Abyssal Circulation From the Yap‐Mariana Junction to the Northern Philippine Basin

The lower deep branch of the Pacific Meridional Overturning Circulation (L‐PMOC) is a crucial element of the ocean's climate and biogeochemical systems through carrying the Lower Circumpolar Water (LCPW). For the first time, the pathway and volume transport of L‐PMOC from the Yap‐Mariana Junction (YMJ) to the Northern Philippine Basin (NPB) are revealed by a six‐mooring array measurement over 2019–2021. The L‐PMOC seasonally intrudes into the western Pacific at the YMJ. Then, it is directed into the West Mariana Basin (WMB) through YMJ‐Northern Channel with 1.41 ± 1.43/0.26 (mean ± standard deviation/total rms error) Sv, and further into the NPB through Kyushu‐Palau Ridge (KPR) Channel with 0.75 ± 0.53/0.18 Sv. Their difference 0.65 ± 1.35/0.28 Sv is the net lateral flux of LCPW into the WMB over the 2.5 yr. Analyses of a data‐assimilative ocean model solution suggest that the L‐PMOC transports through the deep channels are consistent with deep pressure gradients forced mainly by upper ocean processes.

Supporting Information may be found in the online version of this article.
the Northern Channel (YMJ-Northern) connects to the WMB. The flow through these channels has been observed through deployed moorings. Siedler et al. (2004) reported the L-PMOC flowing northward into the WMB with a mean speed of 10.4 cm s −1 at 4,220 m over 1996-1998 on the western side of YMJ-Northern. Recently, based on observations with a five-mooring array, Wang et al. (2020Wang et al. ( , 2021 revealed the existence of return flow in the YMJ-Northern, and the seasonal variation of the L-PMOC volume transports through the YMJ three channels and their driving mechanisms. During November-April, the L-PMOC intrudes into the YMJ from the EMB through the YMJ-Eastern, and then flows into WMB and WCB through the YMJ-Northern and YMJ-Southern, respectively. During May-October, the westward intrusion through the YMJ-Eastern nearly vanishes, and the L-PMOC mainly flows northward through the YMJ-Southern and YMJ-Northern from the WCB to the WMB. Such seasonal intrusion of the L-PMOC is forced by the deep pressure gradient that is mainly determined by the wind forcing and water mass properties over the upper 2,000 m. Intensified deep intraseasonal variability at the YMJ was also observed and attributed to the topographic Rossby waves (Ma et al., 2019).
After passing the YMJ, the pathway and variability of the L-PMOC in the WMB and PB remain unknown, hindering a comprehensive description of the LCPW transport in the whole WPO. Here, we report the first measurement of flow through the deep choke channel at the Kyushu-Palau Ridge (KPR) connecting the Northern Philippine Basin (NPB) with the WMB (Figure 1). The simultaneous mooring measurements in the YMJ-Northern and KPR enable a quantification of the net lateral flux of LCPW into the WMB over 2019-2021. We also add 1.5 yr records of five mooring measurements at the YMJ to the 1 yr time series reported by Wang et al. (2021), leading to a more robust quantification of the variability of L-PMOC transports. The observational results are also compared and interpreted with the solutions of a data-assimilative global ocean modeling product.

Mooring Data
One mooring was deployed in the KPR (hereafter referred to as KPRM). The five-mooring array at the YMJ includes one in the YMJ-Eastern close to the Mariana Trench (MarM), one in the YMJ-Southern close to the Yap Trench (YapM), and three on the western, central, and eastern sides of the YMJ-Northern (WM, CM, and EM), respectively. The locations and durations of the above six moorings, and water depths of mooring sites are given in Table S1 in Supporting Information S1.
Each mooring was equipped with 5-7 discrete deep current meters, and 7-11 discrete conductivity-temperature-depth (CTD) or temperature (T) sensors between 2,500 m and the bottom (Table S1 in Supporting Information S1). Current meters sampled point velocity hourly. The CTDs measured temperature, salinity, and pressure every 10 min. The Ts returned records of temperature every 10 min. The measurement depths of Ts are obtained using the recorded depths of CTDs and the nominal separation depths between the Ts and adjacent CTDs. The potential temperature (θ) corresponding to a measured in situ temperature is derived using the salinity value from the CTD cast carried out during the mooring deployment. All data are low-pass filtered with a 72 hr Lanczos filter to remove the effects of inertial fluctuations and tides and are then averaged between noon times of two consecutive days to derive the daily means.
Velocity vectors are rotated to the along-and cross-channel directions (x and y axes) with positive x toward −50.94°, 90°, 23.25°, and 0° (clockwise from the north orientation) in the KPR, YMJ-Eastern, YMJ-Southern, and YMJ-Northern, respectively. The cross-channel sections along the y-axis are denoted by solid purple lines in Figure 1. In the YMJ-Northern, we assume that the observed along-channel velocities (U along ) from three moorings represent the flow across the 11.37°N section, although their latitudes are slightly different. The seasonal variations are quantified by fitting the annual and semi-annual harmonics to the daily time series.

Ocean Model Data
The global ocean modeling product PSY4V3R1, created by Mercator Ocean International, is obtained from the Copernicus Marine Service. This product is created through extensive data assimilation using a global ocean model, having a nominal horizontal resolution of 1/12° in longitude/latitude and 50 vertical levels with a thickness of 1 m at the surface and 450 m near the bottom. The model uses the ETOPO1 bathymetry data set in regions deeper than 300 m. Details on this product are documented by Lellouche et al. (2013Lellouche et al. ( , 2018. We take the daily temperature, salinity, and velocity from PSY4V3R1 during our observation period.

L-PMOC Transport Estimation
For estimating the L-PMOC transport, we grid the across-channel section with a resolution of 200/20 m in the horizontal/vertical directions, and assume zero velocity in the grid nearest the sidewall and bottom. The ETOPO1 bathymetry data are used. We first interpolate the velocity profiles vertically employing a shape-preserving cubic method (denoted as pchip; Fritsch & Carlson, 1980) among the moored sensors (model grids) and between the deepest measured (modeled) value and zero bottom velocity. Then, horizontal interpolation is carried out also using the pchip method. Integrating the velocity with θ ≤ 1.2°C yields the L-PMOC transport (denoted as V).
For observed V, we evaluate the error sources from the finite length of observation (E T ), limited moorings and different horizontal interpolation methods (E H ), and different vertical interpolation methods (E V ). The total rms error of V is obtained as E = (E T 2 + E H 2 + E V 2 ) 1/2 under the assumption that the three individual errors are independent. E T is estimated from the standard deviation and the number of degrees of freedom of the time series of V with the time-mean value and seasonal variation subtracted. E H is estimated from the difference between the "true transport" V true and V pchip using the model velocities along the cross-channel section with a resolution of 200 m and at the mooring sites only, respectively. The velocities along either the cross-channel section or at the mooring sites are obtained by applying a bilinear interpolation from the velocities on model grids. Then, both V true and V pchip are obtained by applying the pchip method vertically and horizontally. The values of E V are taken as the largest difference among the time-mean values of V derived by applying three commonly used methods for vertical interpolation on the observed data. See Text S1 in Supporting Information S1 for more details about the evaluation on E and the selection of interpolation method.

Observed Variations
According to observations, the upper boundary of the LCPW layer is deepened from the YMJ to the KPR and varies with the abyssal flow velocity. Figure 2 presents time-depth variations of observed θ and U along over depths greater than 3,000 m. We use θ = 1.2°C as the boundary to separate the LCPW and Upper Circumpolar Water (UCPW) layers (e.g., Siedler et al., 2004). The upper deep branch of the PMOC (U-PMOC) carries the UCPW  northward from the Southern Ocean, and can arrive at the WPO through a different route from the L-PMOC (Kawabe & Fujio, 2010). The North Pacific Deep Water occupies almost the same layer as the UCPW, and we hereafter refer to both water masses as UCPW for convenience. At the YMJ, the upper boundary of LCPW layer is located at 3,750 m on average in the YMJ-Eastern and YMJ-Southern. In the YMJ-Northern, the θ = 1.2°C isotherm tilts upward from 3,900 m at WM to 3,750 m at EM, consistent with the geostrophic balance for a deep northward flow (Wang et al., 2020). After crossing the YMJ, the LCPW arrives at KPRM with the upper boundary descended to 4,300 m on average. At MarM, the flow in the LCPW layer is directed westward and has a strong core speed exceeding 30 cm s −1 during November-April, and reverses to a weak eastward flow during May-October. The flow in the UCPW layer is generally weak but with the same phase as the LCPW layer. At YapM, the southward (northward) flow with a core speed of 5 cm s −1 roughly corresponds to the westward (eastward) flow at MarM. In the YMJ-Northern, the L-PMOC flows mainly northward with higher (lower) speed during November-April (May-October) at WM and CM. The southward deep flow at the LCPW layer of EM can be regarded as the return flow of the L-PMOC, as detailed in Wang et al. (2020). The deep flow at the UCPW layer of the YMJ-Northern is mainly directed southward, possibly originating from a deep channel northeast of the YMJ-Northern (Wang et al., 2020, their Figure  12). At KPRM, the deep flows are northward (southward) at depths with θ lower (larger) than 1.25°C, and the core speed of the L-PMOC can reach 15 cm s −1 .
The YMJ is the primary region for the L-PMOC intruding into the WPO. Figures 3a-3c show the time series of V through the YMJ-Eastern, YMJ-Southern, and YMJ-Northern. The observed estimates during the mooring measurement period are −1.84 ± 2.05/0.34 (mean ± standard deviation/total rms error, same hereinafter), 0.41 ± 1.18/0.24, and 1.41 ± 1.43/0.26 Sv, respectively. Among the three channels, the YMJ-Eastern is the first gateway for the L-PMOC flowing into the YMJ, with V exhibiting a strong seasonality. The fitted seasonal cycle can explain 56% of the total variance, and the phase can be divided into November-April and May-October. Taking a broad view of the three channels, we can distinguish the seasonal characteristics of the L-PMOC transports at the YMJ, consistent with the summary in Wang et al. (2021). That is, during November-April, the L-PMOC intrudes westward into the YMJ from the EMB through the YMJ-Eastern, and then bifurcates into southern and northern branches: the southern (northern) branch passing through the YMJ-Southern (YMJ-Northern) into the WCB (WMB) (Figure 1b). During May-October, the L-PMOC is directed mainly northward from the WCB through the YMJ-Southern and into the WMB through the YMJ-Northern; the L-PMOC through the YMJ-Eastern oscillates

between YMJ-SW and YMJ-NE sites in panels (a-b), between YMJ-N-S and YMJ-N-N sites in panel (c), and between KPR-E and KPR-W sites in panel (d), respectively. Energy-preserving spectra of (f) observed and (g) PSY4V3R1 modeled L-PMOC transports through YMJ-Eastern (red), YMJ-Southern (pink), YMJ-Northern (blue), and KPR (black) Channels and transport differences between YMJ-Northern and KPR Channels (green).
in eastward/westward directions (Figure 1c). The seasonality of V through the YMJ-Southern explains 32% of the total variance, while that through the YMJ-Northern only explains 18% of the total variance.
After passing through the YMJ-Northern, the LCPW can arrive at the KPR. According to Figure 3d, the LCPW is directed northwestward from the WMB into the NPB during the whole measurement period. The observed estimate of V over the 2.5 yr is 0.75 ± 0.53/0.18 Sv. The seasonal variation explains 17% of the total variance, similar to the YMJ-Northern.
The energy-preserving spectra of the observational estimates of V are computed (Figure 3f). The L-PMOC transports through the YMJ-Eastern and YMJ-Southern contain significant energy in the low-frequency band at periods longer than 100 days, consistent with their larger seasonal variations. In contrast, V through the YMJ-Northern is dominated by variations at intraseasonal scales with peak energy at periods of 10-30 days and 50-85 days. For the KPR, the intraseasonal and low-frequency bands have similar energy. The integrated spectral energy over the whole frequency range has a descending order from the YMJ-Eastern to YMJ-Southern, YMJ-Northern, and KPR.
The difference between the L-PMOC transports through the YMJ-Northern and KPR (denoted as ΔV, Figure 3e) approximately represents the net LCPW lateral flux into the WMB. The ΔV during the mooring measurement period is 0.65 ± 1.35/0.28 Sv. Its spectrum shows dominant variations at intraseasonal time scales, similar to that of V through the YMJ-Northern which has a stronger variability than V through the KPR (Figure 3f). The seasonal cycle of ΔV can only explain 10% of the total variance. The YMJ-Northern and KPR are the two main deep channels for the WMB to exchange LCPW with surrounding basins. According to the bathymetry, there are two other deep channels possibly allowing the LCPW exchange, located at the KPR to the north and south of KPRM sites (136.9°E/23.29°N and 134.9°E/12°N). However, V values through both of them are much smaller than V through the KPR according to the PSY4V3R1 data ( Figure S5 in Supporting Information S1). Nearly zero transport through the channel at 134.9°E/12°N is also consistent with the observed fact that the LCPW cannot penetrate the Southern Philippine Basin (Tian et al., 2021). Consistent with geostrophic velocity inferred from hydrographic measurement at 137°E (Tian et al., 2021, their Figure 8), the modeled U along at 136.9°E/23.29°N has a core speed of ∼3 cm s −1 , much smaller than that at KPRM. This yields nearly zero transport through this channel during the measurement period. Thus, positive (negative) values of ΔV can be regarded as the net lateral LCPW flux into (out of) the WMB, with a net mean influx during the 2.5 yr. Similarly, the time-mean transport of 0.75 Sv through the KPR can be taken as the mean net lateral flux of LCPW into the NPB.

Interpretation With Model Data
The PSY4V3R1 model data are analyzed to interpret the observed L-PMOC variations. We first evaluate the model performance by comparing the daily time series of the L-PMOC transports derived from the model and observation (Figures 3a-3d). For the YMJ-Eastern, YMJ-Southern, YMJ-Northern, and KPR, the modeled V are 0.99 ± 2.40, 0.69 ± 2.54, 0.81 ± 1.18, and 0.13 ± 0.51 Sv during the mooring measurement period, respectively. The modeled time-mean transports have some differences with the observed estimates. Notably, the mean transport through the YMJ-Eastern is eastward from the model while westward according to observations. The seasonal variations can explain 67%, 53%, 34%, and 21% of the total variance of the modeled transports through the four channels, comparable to observations. The energy-preserving spectra of the modeled V are broadly consistent with the observations in terms of the relative energy levels at seasonal and intra-seasonal time scales, but with evident differences in details (Figures 3f-3g) that will not be discussed further. The normalized root-mean-square deviations for daily transport time series (NRMSE = where n is the number of data values) are calculated. For the four channels in the same order, the NRMSEs are 31%, 30%, 22%, and 22%, respectively. Correspondingly, the observed and modeled variations of V have correlations of 0.87, 0.64, 0.38, and 0.73 at the 0.05 level, respectively. Finally, the difference of modeled V through the YMJ-Northern and KPR is ΔV = 0.68 ± 1.19 Sv (Figure 3e). Compared with the observed values of 0.65 ± 1.35 Sv, the model obtains a similar time-mean and variation magnitude. The above analyses suggest that the model solution can be used to study the dynamics of L-PMOC transport variation.
Interbasin transport through the strait can be linked to along-strait pressure difference according to several theories, as synthesized by Song (2006). Guided by these theories, we now examine the relationship between time Variations of V through deep channels and deep pressure gradients can be linked to upper ocean forcings. A detailed analysis with 26 yr of model data suggests that the differences in the sea surface height (SSH) and water density over the upper 2,000 m between YMJ-SW and YMJ-NE sites play a decisive role in forming the deep pressure gradient and variation of V through the YMJ channels   Hughes et al., 2018;Vinogradova et al., 2007). The PSY4V3R1 solution is skilled in simulating the upper ocean variations because it assimilates SSH from satellite altimeters and temperature and salinity from Argo observations (Gasparin et al., 2018). It should possess reasonable skill in simulating the variations of L-PMOC transports if they are indeed mainly forced by variations in the upper ocean. In sum, the L-PMOC transports through the deep channels correspond to the deep pressure gradients that are mainly forced by upper ocean processes.
The above sites for calculating the deep pressure gradient are chosen at the upstream and downstream sides of the deep channels according to the mean deep flows. Figure 4a shows the PSY4V3R1 mean pressure and flow vectors at 4,405 m averaged over the whole measurement period. The deep flows show evident correspondence with the pressure gradients, following the geostrophic relationship. On the other hand, the complicated flow structure makes it difficult to distinguish a clear pathway of L-PMOC from the YMJ-Northern to the KPR in the WMB. The complexity is related to the strong variability in both deep flows and pressure, evident from comparing the time-mean ( Figure 4a) and a daily (Figure 4b) fields. The daily deep flows show larger magnitudes corresponding to the larger pressure gradients, with more evident correspondence following the geostrophic relationship.

Discussions and Conclusions
Through decades of mooring measurements, the pathways of L-PMOC are depicted and the volume transports are quantified. After separating from the Antarctic, about 15.8 Sv of L-PMOC flows northward into the Pacific Ocean, according to mooring results east of New Zealand and the Tonga-Kermadec Ridge at 32°S over 1991-1992(Whitworth et al., 1999. This value is estimated for all the circumpolar water deeper than 2,000 m, but the transport over 2,000-3,250 m is minor (their Figure 17). Then the L-PMOC was observed to flow northward through the Samoan Passage with 5.4-6.0 Sv over 1992-1994and 2012-2013(Rudnick, 1997Voet et al., 2016Voet et al., , 2015. Our previous studies (Wang et al., 2020 and new extended observations here reveal that the L-PMOC can intrude into the WPO through the YMJ-Eastern with 1.84 ± 2.05/0.34 Sv. Through the YMJ-Northern, 1.41 ± 1.43/0.26 Sv of the lower deep flow enters the WMB. Furthermore, here we for the first time reveal that 0.75 ± 0.53/0.18 Sv of the lower deep flow is directed from WMB to the NPB through the KPR. The difference between the transport through the YMJ-Northern and KPR is 0.65 ± 1.35/0.28 Sv. Due to geography constriction, 0.75 and 0.65 Sv are the net lateral LCPW flux into the WMB and NPB averaged over the 2.5 yr. Finally, we note that the mooring measurements and L-PMOC transport estimation here are largely exploratory, leaving future work to validate the transport uncertainties. It is hoped that our study can stimulate the planning for long-term observations to reveal the variations of deep transports at interannual and longer time scales, and to derive the long-term mean rate of LCPW diapycnal upwelling in combination with deep water mass observations in these basins. Furthermore, the relevance of these estimated deep-water transports to ocean climate change and biogeochemical cycling processes deserves attention.