The Argo program is a joint international effort involving more than 30 countries and organizations and having deployed over 15 000 freely drifting floats since 2000. The accumulated total of collected profiles exceeds 2 million profiles of conductivity, temperature, and depth (CTD) along with other geobiochemical parameters. The Argo program has become the main data source for many research and operational predictions of oceanography and atmospheric science (http://www.ARGO.ucsd.edu, last access: 8 May 2021). We screened the profiles from the Philippines Sea with quality control and estimated diapycnal diffusivity and dissipation rates from them using a fine-scale parameterization.

The diapycnal diffusivity and turbulent kinetic energy dissipation rate can be estimated from a fine-scale strain structure. This is based on a hypothesis that the energy can be transported from large to small scales. In such scales, waves break due to shear or convective instabilities by weakly nonlinear interactions between internal waves (Kunze et al., 2006). Presently, this method has been widely used for the global ocean (e.g., Wu et al., 2011; Kunze, 2017; Whalen et al., 2012; Fer et al., 2010; Waterhouse et al., 2014). The dissipation rate ε can be expressed as follows: 1ε=ε0N2‾N02〈ξz2〉2〈ξz2GM〉2hRωLf,N‾, where ε0=6.73×10-10 W kg-1, N0=5.24×10-3 s-1, and N2‾ represents the averaged buoyancy frequency of the segment. 〈ξz2GM〉 and 〈ξz2〉 are strain variance from the Garrett–Munk (GM) spectrum (Gregg and Kunze, 1991) and the observed strain variance, respectively. The angle brackets indicate integration over a specified range of vertical internal wavenumbers (see Eqs. 4 and 5). The function hRω accounts for the frequency content of the internal wave field, and Rω represents shear and strain variance ratio. Rω is fixed at 7, which is a global mean value (Kunze et al., 2006). 2hRω=162RωRω+1Rω-1. The function Lf,N‾ corrects for a latitudinal dependence; here, f is the local Coriolis frequency, so f30 is the Coriolis frequency at 30∘, and N‾ is the vertically averaged buoyancy frequency of the segment. 3Lf,N‾=farccosh(N‾f)f30arccosh(N‾f30). Strain ξz was calculated from each segment as follows: 4ξz=N2-Nref2N2‾5〈ξz2〉=∫kminkmaxSstrkzdkz≤0.2. We derived N from 2 to 10 dbar processed temperature, salinity, and pressure data according to the Argo float resolution. Nref, as a smooth, piece-wise quadratic fit to the observed N profile, is fitted to 24 m. Here we remove segments that vary in the range of 〈N2〉>5×10-4 s-2 or 〈N2〉<1×10-9 s-2, since the strain signal at these levels is dominated by noise (Whalen et al., 2018). By applying a fast Fourier transform (FFT) on half-overlapping 256 m segments along each vertical ξz profile, we computed the spectra Sstrkz and integrated them to determine the strain variance. We integrated these spectra between the vertical wavenumbers kmin=0.003 cpm and kmax=0.02 cpm, according to typical global internal tidal scales and Eq. (5), respectively. Substituting 〈ξz2〉 into Eq. (1) ultimately yields 32 m resolved vertical sections of each observed profile. The dissipation rate ε is related to the diapycnal diffusivity Kz by the Osborn relation as follows: 6Kz=ΓεN2‾, where the flux coefficient Γ is fixed at 0.2 generally.

The near-inertial energy flux for each observation profile was calculated using the 10 m wind speed product from ERA-Interim (https://www.ecmwf.int/en/forecasts/datasets, last access: 8 May 2021), which is a 6 h wind speed on a grid of 0.75∘ × 0.75∘. We selected the mean near-inertial flux of 30–50 d before the time of each diapycnal diffusivity estimation as our measure of the near-inertial flux, with the consideration of the propagation of NIWs.

The wind-driven NIW energy flux can be directly estimated using a slab model, which assumes that the inertial oscillations in the mixed layer do not interact with the background fields. The mixed layer current velocity can be described by the following: 7dZdt+r+ifZ=TρH, where Z=u+iv is the mixed layer oscillating component of full current, and i is an imaginary number to indicate the latitudinal component. T=τx+iτy is the wind stress on the sea surface, f is the local Coriolis parameter, and r is the frequency-dependent damping parameter, which was fixed at 0.15f for these calculations. ρ is sea water density and fixed at 1024 kg m-3. H is the mixed-layer depth and was set to a constant 25 m. We can calculate the oscillating component of the full velocity from Eq. (7) and obtain the near-inertial component through a bandpass filter of [0.85,1.25]f. The near-inertial energy flux is calculated as follows: 8EΠ=ReZ⋅T∗. The asterisk (*) indicates the complex conjugate of a variable.

The eddy kinetic energy (EKE) is estimated based on geostrophic calculation as follows: 9EKE=12Ug′2+Vg′210Ug′=-gfΔη′ΔyVg′=-gfΔη′Δx, where Ug′ and Vg′ are the geostrophic velocities in the east–west and north–south directions, respectively. They are taken from the Aviso+ (http://www.aviso.altimetry.fr/duacs/, last access: 8 May 2021) geostrophic velocity product. η′ indicates the sea level anomaly (SLA).

The internal tidal conversion rate was provided by SEANOE (SEA scieNtific Open data Edition; https://www.seanoe.org/recordview, last access: 8 May 2021, Falahat et al., 2018), including eight main tidal constituents. We used the mode-summed internal tidal conversion rates of M2 and K1, and integrated eight main tidal constituents in the present study.

The diapycnal diffusivities were used as indicators of ocean diapycnal mixing. The pattern averaged within 250–500 m is shown in Fig. 1a. The Kz was estimated from the Argo profiles, with an average on each cell of 0.5∘ × 0.5∘. The magnitude of diapycnal diffusivities increased with latitude, reaching 10-4 m2 s-1 in the northern part of this area (30–36∘ N). The mean value of Kz was about O(-6)–O(-5) at lower latitudes, while it was remarkable that the magnitude of Kz also increased significantly in some low-latitude regions, reaching O(-4) or higher, such as in the Luzon Strait (Xu et al., 2014). Reviewing the influence of topography, wind, and internal tides (Fig. 1b–d) on ocean mixing, it was found that the latitudinal variability in Kz was consistent with the wind intensity distribution. Upper ocean mixing was significantly enhanced at midlatitudes due to the presence of westerlies. In addition, Kz was also enhanced near several key internal tide sources, such as the Luzon Strait and Bonin, Izu, and Dadong ridges, etc. At these sites, the magnitude of Kz was obviously larger than other areas at the same latitude, indicating a significant role of internal tides. Additionally, the enhancement of deep ocean mixing at these sites was even more obvious (not shown).

It can be noted that the pattern of diapycnal diffusivities was not completely consistent with those of either internal tides or winds. This suggests that the ocean mixing was modulated by factors other than tides and winds. The magnitudes of Kz also vary for internal tide source sites. Considering that the Philippine Sea is a region with energetic mesoscale motions (Fig. 2), the influences of mesoscale features in turbulent mixing should be taken into account. The existence of mesoscale features can alter the propagation and dissipation of internal tides. Therefore, the Philippine Sea is an ideal region for studying the modulation of background flows on turbulent mixing associated with strong internal tides.

The seasonal cycle for diapycnal diffusivities also differs zonally. Here, we divided the Philippine Sea into two portions, i.e., low latitude (10–25∘ N) and midlatitude (25–35∘ N). The diapycnal diffusivities Kz were averaged in each latitude band (Fig. 3). At the depth of 250–500 m in the midlatitude, the diapycnal diffusivities had a significant seasonal trend that was strong in winter and weak in summer. This is consistent with the seasonal fluctuation of near-inertial energy from wind. Such a seasonal cycle could also be found at 500–1000 and 1000–1500 m in the midlatitudes, but it was relatively weaker, especially after 2016. In the lower latitudes, the NIW energy was still strong in winter and weak in summer, but a seasonal dependence of turbulent mixing was not obvious, even in the upper ocean. Consequently, the wind was found to play a significant role in driving turbulent mixing at midlatitudes but was insignificant at low latitudes. Other factors drove and modulated turbulent mixing in low latitudes.

Focusing on the low latitudes where tidal mixing dominated, the diapycnal diffusivities, Kz related to internal tides, and EKE are shown (Fig. 8). The combined influences of mesoscale features and internal tides on mixing are indicated. The increasing internal tide conversion rates significantly enhanced turbulent mixing. We find a correlation between elevated EKE and the averaged diapycnal diffusivities for a given internal tide conversion rate level. When the conversion rate was 10-3 W m-2, the magnitudes of Kz were about 3 × 10-6 m2 s-1, 3 × 10-6 m2 s-1, and 1 × 10-6 m2 s-1 at depths of 250–500, 500–1000, and 1000–1500 m, respectively. When the internal tide conversion rates reached O(-1)–O(0), Kz reached 10-5 m2 s-1 at both depths of 250–500 and 500–1000 m and even exceeded 10-4 m2 s-1 at some internal tide source sites. In addition, there was a positive correlation between EKE and diapycnal diffusivity. A higher EKE can further increase Kz under the same magnitude of the internal tide conversion rate. Such an enhancement was more significant with strong internal tide conversion rates greater than 10-3 W m-2.

M2 and K1 tidal constituents were analyzed to clarify the response of Kz to internal tides in the regions of high EKE (where the EKE is larger than the regional average value) and low EKE (Fig. 9). The results integrated eight main tidal constituents (Fig. 9a, b, and c) and showed that the slopes in a weak (strong) mesoscale field were smaller (larger), i.e., 0.081 (0.105), 0.103 (0.134), and 0.103 (0.142), at depths of 250–500, 500–1000, and 1000–1500 m, respectively. The turbulent mixing was more sensitive to the internal tide magnitude in the presence of an energetic mesoscale field. Moreover, such a response was more obvious in the region with strong internal tides (such as the >10-2 W m-2 conversion rate). In some regions with weak internal tides, such as those with internal tide conversion rates less than 10-3 W m-2, the modulation of mesoscale eddies was less significant.

A similar conclusion can be drawn when only considering M2 or K1. In regions of high EKE, the change in diffusivities in response to internal tides was significant, and the increase was more sensitive to the M2 internal tide. The enhancement related to the M2 internal tide was more significant below 500 m (Fig. 9d and e), while the enhancement of the K1 internal tide was similar at all depths. This may be due to different features and structures of M2 and K1 internal tides. In this area, the modal structure and propagation path of the M2 internal tide are more complicated and more prone to breaking, but those of K1 were relatively stable, and this area includes the K1 critical latitude range, which can be broadened by mesoscale currents (Robertson and Dong, 2019).

The modulation of cyclonic and anticyclonic eddies on tidal mixing also differs. The increase in Kz by internal tides in regions with cyclonic eddies (vorticity >3×10-6 s-1) and anticyclonic eddies (vorticity <-3×10-6 s-1) are both shown (Figs. 10 and 11). Under the same magnitude of internal tides, the Kz increases more significantly in the presence of anticyclonic eddies, which is obvious at 250–500 m, and can also be seen at 500–1000 m. Below 1000 m, there are no significant differences between the regions with cyclones and anticyclones.

Considering that mixing driven by eddies is relatively significant in regions where the tidal mixing is very weak, we only analyze the cases of internal tide conversion rates larger than 10-3 W m-2. When the conversion rates become larger than this value, the diapycnal diffusivities in the presence of high EKE increase faster with internal tides (Fig. 9). It was found that the response of turbulent mixing to internal tides was more sensitive in the presence of anticyclones above 1000 m, while, below 1000 m, the influence of cyclones is slightly stronger than that of anticyclones.