeros. R is ~95 km in WNP-N and is ~100 km in WNP-S. A reasonable constant mixed layer depth was also adopted by other studies (Hausmann and Czaja, 2012). The subscripts trap and stir denote heat transport generated from monopole SSTa moving along with the eddy, and heat advection from both the dipole SSTa and the swirl velocity, respectively. The superscripts meridional and zonal denote meridional and zonal heat transports across the composite eddy. U_p and V_p are them mean of the eddy propagation velocities in the zonal and meridional directions (shown in Table 2). y=0 and x=0 denote the east-west section and north-south section through the eddy center, respectively. The integration range is set to two normalized radii where the eddy swirl velocity is approaching zero. The resulting Q stands for heat transport associated with an individual eddy (over a distance of four normalized radii).

Then the total time-mean meridional transports by eddy trapping and stirring are estimated by multiplying the average individual eddy transport by the detected population number density N_e in a certain area, where N_e is the total detected eddy number multiplied by the sampling interval of altimetry data (daily) and divided by the time length of the analysis period (Dong et al., 2014); N_e is shown in Table 3.

The resulting estimate (Table 3) of the stirring component of the meridional heat transport for individual eddies associated with (Q_s~10¹² W) is 2 orders of magnitude larger than that with trapping (Q_t~10¹⁰ W). Whereas the zonal heat transports associated with stirring and trapping have the same order of magnitude, due to the similar magnitude of eddy westward propagation speed and rotational velocity in these regions. Trapping-induced zonal and meridional heat transport in this way is similar to that estimated by Dong et al. (2014). It is denoted that stirring-induced heat transport is obviously larger in WNP-N than in WNP-S, especially for the anticyclonic eddies.

In addition to the background SST gradient intensity we have proved, there are other factors that would affect the eddy heat transport. For example, large-scale wind stress curl may inject energy into the mesoscale eddy field and promote the eddy evolution, and finally affect the eddy heat transport (Ding et al., 2018).

4 CONCLUSION

Eddy characteristics in the area associated with STCC are analyzed in the study. We confirmed that in this region, eddies tend to concentrate toward the latitudes around 21°N. Firstly, eddies in the latitudes around 21°N have larger sizes and magnitudes. In addition, more eddy termination than birth at these latitudes also confirms the eddy convergence towards 21°N from lower and higher latitudes. While the reason why the eddies tend to concentrate at around 21°N needs to be explored, we separate the study area, we further divided the region into two subareas, WNP-N and WNP-S, that lies north and south of 21°N. The two subareas are featured as different background SST fields: WNP-N has large meridional SST gradient and WNP-S has relatively uniform SST. We found that in the study region, the eddy-induced SSTa structure could be regarded as a superposition of a monopole core of cold (warm) water in the eddy interior and a dipole structure caused by horizontal advection of the background SST field. Comparison between both components shows that eddy's stirring-induced SSTa structure plays a far more important role in WNP-N than in WNP-S.

After reclassifying and recomposing the eddies according to the intensity of their background SST gradient, we find that contribution of stirring-induced SSTa increases as the intensity grows and upwelling-/downwelling-induced SSTa decreases correspondingly, which illustrates further that SSTa pattern caused by eddies relies largely on background SST gradient. Eddies are also reclassified and recompiled according to the month during which they are detected. We find that eddy-induced SSTa varies seasonally with small amplitude in summer and autumn and large amplitude in winter and spring. The seasonal variation of dipole SSTa matches the seasonal variation of the background SST gradient very well.

Dividing eddy-induced SSTa according to two mechanisms, eddy meridional heat transport in mixed-layer is estimated. The stirring component of eddy meridional heat transport is 2 orders larger than the trapping one. Besides, the stirring meridional heat transport is much larger in the area which possesses intense isotherms and larger SST gradient.

5 ACKNOWLEDGMENT

We thank Remote Sensing Systems (available at http://www.remss.com) for providing the AMSR-E SST observations, Collected Localis Satellites (https://www.aviso.altimetry.fr/) for the AVISO SSH observations, and the global dataset of observed mesoscale eddy tracks (available at http://wombat.coas.oregonstate.edu/eddies/).