Decadal variations of Pacific North Equatorial Current bifurcation from multiple ocean products

Authors


Abstract

In this study, we examine the decadal variations of the Pacific North Equatorial Current (NEC) bifurcation latitude (NBL) averaged over upper 100 m and underlying dynamics over the past six decades using 11 ocean products, including seven kinds of ocean reanalyzes based on ocean data assimilation systems, two kinds of numerical simulations without assimilating observations and two kinds of objective analyzes based on in situ observations only. During the period of 1954–2007, the multiproduct mean of decadal NBL anomalies shows maxima around 1965/1966, 1980/1981, 1995/1996, and 2003/2004, and minima around 1958, 1971/1972, 1986/1987, and 2000/2001, respectively. The NBL decadal variations are related to the first Empirical Orthogonal Function mode of decadal anomalies of sea surface height (SSH) in the northwestern tropical Pacific Ocean, which shows spatially coherent variation over the whole region and explains most of the total variance. Further regression and composite analyzes indicate that northerly/southerly NBL corresponds to negative/positive SSH anomalies and cyclonic/anticyclonic gyre anomalies in the northwestern tropical Pacific Ocean. These decadal circulation variations and thus the decadal NBL variations are governed mostly by the first two vertical modes and attribute the most to the first baroclinic mode. The NBL decadal variation is highly positively correlated with the tropical Pacific decadal variability (TPDV) around the zero time lag. With a lead of about half the decadal cycle the NBL displays closer but negative relationship to TPDV in four ocean products, possibly manifesting the dynamical role of the circulation in the northwestern tropical Pacific in the phase-shifting of TPDV.

1. Introduction

The North Equatorial Current (NEC) in the Pacific Ocean flows westward and bifurcates into the northward flowing Kuroshio Current (KC) and the southward flowing Mindanao Current (MC) off the Philippine coast (Figure 1) [e.g., Nitani, 1972; Qu et al., 1998]. Observations have proven that the westward NEC transport equals to the sum of the southward MC transport and northward KC transport, inducing a mass balance in regions around the bifurcation of the NEC [e.g., Qu et al., 1998]. Thus, the NEC bifurcation latitude (NBL) is an important indicator of the partition of the NEC mass, heat and salt between the MC and KC, which flow into the tropical gyre and subtropical gyre, respectively. This partition is very important in determining the meridional water property exchanges and thus regulating the warm pool [e.g., Qu et al., 1997], regional biological processes [e.g., Kimura et al., 2001; Amedo et al., 2002], and so on.

Figure 1.

Annual mean horizontal currents (m s−1) averaged within the upper 100 m in the northwestern tropical Pacific Ocean off the Philippine coast (120°E–140°E, 5°N–20°N) from the 11 ocean products and altimetry observations.

In the literature, the NBL and its variability have been extensively studied based on observations [e.g., Toole et al., 1990; Qu and Lukas, 2003; Yaremchuk and Qu, 2004; Wang and Hu, 2006; Qiu and Chen, 2010; Zhai and Hu, 2012; Zhao et al., 2012], high-resolution numerical model simulations [e.g., Qiu and Lukas, 1996; Kim et al., 2004; Chen and Wu, 2011] and also data assimilation [e.g., Chen and Wu, 2012]. For the long-term mean, the NBL occurs at the zero zonally integrated wind stress curl line, consistent with the Sverdrup theory [e.g., Qu and Lukas, 2003]. On the seasonal time scale, the NBL is revealed to reach its northernmost position in winter and its southernmost position in summer by analyzes based on historic hydrographic observations [e.g., Qu and Lukas, 2003], satellite altimeter measurements [e.g., Wang and Hu, 2006; Qiu and Chen, 2010] and numerical model simulations [e.g., Kim et al., 2004; Chen and Wu, 2011]. Accompanying the northward/southward migration of the NBL, the KC (MC) transport tends to decrease/increase (increase/decrease) [e.g., Qiu and Lukas, 1996; Yaremchuk and Qu, 2004].

On the interannual time scale, the NBL variation is mainly related to El Niño-Southern Oscillation (ENSO) events [e.g., Qiu and Lukas, 1996; Wang and Hu, 2006; Qiu and Chen, 2010; Zhai and Hu, 2012]. The NBL generally occurs at northerly latitude in El Niño years while at southerly latitude in La Niña years. In general, the NBL interannual variation is highly positively correlated with the NEC transport [e.g., Qiu and Lukas, 1996; Kim et al., 2004; Zhai and Hu, 2012]. As for its relations to the KC and MC transports, Kim et al. [2004] pointed out that it was moderately positively (r = 0.54) correlated with the MC transport but highly negatively (r = −0.83) correlated with the KC transport. Their results are roughly consistent with the analyzes of Zhai and Hu [2012, 2013]. The latter indicate that in El Niño/La Niña years, a cyclonic/anticyclonic gyre anomaly forms in the northwestern tropical Pacific Ocean and tends to push the NBL northward/southward and increase/decrease the NEC and MC transports when decreasing/increasing the KC transport just off the Philippine coast.

More recently, the linear trend of the NBL on multidecadal time scale has been studied by Qiu and Chen [2012] using altimetry observations for the period of 1993–2009 and by Chen and Wu [2012] with outputs from the Simple Ocean Data Assimilation (SODA) version 2.2.4 for the period of 1950–2008. Their results indicated that the NBL has experienced different long-term trends over different periods in the last six decades. From 1950 to 2008, Chen and Wu [2012] indicated that the NBL averaged within the upper 381 m has migrated southward by about 1.6°N. This southward migration is slightly smaller than that during 1993–2009, which is about 2°N as reported by Qiu and Chen [2012].

Although many studies of the NBL have already been achieved, its variability on decadal time scale longer than 8 years has been hampered by a dearth of long-term observations. Up to now, however, the availability of multiple operational ocean analyzes constrained by limited in situ observations and high-resolution numerical simulations produced around the world is an opportunity for us to examine the variability of the NBL on decadal time scale. In this study, we would use results from a group of models with or without observations assimilated and two objective analyzes obtained from in situ hydrographic observations only to study the decadal variations of the NBL and the underlying dynamics. Here we do not attempt to quantify errors in each product by validating against in situ observations, which are not available for us. Instead, we mainly focus on the similarities of the NBL decadal variations and associated dynamical mechanisms among the ocean products. The errors in the NBL decadal variations may be roughly estimated from the differences between different ocean products. Accuracy of the estimated NBL decadal variations based on numerical model simulations, data assimilation systems and objective analyzes can be affected by various factors, such as uncertainties in sea surface forcing fluxes, biases in ocean models, deficiencies in data assimilation schemes and objective analysis methods, changes in the ocean observing system, and so on. The adoption of multiple ocean products is superior to that of any individual ocean product [e.g., Zhu et al., 2012] and would provide us a multiproduct mean overview of the NBL decadal variations. In this case, the errors may also be considerably reduced.

The rest of the paper is organized as follows. Section 'Ocean Products, Comparison With Observations, and Methods' describes the numerical simulations, ocean reanalyzes and objective analyzes, and their validations with altimeter measurements and tidal gauge measurements. Section 'Results' gives the results, which are summarized and discussed in section 'Summary and Discussions'.

2. Ocean Products, Comparison With Observations, and Methods

In order to get a multiproduct mean overview of the decadal variation of the NBL, we utilize seven kinds of ocean reanalyzes, two kinds of numerical simulations without observations assimilated and two kinds of objective analyzes in total. The seven kinds of ocean reanalyzes are the ensemble coupled data assimilation (ECDA) system version 3.1 [e.g., Chang et al., 2011, 2012] developed by the Geophysical Fluid Dynamics Laboratory, the German partner of the consortium for Estimating the Circulation and Climate of the Ocean (GECCO) [e.g., Köhl and Stammer, 2008], the Global Ocean Data Assimilation System (GODAS) [e.g., Behringer et al., 1998; Behringer and Xue, 2004] produced at the National Centers for Environmental Prediction (NCEP), the European Center for Medium-Range Weather Forecasts (ECMWF) ocean analysis/reanalysis system 3 (ORAS3) [Balmaseda et al., 2008] and system 4 (ORAS4) [Balmaseda et al., 2012], and the Simple Ocean Data Assimilation version 2.2.4 (SODA224) and version 2.2.6 (SODA226) [e.g., Carton and Giese, 2008], respectively. The two kinds of numerical simulations are based on the Ocean General Circulation Model (OGCM) for the Earth Simulator (OFES) and the Max-Planck Institute Ocean Model (MPIOM) within the German consortium project STORM, respectively. For more details about the configurations and evaluations of the models adopted in the current study, one can refer to Masumoto et al. [2004] and Sasaki et al. [2004] for the OFES simulation and to von Storch et al. [2012] for the STORM simulation.

The two kinds of ocean objective analyzes are the newly developed Decadal Climate Prediction System (DePreSys) [Smith and Murphy, 2007; Smith et al., 2007] and the newest version of the Enhanced Ocean Data Assimilation and Climate Prediction (ENACT) archive version 3 (EN3v2a) [Ingleby and Huddleston, 2007; Wijffels et al., 2008; Guinehut et al., 2009], respectively. Details about the time period, spatial domain, horizontal resolution and number of vertical layers within the upper 2000 m of the ocean products utilized in our study are summarized in Table 1.

Table 1. Time Period, Spatial Domain, Horizontal Resolution, and Number of Vertical Layers Above 2000 m of the 11 Ocean Products
ProductPeriodDomainHorizontal ResolutionNumber of Vertical Layers Above 2000 m
DePreSys1950–2006120°E–290°E, 0–30°N1.25° × 1.25°14
ECDA1961–2012120°E–290°E, 0–30°N0.5° × 0.5°39
EN3v2a1950–2012120°E–290°E, 0–30°N1° × 1°30
GECCO1952–2001120°E–290°E, 0–30°N1° × 1°16
GODAS1980–2012120°E–290°E, 0–30°N1° × 1/3°34
OFES1950–2004120°E–270°E, 0–30°N0.1° × 0.1°40
ORAS31959–2009120°E–290°E, 0–30°N1° × 1°25
ORAS41958–2011120°E–290°E, 0–30°N1° × 1°30
SODA2241950–2010120°E–290°E, 0–30°N0.5° × 0.5°26
SODA2261950–2008120°E–290°E, 0–30°N0.5° × 0.5°26
STORM1981–2010120°E–280°E, 0–30°N0.2° × 0.2°56

The monthly gridded temperature and salinity in the two kinds of ocean objective analyzes are used to calculate the sea surface dynamic height and geostrophic velocity. The reference level was chosen to be 1500 db following Qu et al. [1998]. When calculating the geostrophic velocity, we have used the slope of a linear fit over every three horizontal grids, centered at the middle grid, to estimate the horizontal gradients of the depth-integrated geopotential anomaly. The three-grid segment was chosen as a suitable trade-off between the need for high horizontal resolution and statistically reasonable regression estimation. Sea surface dynamic height is also calculated for ECDA using its temperature and salinity according to the same method as we could not obtain the sea surface height (SSH) from the data source http://www.gfdl.noaa.gov/ocean-data-assimilation-model-output. The calculated sea surface dynamic height is then used as SSH for further evaluation and analysis.

Before using these products to examine the NBL decadal variations, we should assess their performance in the tropical North Pacific Ocean through comparing their SSH with satellite altimeter measurements and tidal gauge observations. Of the 11 products, only several kinds have been evaluated using observations in the tropical North Pacific Ocean [e.g., Sasaki et al., 2004; Chen and Wu, 2012; Zhai and Hu, 2013; Zhai et al., 2013]. Figure 1 shows the annual mean horizontal current averaged within the upper 100 m and over the whole time period of each data set in the northwestern tropical Pacific Ocean off the Philippine coast (120°E–140°E, 5°N–20°N). For comparison, we also show the altimetry observed mean horizontal currents averaged over October 1992–December 2012 in Figure 1m. The altimetry observations are from Archiving, Validation and Interpretation of Satellite Oceanographic Data (AVISO) [Ducet et al., 2000] and obtained through combining Topex/Poseidon, Jason, ERS-1 and 2 and ENVISAT in a 1/3° × 1/3° Mercator grid and 1 week time interval since October 1992. In the current study, the altimeter products are downloaded from http://www.aviso.oceanobs.com/en/. They include the gridded sea surface heights above geoid and accordingly calculated absolute geostrophic velocities. The sea surface heights are the sum of altimetry observed sea level anomalies and mean sea surface height above geoid from CNES-CLS09_v1.1 [Rio et al., 2011]. The data set is averaged to form the monthly data set.

As can be seen from the figure, the annual mean horizontal currents from the 11 products agree reasonably well with the altimetry observations in both the spatial pattern and the velocity magnitude. We have also calculated the linear correlations between the altimeter observed SSH and the SSH from the 11 products (figure not shown). Results show that the correlations are generally high (r > 0.7 above the 95% confidence level) in the equatorward region of the tropical North Pacific Ocean and even higher in the northwestern tropical Pacific Ocean.

Figure 2 gives the linear correlations between the SSH from the 11 products and the sea level observed at six tidal gauge stations in the tropical North Pacific Ocean. We should note that the tidal gauge observations are independent of the 11 products and thus can be used for validation. Overall, all the products give high correlations above the 95% confidence level with sea level observations at all the six tidal gauge stations. Of them, the seven kinds of ocean reanalyzes constrained by various in situ observations and the two kinds of ocean objective analyzes give relatively higher correlations than the two kinds of numerical simulations as one can expect. The above comparisons indicate that the products used here generally capture well the SSH variations, thus the upper layer circulation variations in the tropical North Pacific Ocean.

Figure 2.

Simultaneously linear correlations between the SSH from the 11 ocean products and the sea level observed at six tidal gauge stations in the tropical North Pacific Ocean: (a) Guam (13.43°N, 144.65°E), (b) Legaspi (13.15°N, 123.75°E), (c) Malakal (7.33°N, 134.46°E), (d) Saipan (15.23°N, 145.74°E), (e) Truk (7.45°N, 151.85°E), and (f) Yap (9.51°N, 138.13°E). The time period of the sea level at each tidal gauge station used for the correlation calculation is indicated in each plot.

In the following, to focus on the variations on decadal time scale, monthly variables are first filtered by a low-pass filter (>8 years). Then the first and last four years of the low-pass filtered variables are excluded. At last, the linear trend over the whole available time period is subtracted from the low-pass filtered time series to get decadal anomalies for each variable.

Following Xue et al. [2012], in comparison of the decadal variations of the NBL and transports of the NEC, MC and KC across ocean products, the spread is calculated. For the temporally varying variables, the spread in each month among ocean products is defined as the standard deviation of the departure of ocean products from the multiproduct mean in that month. At the same time, signal to noise ratios (S/N ratio hereafter) of the decadal anomalies of the NBL and current transports are also calculated. The signal is the temporal standard deviation of the multiproduct mean of the decadal anomalies of the NBL and current transports over a time period, while the noise is the average of the spread among ocean products over the same time period. According to Xue et al. [2012], the S/N ratio being larger than 1 implies that the signal in the quantity analyzed is larger than the noise, and more confidence can be then placed in the multiproduct mean as an estimate of the climate signal. Here, the multiproduct mean and spread across ocean products in each month of the decadal anomalies of the NBL and transports of the NEC, MC, and KC are calculated only when there are at least five products available for that month. The S/N ratio is calculated over every 20 years, centered in the month. Within each 20 years, the S/N ratio will not be calculated as long as there is 1 month when the multiproduct means of variables are not available.

3. Results

3.1. Overall Characteristics

As the first step, we would only focus on the decadal variations of the NBL and circulation averaged within the upper 100 m in our current study. Following previous studies [e.g., Kim et al., 2004; Chen and Wu, 2012], the NBL is defined as the latitude where the meridional velocity averaged in a 2°-longitude band off the Philippine coast is zero. In ORAS3, as the simulated western boundary currents (Figure 1h), the KC and MC, cover much wider regions than in other ocean products and altimetry observations, the NBL is therefore defined as the latitude where the meridional velocity averaged in a 3°-longitude band off the Philippine coast is zero.

Figure 3a shows the NBL decadal anomalies in the 11 products. The standard deviations of the NBL decadal anomalies in the 11 products are between 0.2°N and 0.4°N, smallest in DePreSys and largest in the STORM simulation. The thick black line in Figure 3b shows the multiproduct mean of the decadal NBL anomalies. From 1954 to 2007, the multiproduct mean of the decadal NBL anomalies shows maxima around 1965/1966, 1980/1981, 1995/1996, and 2003/2004, and minima around 1958, 1971/1972, 1986/1987, and 2000/2001, respectively. The decadally northerly NBL in 1995/1996 and 2003/2004 and southerly NBL in 2000/2001 are generally consistent with those derived from altimetry observations [Qiu and Chen, 2010], which are 1997/1998 and 2003/2004 for northerly NBL and 1999/2000 for southerly NBL, respectively. The decadal NBL extremes after 1975 correlate quite well with those of the NEC decadal anomalies across 137°E [Zhai et al., 2013, Figure 4a] at 0–2 years lead or lag. After 1970, the magnitudes of the extremes of the multiproduct mean of the NBL decadal anomalies are generally larger than the spread, except those around 1980/1981 and 2000/2001, which are comparable with the latter. The S/N ratio generally increases with time and is <1 before the summer of 1982 and >1 thereafter. It implies that decadal variations of the NBL after the summer of 1982 obtained here can be placed more confidence as an estimate of true signals.

Figure 3.

Decadal anomalies of NBL and transports of NEC, MC, and KC in the ocean products and corresponding multiproduct means. (a) Decadal anomalies of NBL in the ocean products. (b) Multiproduct mean (black line) and spread (red line) of the NBL decadal anomalies and the S/N ratio (green line). (c) and (d) same as (a) and (b) but for the NEC transport decadal anomalies. (e) and (f) same as (a) and (b) but for the MC transport decadal anomalies. (g) and (h) same as (a) and (b) but for the KC transport decadal anomalies. In Figures 3a, 3c, 3e, and 3g, the solid red line presents results from DePreSys, the solid green line from ECDA, the solid blue line from EN3v2a, the solid black line from GECCO, the solid gray line from GODAS, the dashed red line from OFES, the dashed green line from ORAS3, the dashed blue line from ORAS4, the dashed black line from SODA224, the dashed gray line from SODA226, and the dashed magenta line from STORM. In Figures 3b, 3d, 3f, and 3h, the blue line is the spread times −1. All labels of the horizontal axes are at the beginning of the years.

Figure 4.

Decadal variations of SSH in the northwestern tropical Pacific Ocean in the ocean products. First column: Standard deviation of decadal SSH anomalies (m) in the tropical North Pacific Ocean. Second column: Spatial pattern of the EOF mode 1 of decadal SSH anomalies (m) in the northwestern tropical Pacific Ocean of 122°E–140°E, 7°N–20°N. Third column: Same as the second column but for the EOF mode 2. Forth column: PCs of the first (red line) and second (blue line) EOF modes. Numbers in the plot titles of the second and third columns denote the percent of the total variance explained by the corresponding EOF modes. In plots of the second and third columns, the white lines indicate the zero contours. For easy viewing and comparison, values of EOF mode 1 are divided by 1.2 for ORAS3, by 1.5 for ECDA, EN3v2a, GODAS, and OFES, by 1.7 for ORAS4 and STORM, and by 2 for SODA224 and SODA226.

Then we examine the relationship between the NBL decadal variations and the decadal variations of the westward NEC transport across 130°E between 8°N and 18°N, the southward MC transport across 8°N between the Mindanao coast and 130°E and the northward KC transport across 18°N between the Luzon coast and 130°E [e.g., Qu et al., 1998]. Here the transports of the three currents are integrated over the upper 1000 m. The MC and KC transports are not calculated with the two ocean objective analyzes as the derived geostrophic velocity field from them is unable to resolve the narrow western boundary currents (Figures 1a and 1c). The decadal anomalies and their multiproduct means and spreads of the transports of the three currents are shown in Figures 3c–3h.

During the studied period of 1954–2007, the multiproduct mean of the NEC transport anomalies basically shows maxima around 1957, 1965/1966, 1980/1981, 1994/1995, and 2004/2005, and minima around 1962, 1969/1970, 1989/1990, and 1999/2000. The multiproduct mean decadal variations of the NEC transport after 1975 obtained here are consistent with those across 137°E as indicated by Zhai et al. [2013, Figure 4a]. The linear correlation coefficient between time series of the multiproduct mean of NBL and NEC transport decadal anomalies reaches its maximum of about 0.18 when the NBL leads NEC transport by about 6 months. Though with low correlation coefficient, we note that after about 1965 the occurring periods of the NEC transport extremes agree well with those of the NBL extremes at 0–2 years lead or lag. As indicated in the following section, both decadal variations of NBL and NEC transport result from the circulation decadal variations in the northwestern tropical Pacific Ocean. The 0–2 years lead or lag between decadal variations of the NBL and NEC transport is reasonable considering the slowly evolution of the circulation in the tropical northwestern Pacific Ocean on the decadal time scale. The decadal evolution of the NEC transport depends on that of the zonal velocity within a wide latitude band (8°N–18°N) and depth range (0–1000 m), while the decadal evolution of the NBL is only determined by that of the meridional velocity off the Philippine coast. These could result in slight phase difference in decadal variations of the NBL and NEC transport. Meanwhile, the S/N ratio is generally smaller than 1, indicating significant divergence among the ocean products, which might induce large uncertainties in NEC transport decadal anomalies and thus also contribute to the phase difference. The spreads of the decadal anomalies of the NEC transport across ocean products are generally larger than magnitudes of the transport anomaly extremes except those around 1980/1981, 2004/2005, 1989/1990, and 1999/2000.

The decadal transport variations of the MC (Figures 3e and 3f) and KC (Figures 3g and 3h) differ much more significantly between different ocean products than that of the NEC, resulting in near-zero multiproduct mean values and much smaller than 1 S/N ratio in most time of the studied period. Maxima of multiproduct mean of MC transport anomalies (Figure 3f) occur around 1965/1966, 1979/1980, 1994/1995, and 2003/2004, while minima around 1969/1970, 1985/1986, and 1998/1999. As for the multiproduct mean of KC transport anomalies (Figure 3h), maxima appear around 1972/1973, 1984/1985, 1995/1996, and 2005/2006, while minima around 1980/1981, 1989/1990, and 2001/2002. The linear correlation coefficient between time series of the multiproduct mean of NBL and MC/KC transport decadal anomalies reaches its maximum/minimum of about 0.36/−0.65 above the 95% confidence level when the NBL lags MC/KC transport by about 1 or 2 years. Though with different correlation coefficients, the multiproduct mean of the MC transport anomalies agrees better with the NBL than that of the KC transport anomalies in the occurring periods of their extremes.

In this part, the relationship between decadal variations of the NBL and volume transports of NEC-MC-KC is discussed using the multiproduct mean of decadal anomalies. However, as the S/N ratios of NBL anomalies (Figure 3b) before the summer of 1982 and transport anomalies (Figures 3d, 3f, and 3h) are all smaller than 1, the relationship discussed here might be with large uncertainties and should be further improved in the future.

3.2. Circulation Variations

It is instructive to further examine the relationship between the NBL decadal variations and the broader-scale horizontal circulation variations in all the 11 ocean products. To do that, we would first explore the decadal variations of the upper ocean circulation in the tropical North Pacific Ocean. Plots in the first column of Figure 4 give the standard deviations of the decadal SSH anomalies in the 11 products. It is interesting to note that all the ocean products reveal similar localized high standard deviations of the SSH in the northwestern tropical Pacific Ocean. It indicates that on the decadal time scale the circulation in this region fluctuates much more significantly than around. To reveal the main modes in the decadal circulation variations, we apply an Empirical Orthogonal Function (EOF) analysis to the decadal SSH anomalies in the region of 122°E–140°E, 7°N–20°N, roughly consistent with the high SSH standard deviation region. Plots in the second and third columns of Figure 4 give the spatial patterns of the first and second EOF modes of the decadal SSH anomalies in the 11 ocean products, respectively. The corresponding principal components (PCs) of the two EOF modes are presented in the fourth column of Figure 4. The PCs have been rescaled according to the spatial patterns of the EOF modes.

The first EOF modes in all the ocean products are characterized by spatially coherently varying SSH over the whole region with the variation center locating south of the mean NBL and explain most of the total variance (>65%). Different from EOF mode 1, EOF mode 2 displays a north-south dipole structure in all the ocean products. In the following, we would show that it is the EOF mode 1 that correlates the most with the decadal NBL variations and thus the exact reasons for the discrepancy in EOF mode 2 between different ocean products are not discussed in current study.

Then we calculate the lead-lag correlations between the decadal NBL time series and the PCs of the first two EOF modes of the decadal SSH anomalies in each ocean product. The results for the 11 ocean products are shown in Figure 5. The most noticeable feature is that all ocean products reveal highly positive correlations between the NBL decadal anomalies and the EOF mode 1 of the SSH decadal anomalies. The highest correlation occurs at about 1 year lagged by the NBL in DePreSys but at zero time lag in other ocean products. As for the EOF mode 2, its correlation with the NBL decadal time series is much weaker and varies significantly from one product to another product. This indicates that the NBL decadal variations are mainly associated with the first EOF mode of the decadal circulation variations in the northwestern tropical Pacific Ocean in all the ocean products. It therefore implies that the northerly/southerly NBL corresponds well to negative/positive SSH anomalies in the northwestern tropical Pacific Ocean.

Figure 5.

Lead-lag correlations between the decadal NBL time series and the PCs of the first (black line) and second (gray line) EOF modes of the decadal SSH anomalies in the northwestern tropical Pacific Ocean. The dashed red lines mark the statistically significant value with 95% confidence level. Positive (negative) lags denote the lag (lead) of the decadal NBL anomalies.

The above correspondence can also be directly seen from the highly negative linear correlations between the decadal NBL anomalies and the decadal SSH anomalies in the northwestern tropical Pacific Ocean as displayed in the plots in the first column of Figure 6. To further clarify the circulation variations associated with the NBL decadal variations, we calculate the regression maps of the decadal anomalies of the SSH and horizontal current averaged within the upper 100 m against the normalized time series of the NBL decadal anomalies for each ocean product at zero time lag. The regression maps for the SSH are given in the second column of Figure 6 while those for the horizontal current vector are not shown as they are generally consistent with the regression maps for the SSH satisfying the geostrophic balance. The resultant regression maps from the 11 ocean products all show negative SSH anomalies and cyclonic gyre anomalies (velocity vectors not shown) in the northwestern tropical Pacific Ocean with the center situating slightly to the north of the climatological tropical gyre center.

Figure 6.

Circulation variations in the tropical North Pacific Ocean associated with the decadal variations of the NBL. First column: Simultaneous linear correlations between the NBL decadal anomalies and the SSH decadal anomalies. Second column: Regression maps of the SSH decadal anomalies (m) against the normalized NBL decadal time series at zero time lag. Third column: Composite maps of the SSH decadal anomalies (m) during periods when the NBL decadal anomalies are larger than 0.2°N. Forth column: Same as the third column but during periods when the NBL decadal anomalies are smaller than −0.2°N.

To make this clearer, we calculate for each ocean product the composite decadal anomalies of the SSH and horizontal currents averaged within the upper 100 m for the periods when the NBL decadal anomalies are larger than 0.2°N and smaller than −0.2°N, respectively. The results agree well with the regression analysis. When the NBL decadal anomalies are larger than 0.2°N, the composite patterns in all products exhibit negative SSH anomalies (plots in the third column of Figure 6) and cyclonic gyre anomalies (velocity vectors not shown) in the northwestern tropical Pacific Ocean. When the NBL decadal anomalies are smaller than −0.2°N, the situations are reversed (plots in the fourth column of Figure 6). The above analyzes fully approve that the meridional migration of the NBL on the decadal time scale is closely related to the gyre anomalies that are present in the northwestern tropical Pacific Ocean. The gyre anomalies on the decadal time scale have already been noted by Zhai et al. [2013]. They pointed out that the strong/weak NEC decadal transport across 137°E is closely related to the cyclonic/anti-cyclonic gyre anomaly accompanying the decadal negative/positive SSH anomaly in this region. In the above section, we note that the time difference between peaks of decadal anomalies of the same sign of NBL and NEC transport is only about 0–2 years, which is quite short compared to the decadal time scale of ∼10 years. Therefore, we could conclude that the northerly/southerly migration of NBL and increase/decrease in NEC transport on the decadal time scale result from the same cyclonic/anticyclonic gyre anomaly. The cyclonic/anticyclonic gyre anomaly induces southward/northward velocity anomaly along the Philippine coast (velocity vector not shown) and thus push the NBL northward/southward.

3.3. Vertical Modes

In this part, we would examine the most important baroclinic modes involved in the decadal NBL variations and thus the decadal circulation variations in the northwestern tropical Pacific Ocean. Of the 11 products, the DePreSys and EN3v2a are excluded from such an examination as there are no simulated velocities in them. The region for the examination is chosen as 120°E–150°E, 5°N–20°N. With the barotropic and baroclinic modes, the decadal anomaly of the simulated horizontal velocity vector can be written as

display math(1)

where inline image is the barotropic velocity anomaly, inline image the vertical structure function of the horizontal velocity for the ith baroclinic mode and divided by its value at the sea surface, and inline image the velocity vector anomaly for the ith baroclinic mode at the sea surface. For each grid point (x, y) and time t, the barotropic velocity anomaly inline image is obtained via averaging inline image throughout the water column from sea surface to bottom. In equation (1), the baroclinic dynamical modes are orthogonal, that is, inline image with i≠j, where H is the ocean depth. Then for each baroclinic mode, the inline image can be determined following inline image. In each ocean product, inline image and inline image of the first six baroclinic modes are determined for each grid point and time moment as the ocean is basically dominated by the first several lowest baroclinic modes [e.g., Wunsch, 1997]. For each ocean product, the Brunt-Väisälä frequency is calculated from the mean temperature and salinity averaged over the whole time period of that ocean product and then is used to compute the vertical structure functions of the horizontal velocity using standard techniques [Gill, 1982].

Figure 7 shows the regression maps of the horizontal velocity decadal anomalies for the first (M1, first column) and second (M2, second column) baroclinic modes, respectively, against the normalized time series of the NBL decadal anomalies. The regressed velocity decadal anomaly of the barotropic mode is much smaller than those of baroclinic modes and thus not shown. Corresponding to positive NBL decadal anomalies, the regression maps for the M1 horizontal velocity decadal anomalies exhibit essentially cyclonic gyre anomalies in the northwestern tropical Pacific Ocean in most products but with the gyre anomaly centers locating at different positions for different ocean products. The magnitudes of the regressed M2 horizontal velocity decadal anomalies are smaller but not negligible than those of the regressed M1 horizontal velocity decadal anomalies in most products. Meanwhile, the patterns of the regressed M2 horizontal velocity decadal anomalies differ between different ocean products, possibly indicating that they are not robust. Current analysis indicates more important contributions from the first baroclinic mode than from the second baroclinic mode to the total circulation variations.

Figure 7.

Regression maps of the horizontal velocity decadal anomalies (cm s−1) for the first (M1, first column) and second (M2, second column) baroclinic modes, respectively, against the normalized time series of the NBL decadal anomalies in the nine ocean products.

We then calculate the explained variance of the total regressed horizontal velocity decadal anomalies (calculated in section 'Circulation Variations' but with no figure shown) by the sum of those of the first several baroclinic modes to quantify the relative contributions from different baroclinic modes. The explained variances are calculated for the zonal inline image and meridional inline image velocity components separately, following

display math(2)

In (2), inline image and inline image are the zonal and meridional components of the total regressed horizontal velocity decadal anomalies, inline image and inline image are the zonal and meridional components of the regressed horizontal velocity decadal anomalies for the ith mode, i=0 denotes the barotropic mode, m is the number of baroclinic modes, and the inline image indicates the spatial average over the region shown in Figure 7 but excluding the region west of the Philippines. Figure 8 shows the inline image and inline image for the nine ocean products. The result indicates that the decadal circulation variations responsible for the decadal NBL variations are dominated by the first two baroclinic modes and own the largest contribution from the first baroclinic mode. The percents of the variance of the horizontal velocity decadal anomalies explained by only the first baroclinic mode (roughly estimated as inline image and inline image) are mostly larger than 60% in all the nine products except GODAS and can be higher than 80% in several ocean products. Including the second baroclinic mode (roughly estimated as inline image and inline image) increases the percents to be higher than 90% for most products. In GODAS (Figure 8c), inline image is only about 49%, much lower than inline image, which is higher than 80%. Also inline image in GODAS is lower than that in other ocean products. The reason for these differences should be further explored in the future study.

Figure 8.

Su (black line) and Sv (gray line) as a function of the number of first several baroclinic modes. Zero values of the horizontal axes denote that only the barotropic mode is involved in (2).

Figure 9 compares the sum of the regressed horizontal velocity decadal anomalies of the first two baroclinic modes with the surface geostrophic velocity derived from the first EOF mode of the decadal SSH anomalies previously shown in the second column of Figure 4. The two kinds of anomalies agree reasonably well with each other in both magnitudes and spatial distributions except in GODAS, where the magnitude of the former is slightly larger than the latter. The analysis here further approves that the decadal ocean circulation variations in the northwestern tropical Pacific Ocean are mainly dominated by the first two baroclinic modes. Zhai et al. [2013] have well reproduced the decadal circulation variations in the northwestern tropical Pacific Ocean using the linear vorticity equation derived from the wind-forced 1.5 layer reduced gravity model, in which only the first baroclinic mode is involved. Our current analysis suggests that involving the second baroclinic mode in the dynamics would better reproduce the decadal circulation variations.

Figure 9.

Comparison of the sum of the regressed horizontal velocity decadal anomalies (cm s−1; blue arrows) of the first two baroclinic modes with the surface geostrophic velocity (cm s−1; red arrows) derived from the first EOF mode of the decadal SSH anomalies previously shown in the second column of Figure 4.

3.4. Relationship to Decadal Variability Modes in the Pacific Ocean

We then discuss the relationship of the decadal NBL variations to the main decadal variability modes in the Pacific Ocean. In the North Pacific Ocean, there is a well-known decadal variability mode named the Pacific Decadal Oscillation (PDO) [e.g., Mantua et al., 1997; Zhang et al., 1997; Mantua and Hare, 2002]. The PDO index is defined as the PC of the first EOF mode of the monthly sea surface temperature (SST) anomaly poleward of 20°N in the North Pacific Ocean and thus the influence center of PDO is mainly at high latitudes. Previous studies have pointed out that the PDO has significant imprints on the decadal variability of the tropical Pacific Ocean [e.g., Mantua et al., 1997; Zhang et al., 1997; Mantua and Hare, 2002]. On the other hand, the decadal variability in the tropical Pacific Ocean has also been paid much attention to [e.g., Tourre et al., 2001; Liu, 2003; White et al., 2003; Rodgers et al., 2004; Yeh and Kirtman, 2004, 2005; Yu and Boer, 2004; Tourre et al., 2005; Hasegawa et al., 2007; Kang and Zhu, 2010; Hasegawa et al., 2013]. The TPDV has also been called Quasi-Decadal (QD) variability with a typical fluctuating period of 8–12 years in several previous studies [Tourre et al., 2001, 2005; Hasegawa et al., 2007, 2013]. Here, we define the corresponding PC of the first EOF mode of the decadal SST anomalies in the tropical Pacific Ocean of 120°E–280°E, 25°S–25°N as the TPDV index, indicating the typical decadal variability mode in the tropical Pacific Ocean.

Zhai et al. [2013] have pointed out the close connection of the decadal variations of the NEC transport across 137°E and thus the circulation in the northwestern tropical Pacific Ocean to the variability modes of PDO and TPDV. They indicated that during the period of 1975–2005 the NEC transport decadal variations highly correlate with the TPDV leading by 7 months and with the PDO leading by 21 months. The different lead times possibly imply different relationships of the decadal variations of the circulation in the northwestern tropical Pacific Ocean with the PDO and TPDV. Because the length of the NEC transport time series in their study is relatively short (1975–2005), it is important to reevaluate the connection based on longer NBL time series derived from the ocean products.

First, we would examine the relationship between PDO and TPDV in each of the ocean products. Of the 11 products, GECCO, GODAS, and the two kinds of numerical simulations are excluded from such an examination. The GODAS and STORM are excluded due to their limit of time lengths, only 30 years after 1980. The GECCO is excluded as it does not cover years after 2000. The OFES is not examined here as its outputs adopted in the current study do not cover the tropical South Pacific Ocean (Table 1). At last, only seven of the 11 ocean products are used for examining the relationship between PDO and TPDV. As a comparison, the SST data from the Extended Reconstruction Sea Surface Temperature (ERSST) version 3b (ERSSTv3b) [Xue et al., 2003; Smith et al., 2008] is also used, which is obtained from all available observations only. The indices of PDO and TPDV are calculated according to their definitions. The PDO index is calculated as the PC of the first EOF mode of the monthly SST anomaly in the North Pacific Ocean of 120°E–260°E, 20°N–60°N as previous studies [e.g., Mantua and Hare, 2002] did. After being low-pass filtered (>8 years), the PDO index is subtracted the linear trend to get decadal anomalies. The TPDV index is calculated as the PC of the first EOF mode of the decadal SST anomalies in the tropical Pacific Ocean of 120°E–280°E, 25°S–25°N.

The first column in Figure 10 compares the indices of PDO and TPDV from multiple ocean products except the EN3v2a. The displayed ocean products basically give the same temporal evolutions of the two indices as derived from the ERSSTv3b. In EN3v2a, the PDO index shows different temporal variations from that in ERSSTv3b mainly during the period of 1980–1986, while the TPDV index is dominated by the variability on a longer time scale than that in ERSSTv3b (figure not shown). Using SST from EN3v2a during the period of 1975–2008, Zhai et al. [2013] have derived the same time series and spatial pattern of the TPDV as those derived with ERSSTv3b. Thus, the longer than decadal time scale dominating the TPDV in EN3v2a obtained here may be not true and the reason should be further explored by the analysis center. In the following, the EN3v2a would be excluded from further examination. In all the left six products and ERSSTv3b, the PDO shows high simultaneous correlation with the TPDV (r ≥ 0.6 above 95% confidence level). The significant differences between the two indices mainly occur during 1980–1985, when the positive maximum of TPDV (around 1980) leads that of PDO (around 1985) by about 4–5 years. This discrepancy should be further studied in the future and beyond our current study.

Figure 10.

Variability modes of PDO and TPDV in (a–f) the six ocean products and (g) ERSSTv3b. First column: Normalized time series of PDO (black line) and TPDV (red line). Second column: Regression maps of monthly SST anomalies (°C) within 25°S–60°N in the Pacific Ocean against the normalized time series of PDO. Third column: Same as the second column but against the normalized time series of TPDV. Numbers in the plot titles of the first column denote the simultaneous linear correlation coefficients between PDO and TPDV indices, all of which are well above 95% confidence level.

The second and third columns of Figure 10 display the regression maps of monthly SST anomalies within 25°S–60°N in the Pacific Ocean against the normalized time series of PDO and TPDV, respectively. Basically the derived regression maps from the six ocean products agree quite well with those from ERSSTv3b and previous studies [e.g., Zhang et al., 1997; White et al., 2003; Yeh and Kirtman, 2004; Zhai et al., 2013], implying that the six ocean products capture well the decadal variability modes of PDO and TPDV. The regression maps against PDO are generally the same as those against TPDV except that SST anomalies of the former are colder than those of the latter, especially in the tropical Pacific Ocean (comparing plots in the second and third columns of Figure 10). It is easy to understand that PDO has its influence center at high latitudes while TPDV has its influence center in the tropical Pacific Ocean [e.g., Hasegawa et al., 2007]. We may conclude that during the studied period the PDO could not fully represent the decadal variability of the tropical Pacific Ocean, even though it may have significant imprints on the latter [e.g., Mantua et al., 1997; Zhang et al., 1997; Mantua and Hare, 2002; Hasegawa et al., 2007].

Figure 11a shows the lead-lag correlations of the PDO time series with the TPDV time series in the six ocean products and ERSSTv3b. All of them display high correlations between PDO and TPDV (r > 0.5 above 95% confidence level). The highest correlations in the ocean products occur when the TPDV leads the PDO by about 1–6 months. The correlation maximum is highest in ERSSTv3b (r = 0.79 above 95% confidence level) and lowest in SODA224 (r = 0.56 above 95% confidence level). Figures 11b and 11c display the lead-lag correlations of the NBL time series with the PDO and TPDV time series, respectively. In general, around the zero time lag the NBL decadal anomalies show higher positive correlations with TPDV (multiproduct mean correlation maximum r ≈ 0.52 above 95% confidence level) than with PDO (multiproduct mean correlation maximum r ≈ 0.08). This discrepancy should be mainly due to the discrepancy between the PDO and TPDV during the period of 1980–1985. Thus, the current result implies that the decadal circulation variations in the northwestern tropical Pacific Ocean are dominantly controlled by TPDV rather than by PDO, consistent with that the circulation variations are mainly induced by the decadal variations of the sea surface wind forcing in the tropical North Pacific Ocean, which are closely related to TPDV [Zhai et al., 2013].

Figure 11.

Lead-lag correlations between (a) PDO and TPDV, (b) NBL and PDO, and (c) NBL and TPDV on decadal time scale in the utilized six ocean products. In Figure 11a, the lead-lag correlations between PDO and TPDV from ERSSTv3b are also displayed.

The time lagged by TPDV when the positive correlation maximum occurs is about −2.25 years in DePreSys, −0.75 years in ECDA, 0.58 years in ORAS3, 0.25 years in ORAS4, 0.33 years in SODA224, and 3 years in SODA226. All the ocean products except the DePreSys and ECDA produce decadal NBL variations that slightly lead the TPDV. Compared to the decadal time scale (∼10 years), the lagged time for positive correlation maximum is negligible. Therefore, decadally northerly/southerly NBL occurs in the warm/cold phase of TPDV and generally gets its northernmost/southernmost position around the mature phase. During the warm/cold phase of TPDV, atmospheric response to the warm/cold SST anomaly in the equatorial Pacific Ocean results in westerly/easterly wind anomaly near the equator and positive/negative sea surface wind stress curl anomaly in off-equatorial region in the western-central tropical North Pacific Ocean [e.g., White et al., 2003; Hasegawa et al., 2007; Zhai et al., 2013]. The wind forcing anomaly induces local upward/downward Ekman pumping anomaly, which triggers westward propagating upwelling/downwelling baroclinic Rossby waves [e.g., Jin et al., 2001; Zhai et al., 2013]. The accumulation of baroclinic Rossby waves combined with local Ekman pumping then generates cyclonic/anticyclonic gyre anomaly in the northwestern tropical Pacific Ocean [Zhai et al., 2013], which pushes the NBL northward/southward.

On the other hand, we also notice that with a lead of about half the decadal cycle the NBL time series display highly negative correlations with the TPDV index in the six ocean products except the SODA224 and SODA226. The multiproduct mean negative correlation minimum is about −0.78 above 95% confidence level, indicating closer correlation than the situation around zero time lag. The highly negative correlation might be one manifestation of the dynamical role played by the circulation in the northwestern tropical Pacific in the phase-shifting of TPDV. That is, the cyclonic/anticyclonic gyre anomaly in the northwestern tropical Pacific Ocean several years ago could possibly result in the TPDV shifting from warm/cold phase to cold/warm phase. Yu and Boer [2004] pointed out that the cyclonic/anticyclonic gyre anomaly in the northwestern tropical Pacific Ocean could induce cold/warm temperature anomalies through oceanic heat transport processes. The temperature anomalies are then advected to the central-eastern equatorial Pacific Ocean [e.g., Yu and Boer, 2004; Hasegawa et al., 2013] or propagate eastward along equator as equatorial coupled waves [e.g., White et al., 2003], shifting TPDV to its cold/warm phase. We also should note that this kind of lead-lag correlation between the NBL decadal time series and the TPDV index is absent in SODA224 and SODA226, implying that it needs more observational evidence.

4. Summary and Discussions

Up to now, the decadal variability (>8 years) of the NBL in the Pacific Ocean has been paid little attention mainly due to the lack of long-term in situ observations. In the present study, we attempt to examine the NBL's decadal variations and underlying mechanisms over the past six decades using 11 kinds of ocean products. The ocean products include seven kinds of ocean reanalyzes based on data assimilation systems, two kinds of numerical simulations without observations assimilated and two kinds of objective analyzes based on in situ observations only.

During the studied period of 1954–2007, the multiproduct mean of the decadal NBL anomalies shows maxima around 1965/1966, 1980/1981, 1995/1996, and 2003/2004 and minima around 1958, 1971/1972, 1986/1987, and 2000/2001, respectively. The S/N ratio generally increases with time and is smaller than 1 before the summer of 1982 and larger than 1 thereafter. This implies that the decadal variations of the NBL after the summer of 1982 obtained here can be placed more confidence as an estimate of true signals.

After about 1965, the occurring periods of the decadal NBL extremes agree well with those of the decadal NEC transport extremes at 0–2 years lead or lag. Especially, the extremes after 1975 correlate quite well with those of the NEC decadal anomalies across 137°E derived from long-term hydrographic observations [Zhai et al., 2013] at 0–2 years lead or lag. This indicates that the NBL and NEC transport in the northwestern tropical Pacific Ocean basically vary in phase with each other on decadal time scale. That is, on the decadal time scale, when the NEC bifurcates at northerly latitudes, its transport is larger than average; while when the NEC bifurcates at southerly latitudes, its transport is smaller than average. As for the relationship of the NBL decadal variations and those of the MC and KC transports, analyzes indicate that the NBL decadal variations correlate better with the MC transport than with the KC transport in terms of the occurring time of their extremes.

We then investigate the relationship of the NBL decadal variations with the decadal variations of the broader-scale circulation in the northwestern tropical Pacific Ocean. All the 11 ocean products display localized higher than around standard deviations of the decadal SSH anomalies in the northwestern tropical Pacific Ocean. An EOF analysis is applied to the decadal SSH anomalies in the region of 122°E–140°E, 7°N–20°N to reveal the main modes in the decadal circulation variations. The first EOF modes in all the ocean products are characterized by the spatially coherently varying SSH over the whole region and explain most of the total variance (>65%). Then lead-lag correlation analysis indicates that it is the EOF mode 1 that correlates with the NBL decadal variations. The highest correlation occurs at about 1 year lagged by the NBL in DePreSys but at near zero time lag in other ocean products. This therefore implies that the northerly/southerly NBL corresponds well to negative/positive SSH anomalies in the northwestern tropical Pacific Ocean.

This kind of relationship is further supported through regression and composite analyzes of the SSH and horizontal currents averaged within the upper 100 m. The regressed maps of the SSH and horizontal currents from the 11 ocean products all show negative/positive SSH anomalies and cyclonic/anticyclonic gyre anomalies in the northwestern tropical Pacific Ocean corresponding to northerly/southerly NBL. Composite analyzes give the same results. When the NBL decadal anomalies are larger than 0.2°N, the composite patterns in all products show negative SSH anomalies and cyclonic gyre anomalies in the northwestern tropical Pacific Ocean. When the NBL decadal anomalies are smaller than −0.2°N, the situations are reversed. These fully approve that the meridional migration of the NBL on the decadal time scale is closely related to the gyre anomaly forming in the northwestern tropical Pacific Ocean [Zhai et al., 2013]. The cyclonic/anticyclonic gyre anomaly produces southward/northward velocity anomaly along the Philippine coast and thus pushes the NBL northward/southward. This is similar to the situation on the interannual time scale [Zhai and Hu, 2012, 2013]. Vertical mode decomposition indicates that the decadal circulation variations and thus the decadal NBL variations are governed mostly by the first two vertical modes and own much more contributions from the first baroclinic mode than from the second baroclinic mode.

At last, we examine the relationship of the decadal NBL variations to the two decadal variability modes in the Pacific Ocean, the PDO and TPDV, with ocean products longer than 30 years and covering years after 2000. Basically, the chosen ocean products except the EN3v2a produce the same spatial patterns and corresponding temporal evolutions of the two variability modes as derived from ERSSTv3b. Overall, spatial patterns of the PDO and TPDV are similar to each other except that SST anomalies of the former are colder than those of the latter in the tropical Pacific Ocean given that the corresponding PCs are normalized. During the studied period, the PDO essentially agrees well with TPDV except during 1980s, when the warm phase of TPDV leads that of PDO by 4–5 years. Though beyond our current study, this discrepancy is interesting as many previous studies have pointed out that the PDO have significant imprints on the TPDV [e.g., Mantua et al., 1997; Zhang et al., 1997; Mantua and Hare, 2002]. Around the zero time lag the NBL decadal anomalies show higher positive correlations with TPDV than with PDO, consistent with that the circulation variations are mainly induced by the decadal variations of the sea surface wind forcing in the tropical North Pacific Ocean, which is related to TPDV [e.g., Yu and Boer, 2004; Zhai et al., 2013].

On the other hand, with a lead of about half the decadal cycle the NBL time series displays highly negative correlations with the TPDV in four ocean products. The multiproduct mean negative correlation minimum is about −0.78 above 95% confidence level. The highly negative correlation might be one manifestation of the dynamical role played by the circulation in the northwestern tropical Pacific in the phase-shifting of TPDV [Yu and Boer, 2004]. However, this kind of lead-lag correlation between the NBL decadal time series and the TPDV index is absent in SODA224 and SODA226 and should be further studied with more observations in the future. Meanwhile, several other mechanisms responsible for the phase-shifting of TPDV have also been proposed, which involve Sverdrup transport and emphasize the role played by the tropical South Pacific Ocean [e.g., Luo and Yamagata, 2001; Hasegawa and Hanawa, 2003; Hasegawa et al., 2007]. Therefore, whether the above highly negative correlation results from a dynamically causal relationship or just from variations of opposite phases but with no dynamical relationship should be further investigated.

Acknowledgments

This study was conducted under the Northwestern Pacific Ocean Circulation and Climate (NPOCE). We thank J.-S. von Storch for providing the STORM simulation, and H. Sasaki and his colleagues from the Earth Simulator for the OFES simulation. We also greatly appreciate the insightful and valuable comments from two anonymous reviewers. The altimeter products were produced by Ssalto/Duacs and distributed by Aviso, with support from Cnes (http://www.aviso.oceanobs.com/duacs/). The present study is sponsored by Project of State Strategic Program of Global Change (2013CB956202) and China Postdoctoral Science Foundation through grant 2013M530331.

Advertisement