Three-dimensional oceanic thermal structures and variability in the western North Pacific (NP) are examined on the interannual to decadal time scales and their relationship to oceanic and atmospheric variability is discussed by analyzing observation and reanalysis data for 45 years (1964–2008), which is much longer than the satellite-altimetry period. It is shown that the meridional shift of the Kuroshio Extension (KE) and subarctic frontal zone (SAFZ) is associated with the overall cooling/warming over the KE and SAFZ region (KE–SAFZ mode). It appears, however, that changes in KE strength induce different signs of thermal anomalies to the south and north of the KE, not extended to the SAFZ (KE mode), possibly contributing to noncoherent variability between the KE and SAFZ. Thus, the KE and SAFZ are dependent on each other in the context of the KE–SAFZ mode, while the KE is independent of the SAFZ in terms of the KE mode. This intricate relationship is associated with different linkages to atmospheric variability; the KE–SAFZ mode exhibits a relatively fast response to the large-scale wind stress curl forcing in the NP, whereas the KE mode is related to a delayed response to the atmospheric forcing via jet-trapped baroclinic Rossby wave propagation. It is suggested that further knowledge of the underlying mechanisms of the two modes would contribute to understanding ocean–atmosphere feedback as well as potential predictability over the western boundary current region in the NP.
The western North Pacific (NP) exhibits extensive air–sea interaction and distinctive climate variability closely related to the western boundary currents transporting anomalous heat and salt from the tropical region [see the reviews by Kwon et al. (2010) and Minobe et al. (2016)]. The western boundary current system in the midlatitude NP includes the Kuroshio Extension (KE) and subarctic frontal zone (SAFZ), which have been often referred to as the Kuroshio–Oyashio Extensions and recognized as one broad zonal structure because of the relatively coarse horizontal resolution of observational datasets and model outputs. Recent high-resolution satellite data and numerical model results have shown that the KE is identified by a strong eastward current located at about 35°N to the east of Japan, whereas the SAFZ is characterized by a broad region of sea surface temperature (SST) fronts around 41°N (to the north of the KE). The KE and SAFZ thus contribute oceanic thermal variations at different depths; the KE-associated variability is observed in the main thermocline (about 200–600 m), while that related to the SAFZ is found in the mixed layer (Deser et al. 1996; Tourre et al. 1999). Accordingly, the KE exhibits a distinct meridional gradient of sea surface height (SSH) and a modest SST gradient, whereas the SAFZ shows a strong meridional gradient in SST, not in SSH (Nonaka et al. 2006).
SST has been one of the most extensively analyzed variables to study climate variability in the NP, partly because of its better availability in both time and space. The most dominant mode of the NP variability [i.e., the Pacific decadal oscillation (PDO); Mantua et al. (1997)] is derived from empirical orthogonal function (EOF) analysis of SST in the NP (north of 20°N) and has been widely used to understand the long-term changes in the physical and biological conditions in the NP. However, the different vertical structures of the KE and SAFZ have raised the importance of looking at both surface and subsurface layers, the three-dimensional structures of the variability in the western NP. Moreover, the relationship between the KE and SAFZ should be investigated considering not only the SST but also upper-ocean (including the main thermocline) variability.
The KE and SAFZ region is mainly affected by large-scale wind stress curl forcing in the NP that is considered to be related to the two dominant modes of NP variability [i.e., PDO and North Pacific Gyre Oscillation (NPGO); Di Lorenzo et al. (2008)]; basin-scale wind stress curl fluctuations associated with the two modes induce sea level anomalies in the central and eastern NP, which propagate westward as first-baroclinic-mode Rossby waves and alter variability around the KE (Qiu 2003; Qiu and Chen 2005, among others). Observational studies have noted that the KE and SAFZ shifted southward from the mid-1970s to the mid-1980s and the SAFZ experienced associated cooling not only in the SST (Nakamura et al. 1997; Nakamura and Kazmin 2003) but also in the subsurface layers over the KE and SAFZ region (Deser et al. 1999). The decadal transition was explained to be associated with the 1976/77 climate transitions of the atmospheric circulation over the NP (i.e., the Aleutian low pressure system; Deser et al. 1999; Miller and Schneider 2000; Seager et al. 2001).
Hindcasts by a high-resolution global ocean general circulation model (OGCM) also reproduced the southward shift of the KE and SAFZ, and associated cooling from the mid-1970s to the mid-1980s (Nonaka et al. 2006; Taguchi et al. 2007). Recently, Sasaki and Schneider (2011) suggested meridional shifts of the KE on decadal time scales are due to the jet-trapped Rossby waves propagating westward along the KE based on the eddy-resolving OGCM results. The eddy resolving (i.e., with spatial resolution of less than about 0.1°) OGCM was able to represent the KE and SAFZ variations more in detail compared to the coarser-resolution observation data, and suggested that KE and SAFZ are not always coherent, possibly driven by different mechanisms (Nonaka et al. 2006). In particular, the decadal temperature variability in the SAFZ is not fully explained by the westward propagation of the baroclinic Rossby waves forced by large-scale Ekman pumping anomalies in the central NP. As the KE and SAFZ exhibit different spatial structures, analyzing three-dimensional and high-resolution dataset will be of particular importance for an in-depth examination of the KE and SAFZ variability.
Since the satellite altimetry data became available in the early 1990s, SSH analysis has been broadly applied to explore the KE and SAFZ variability on climate time scales (Qiu and Chen 2005, 2010; Sasaki et al. 2013, among others). The near 20-yr satellite SSH data, however, do not extend to the 1970s when the 1976/77 climate regime shift occurred, which could limit our understanding of the decadal-scale variability. The different periods of analysis indeed would have led different views on the independence of the two modes of forcing because the PDO and NPGO indices may not be very well separated in the recent 20-yr period (Qiu and Chen 2010).
This study, thus, analyzes the reanalysis dataset together with the gridded observation dataset for 45 years (1964–2008) to investigate the three-dimensional structure of the KE and SAFZ variability on interannual to decadal time scales, for longer than the satellite-altimetry period. To extend our analysis from surface to subsurface layer of the western NP, the present study employs ocean heat content as a main variable. Different from the heat content, SST, especially on climate time scales, is a consequence of air–sea interaction through surface heat fluxes in most of midlatitudes, not representing ocean dynamics itself; SST anomalies decorrelate from themselves rapidly on seasonal time scales, particularly in the extratropical oceans (Qiu and Kelly 1993; Kelly and Qiu 1995; Deser et al. 2003). Ocean heat content, on the other hand, was suggested as a better climate indicator because of its long-term memory (Sutton and Mathieu 2002; Kelly et al. 2010).
The following section describes the datasets and the analysis procedures employed in this study. Variability of the upper-ocean heat content, its relationship with the KE and SAFZ variability, and that with the atmospheric variability are examined in sections 3, 4, and 5, respectively. Discussion and conclusions are presented in section 6.
2. Data and methods
Temperature, salinity, current velocity, SSH, and wind stress data with a spatial resolution of 0.5° in the western NP are taken from the Simple Ocean Data Assimilation (SODA) reanalysis, version 2.1.6 (Carton and Giese 2008), for the time period of 1964 to 2008 (45 years). SODA, version 2.1.6, is forced by ERA-40 wind stress (Uppala et al. 2005) until 2001 and by ERA-Interim wind stress (Dee et al. 2011) during 2002–08. It is assimilated with hydrographic observations from World Ocean Database 2009 (Boyer et al. 2009) that have been corrected for the drop-rate error of XBT as noted by Levitus et al. (2009). The SODA dataset has been widely used to investigate long-term ocean dynamics as it provides subsurface structures and their time variability. Near-surface (5.01 m) temperature of SODA is compared with the Extended Reconstructed SST (ERSST, https://www.ncdc.noaa.gov/data-access/marineocean-data/extended-reconstructed-sea-surface-temperature-ersst-v3b) in Fig. 1.
The SODA monthly mean temperature and salinity data at 21 vertical levels (5.01, 15.07, 25.28, 35.76, 46.61, 57.98, 70.02, 82.92, 96.92, 112.32, 129.49, 148.96, 171.4, 197.79, 229.48, 268.46, 317.65, 381.39, 465.91, 579.31, and 729.35 m) are used to calculate ocean heat content in the upper 700-m layer. Upper 700-m ocean heat content from monthly mean gridded observational dataset with a spatial resolution of 1° (Ishii and Kimoto 2009; this dataset is hereafter referred to as the IK09) is also used to compare with the ocean heat content derived from SODA. The SODA monthly mean velocity at the 21 vertical levels, SSH, and wind stress are used to investigate the relationship with the heat content variability. Additionally, sea level pressure (SLP) data from the NCEP–NCAR reanalysis (Kalnay et al. 1996) is used to examine the relationship between the ocean heat content and the atmospheric forcing in the NP. (The PDO index and the NPGO index are taken from http://jisao.washington.edu/pdo/ and http://www.o3d.org/npgo/index.html, respectively.)
b. Calculation of heat content
To calculate upper-ocean heat content, the SODA temperature and salinity at the upper 21 levels from 5.01 to 729.35 m are linearly interpolated at every 1-m depth. Upper 700-m-integrated heat content, , is then calculated via
where is temperature, is salinity, is density, and is specific heat capacity at constant pressure. The mean and standard deviation of the SODA heat content for 45 years (1964–2008) is presented in Figs. 1e and 1f. Note that the standard deviation in Fig. 1 is calculated after applying a 13-month running mean filter to focus on long-term variability. Temporal variability is particularly large along about 35°–37°N. A major difference between the spatial pattern of heat content and that of SST (or near-surface temperature, the first two rows in Fig. 1) is that SST shows large variability along about 37°–42°N associated with the SAFZ (Yasuda et al. 1996; Nonaka et al. 2006), which is farther north compared with the heat content.
The SODA heat content is also compared with the heat content of IK09 during the same period in Figs. 1g and 1h. It is obvious that the SODA heat content (the third row in Fig. 1) captures smaller-scale variability than does the IK09’s heat content (the bottom row in Fig. 1), reflecting that the IK09 is substantially smoothed by optimal interpolation (Ishii and Kimoto 2009). They, however, generally agree well in the overall structures; both show strong variability to the east of Japan at about 35°–37°N.
c. Cyclostationary EOF analysis
The SODA heat content and IK09 heat content for the region in Fig. 1 (25°–50°N, 140°E–180°) are analyzed by conducting cyclostationary empirical orthogonal function (CSEOF) analysis (Kim et al. 1996; Kim and North 1997), which is more appropriate for many geophysical variables compared to conventional EOF analysis based on the assumption of stationarity (Kim et al. 2015). The SODA and IK09 heat content are decomposed into cyclostationary loading vectors (CSLVs), , and their corresponding principal component (PC) time series, , as in (2):
where , , and denote the mode number, space, and time, respectively. CSLVs are orthogonal and PC time series are uncorrelated with each other as in conventional EOF analysis. However, CSLVs are periodic with the nested period , which is set to be 12 months in this study:
Thus, each CSLV represents a physical evolution over the period of d and corresponding PC time series shows time-varying strength of the physical evolution. The 12-month nested period naturally captures the seasonal cycle as one mode; if the seasonal cycle is the most dominant mode, it is extracted as the first mode. One advantage of applying the CSEOF technique in the present study is that dominant modes are obtained without employing a filter to remove the seasonal cycle.
The relationships between the heat content variability, temperature, salinity, and current velocity are investigated via multiple regression analysis using the CSLVs and PC time series of each variable:
where is the target PC time series for mode , is the predictor PC time series for mode , and is regression error. Target variable is the heat content in the western NP and predictor variables are the temperature, salinity, and current velocity at each layer in the upper 700 m of the same region in the western NP (25°–50°N, 140°E–180°). Note that the multiple regression in (4) is conducted by using the CSEOF PC time series of the predictor variables at each layer in order to find a common long-term undulation between the target heat content mode and each predictor variable at each layer. Regressed anomalies of the predictor variables, , are obtained using the regression coefficients in (4):
where are the CSLVs of the predictor variable. The resulting spatial patterns, , share the long-term undulation of the target variable and thus the evolution of the predictor variable is physically consistent with the evolution of the target variable. A detailed explanation on the regression analysis in CSEOF space is presented in Kim et al. (2015).
3. Upper-ocean heat content variability
Each CSEOF mode of the upper-ocean heat content is composed of 12 monthly spatial patterns and corresponding PC time series, as described in section 2c. The first CSEOF mode captures the typical seasonal variations with positive anomalies in summer months and negative anomalies in winter months (not shown). The second and third CSEOF modes explain about 19% and 16% of the interannual variance (total variance after applying a 13-month running mean filter), respectively. Because the 12 monthly spatial patterns show little monthly evolution for each of the second and third modes, 12-month averages of the loading vectors, and in (2), are presented in Figs. 2 and 3. These two modes are nearly identical with the first and second EOF (not CSEOF) modes of the 13-month running averaged upper-ocean heat content. The EOF PC time series are also displayed as black curves in the bottom panels of Figs. 2 and 3 for reference.
The spatial pattern of the second CSEOF mode exhibits generally the same sign of strong anomalies along the latitude band between 33° and 42°N (i.e., along the KE and SAFZ; Fig. 2); henceforth, this mode is referred to as the KE–SAFZ mode. The corresponding PC time series (bold blue line in the bottom panel of Fig. 2) shows interannual variability with a negative to positive phase shift during 1964–2008. The KE–SAFZ mode is also captured from the IK09 heat content explaining 18% of the interannual variance, and the PC time series from the IK09 (dotted blue line in the bottom panel of Fig. 2) agrees well with that from the SODA (correlation coefficient, r = 0.78) confirming the validity of SODA.
The positive trend in the PC time series of the KE–SAFZ mode together with negative anomalies in the spatial pattern indicates that the KE–SAFZ mode is associated with a reduction in upper-ocean heat content over the KE region and SAFZ during the 45 years. The change of sign from negative to positive in the early 1980s suggests that the KE–SAFZ mode may be related to the 1976/77 climate transition and the subsequent southward shift of the KE and SAFZ suggested by previous studies (e.g., Deser et al. 1999; Taguchi et al. 2005; Nonaka et al. 2006; Taguchi et al. 2007, among others). The relationship between the KE–SAFZ mode and the KE and SAFZ variability is investigated in the following section.
The spatial pattern of the third CSEOF mode depicts positive and negative anomalies to the south and north of about 35°N in the upstream region of the KE (west of 155°E) (top panel of Fig. 3); thus, this mode is referred to as the KE mode hereafter. The corresponding PC time series (bold red line in the bottom panel of Fig. 3) shows decadal variability with relatively small fluctuations before the early 1980s and larger fluctuations in the later years. The KE mode is also captured from the IK09 heat content explaining 26% of the interannual variance, and the PC time series from the IK09 (dotted red line in the bottom panel of Fig. 3) agrees well with that from the SODA (correlation coefficient, r = 0.81).
The KE mode represents warming to the south and cooling to the north of about 35°N (i.e., the mean KE latitude; or cooling to the south and warming to the north depending on the sign of corresponding PC time series), while the KE–SAFZ mode indicates general cooling or warming over the KE region and SAFZ. The KE–SAFZ and KE modes are not only uncorrelated (zero correlation between the two PC time series), but also nearly independent (insignificant lead–lag correlations between the two PC time series). It is also noteworthy that the KE mode becomes a more dominant mode than the KE–SAFZ mode in terms of the explained variance when similar analysis is conducted using only the recent SODA data from 1993 (the satellite era). This is partly because a large portion of variance in the KE–SAFZ mode comes from the phase shift itself in the early 1980s; the KE–SAFZ mode becomes less dominant when the analysis period is limited to the recent and the possible reason is discussed in sections 5 and 6. The spatial patterns and PC time series of the KE–SAFZ and KE modes derived from the shorter period (not shown) are, however, consistent with those from the 45-yr record, indicating robustness of these modes.
4. Relationship with the KE and SAFZ variability
The KE (not SAFZ) variability is well represented by SSH variability in the western NP, and the meridional shift of the jet axis and changes in the jet strength have been considered as the two main characteristics of the observed KE variability (Qiu and Chen 2005, 2010; Sasaki et al. 2013, among others). The two characteristics are also captured as the two dominant modes of SSH variability by the numerical model results (see Fig. 4 in Taguchi et al. 2007). Figure 4a shows zonally averaged SSH over the upstream KE and SAFZ (145°–155°E) after applying a 13-month running mean filter. The black bold line is the latitude of maximum meridional gradient and its normalized time series is compared with the PC time series of the KE–SAFZ mode in Fig. 4b. Their negative correlation (r = −0.46) implies that a southward shift of the KE jet is related to general cooling of the upper ocean over the KE region and SAFZ. Correlation between the latitude of KE axis and the KE mode is not statistically significant (r = −0.12). Rather, the KE mode is correlated with the KE strength (r = 0.70, Fig. 4c), while the KE–SAFZ mode is not significantly correlated with the KE strength (r = 0.22).
As only the KE variability, not the overall KE and SAFZ variability, is well reflected on SSH fields, other ocean variables are also analyzed to look into vertical structures of the KE and SAFZ. Mean temperature, salinity, and zonal velocity are shown over the upstream (145°–155°E) and downstream (155°–165°E) regions in Fig. 5. The upstream mean temperature shows warmer temperatures to the south and colder temperatures to the north with a large meridional gradient over 33°–42°N, which encompasses the KE along about 35°N, the SAFZ around 41°N, and a mixed water region in between (Yasuda et al. 1996; Yasuda 2003). A similar, but less strong meridional gradient is observed over the downstream KE and SAFZ (Fig. 5b). Mean salinity shows saline waters to the south and fresh waters to the north with a large meridional gradient around the same latitude as the temperature (Fig. 5c). Over the downstream, low salinity water to the north is found at a shallower depth with a larger vertical gradient in comparison with the upstream region. Eastward velocity is strongest at about 35°N as inferred from the meridional temperature gradient, but a double-core structure is found over the downstream region (Fig. 5f).
Regression analysis of these ocean variables onto the target upper-ocean heat content provides more in-depth understanding of the KE–SAFZ and KE modes and their associated ocean dynamics in the western NP, which is one of the main goals of this study. Upper-ocean temperature, salinity, and current velocity at each level are decomposed into the CSEOF modes and multiple regression analysis was conducted using their PC time series, as explained in section 2c. Figure 6 shows R-squared values of the multiple regression of PC time series from each variable at each depth onto the KE–SAFZ and KE modes. As a result of this exercise, the regressed anomalies shown in Figs. 7 and 8 exhibit long-term modulation according to the PC time series of the KE–SAFZ mode and the KE mode, respectively; the regression error can be inferred from the R-squared values in Fig. 6. It is apparent that the temperature regression exhibits higher R-squared values because the upper-ocean heat content is mainly governed by temperature. The highest R-squared values of temperature regression are found in the subsurface layer, particularly at 200–400-m depth (Fig. 6a). Salinity regression generally shows lower R-squared values compared to those of temperature regression, with the highest values at 300–400-m depth (Fig. 6b). In the case of current velocity, change in R-squared values with depth is smaller than those of temperature and salinity regression. Obtaining regressed anomalies that share significantly correlative long-term modulation with the target modes so that those can be compared with the spatial patterns of target modes, greatly helps in understanding the relationship between the different variables and is one of the key advantages of applying CSEOF technique.
Regressed anomalies targeting the KE–SAFZ mode are presented with the mean fields superposed in Fig. 7. Temperature regression onto the KE–SAFZ mode shows the same sign of anomalies with a maximum at about 350 m near 36°N in the upstream (color shading in Fig. 7a), where the meridional slope of the mean isotherms is the steepest. The subsurface negative temperature anomalies generally follow the depths of large mean temperature gradient, which become shallower to the north and reach the surface at around 41°N. The regressed salinity anomalies onto the KE–SAFZ mode also exhibit negative anomalies where the mean meridional salinity gradient is large, but the maximum anomalies are observed near the surface at around 41°N (Fig. 7c). These same signs of anomalies both along the KE and SAFZ are similar to those of Sugimoto et al. (2014), but their analysis was mainly focused on the SAF for a shorter time period during 1982–2011. The downstream KE and SAFZ region (155°–165°E) generally exhibit consistent vertical structures compared to those over the upstream; large temperature and salinity anomalies follow the depths of the strong mean meridional gradient, but maximum values are located farther north at a shallower depth compared to those over the upstream KE and SAFZ region. It is notable that the regressed salinity anomalies are larger in the downstream (eastward) region than those in the upstream (westward) region.
The negative temperature anomalies (not a pair of different signs), which are consistent with those of the KE–SAFZ mode in Fig. 2, are related to the southward shift of the isotherms and the axis of the KE and SAFZ. The PC time series of the KE–SAFZ mode indicates meridional shifts on interannual time scales, specifically a pronounced southward phase shift in the early 1980s. Thus, the decrease in upper-ocean heat content during the 45 years is associated with subsurface-intensified cooling at about 34°–37°N and surface-intensified cooling at about 41°N induced by a southward shift of the KE and SAFZ, respectively. Previous studies also reported cooling in the western NP from the 1970s to the 1980s by analyzing the observational data without detailed three-dimensional structure (Deser et al. 1999; Joyce and Dunworth-Baker 2003) and by numerical model experiments (Taguchi et al. 2005; Nonaka et al. 2006; Taguchi et al. 2007). The southward shift of the KE is consistent with positive and negative zonal velocity anomalies, respectively, to the south and to the north of the mean current axis as seen in Figs. 7e and 7f. The anomalies are evident at the latitude of the KE, but are not apparent near the latitude of SAFZ, which is different from those of temperature and salinity.
Additionally, both the negative anomalies of temperature and salinity of the KE–SAFZ mode are consistent with the decadal changes in temperature and salinity from the earlier period (1968–72) to the later period (1984–88) shown by the numerical model results (see Fig. 8 in Nonaka et al. 2006). In fact, the KE mode exhibits the same phase during the two periods according to the PC time series in Fig. 3, which makes the KE–SAFZ mode a primary contributor to the difference between the two time periods. Indeed, the KE–SAFZ mode exhibits a negative phase in the earlier period (1968–72) and a positive phase in the later period (1984–88), which affirms that the model captured the decadal change associated with the KE–SAFZ mode. Some studies, which focused on the recent 20-yr period when the satellite altimetry data are available (Qiu and Chen 2005, 2010; Sasaki et al. 2013, among others), mainly accounted for the KE mode since it is more dominant of the two during the recent period.
Regressed anomalies targeting the KE mode are presented in Fig. 8. Temperature and salinity regression exhibits different signs of anomalies in the upstream, unlike the same sign of anomalies regressed onto the KE–SAFZ mode in Fig. 7. The positive and negative anomalies to the south and north of about 35°N are consistent with the spatial pattern of the KE mode (Fig. 3). These different signs of anomalies contribute to a steeper slope of the isotherms, thereby leading to a stronger eastward velocity of the KE. It is noteworthy that the regressed temperature anomalies are not extended to the latitude of SAFZ nor to the surface layer (different from those targeting the KE–SAFZ mode), suggesting that the KE mode represents the changes in the KE strength. Strong positive anomalies of zonal velocities are obtained along the mean KE axis, suggesting that the jet is strengthened. The negative anomalies to the south might be related to the recirculation of the jet (Qiu and Chen 2010).
The vertical sections for the upstream KE and SAFZ are zonally averaged over 145°–155°E, slightly to the east of where the KE–SAFZ and KE modes show the largest anomalies in Figs. 2 and 3. Note that consistent patterns of temperature anomalies are obtained along the 143°E meridional section where the KE–SAFZ and KE modes show the largest anomalies; the negative anomalies for the KE–SAFZ mode and the different signs of anomalies for the KE mode are seen about 2°–3° farther north compared to those of the 145°–155°E zonal mean meridional sections (not shown).
Temperature anomalies over the downstream region regressed onto the KE mode (Fig. 8b) exhibit only positive anomalies to the south of about 35°N, which is different from those over the upstream region (Fig. 8a), but correspond with the spatial pattern of the KE mode that shows negative anomalies confined in the upstream region (Fig. 3). Negative salinity anomalies are observed where the mean meridional salinity gradient is large, same as those over the upstream region. Downstream mean zonal velocity is considerably weak compared to that over the upstream, but the regression still show weak positive anomalies at the latitude of the strongest eastward flow, suggesting that the jet is strengthened also in the downstream region but with a much weaker magnitude.
5. Relationship with atmospheric variability
The decadal variability in the western NP has been understood to be affected by large-scale wind stress curl forcing in the NP [see the reviews by Kwon et al. (2010) and Minobe et al. (2016)]. Lag correlations of the KE–SAFZ mode and KE mode with the PDO index and with the NPGO index (Fig. 9), thus, would provide an insight into the relationship between the decadal KE and SAFZ variability and the atmospheric variability. The KE–SAFZ mode exhibits significant correlation with the PDO index at a half-year lag, while the KE mode shows significant correlation with the NPGO index when the NPGO leads the KE mode by about 3 years. The half-year time lag with the PDO index is different from the previous understanding that the basin-scale wind stress curl fluctuations associated with the two modes induce sea level anomalies in the central and eastern Pacific, which propagate westward and alter variability of the KE in about 3–5 years (Deser et al. 1999; Qiu 2003; Qiu and Chen 2005; Taguchi et al. 2005; Seager et al. 2001; Taguchi et al. 2007; Ceballos et al. 2009; Sasaki et al. 2013, among others), although Minobe (2002) reported simultaneous changes in SST and heat content between the central NP and the western NP region in the late 1990s.
The lag-correlation maps, rather than lag correlation with a single index, allow more detailed inspection of the relationship between the upper-ocean heat content modes and atmospheric variability. Figure 10 shows lag correlations of the KE–SAFZ mode with SLP, wind stress curl, and SSH at every 6-month lag, which is chosen based on the lag correlation with the PDO index. The SLP, wind stress curl, and SSH are smoothed via a 13-month running mean filter and correlations are calculated after removing the linear trends in the variables. The second row from the bottom in Fig. 10 is for the zero-lag correlation and strong negative correlation with SSH along about 35°–40°N is dominant, indicating that a southward shift of the KE (positive phase of the KE–SAFZ mode) is related to a decrease of SSH over the region. The related wind stress curl forcing reaches its maximum one year earlier and appears to be related to SLP variability of the Aleutian low. The correlation between the north–south shift of the Aleutian low and that of SAFZ was previously reported by Seo et al. (2014) to be a longer lead–lag relationship (3-yr lead by Aleutian low), and it was mainly focused on the SST analysis for a shorter time period during 1982–2011.
Figure 11 shows lag correlations of the KE mode with the SLP, wind stress curl, and SSH at every 1-yr lag, which is chosen based on the lag correlation with the NPGO index. The second row from the bottom is for the zero-lag correlation and strong positive correlation with SSH to the south of KE agrees with a stronger KE (positive phase of the KE mode). It appears that SSH anomalies propagate from east to the KE region in about 3 years and the spatial patterns of the westward-propagating signal correspond to those of jet trapped Rossby waves in Sasaki et al. (2013),which is based on the satellite altimetry SSH during 1993–2010. A significant atmospheric signal is also found with about 3–4-yr time lead. Interestingly, the SLP pattern shows a meridional-dipole-like structure in the eastern NP at a 4-yr lead, in agreement with Di Lorenzo et al. (2008) and Ceballos et al. (2009), while the SLP pattern at a 3-yr lead exhibits a monopole-like structure in the eastern NP. At both lead times, however, correlations with wind stress curl commonly exhibit negative anomalies around 35°–40°N accompanied by positive SSH correlations roughly at the same locations.
The lag correlation analysis indicates that the KE–SAFZ and KE modes exhibit different dynamics of the oceanic response to the atmospheric forcing. Figure 12 highlights the difference by presenting lag correlations of the two modes with SSH averaged over 32°–38°N. In the case of the KE mode (Fig. 12b), SSH anomalies of the upstream KE–SAFZ region appear to be linked with those over the region of atmospheric forcing with about a 3–4-yr time lag as seen in Fig. 11c. In the case of the KE–SAFZ mode (Fig. 12a), however, the same-signed anomalies over the upstream KE–SAFZ region and over the region of atmospheric forcing exhibit time lags shorter than one year as seen in Fig. 10c. This suggests that the mechanism causing the meridional shift of the KE and SAFZ may be different from that affecting the strength of KE. The seemingly disconnected propagating signals in both Figs. 12a and 12b are partly due to the meridional averages taken for the presentation.
It was noted that the Kuroshio transport in the East China Sea exhibits zero-lag correlation with the PDO index; during the positive phase of the PDO, the upstream Kuroshio transport increases because of negative SSH anomalies arriving at the onshore side of the East China Sea–Kuroshio as barotropic Rossby waves from the latitude of KE and SAFZ (Andres et al. 2009, 2011). The increased East China Sea–Kuroshio transport is reported to affect flow variability south of Japan and downstream Kuroshio. Thus, the faster response of the KE–SAFZ mode may be related to the barotropic component in association with the upstream Kuroshio variability. Further studies are necessary for understanding the mechanism of this fast response. Numerical experiments may also be helpful to understand the possible differences in the response characteristics of the KE and SAFZ to different wind forcings.
6. Discussion and conclusions
Upper-ocean thermal variation in the western part of the NP on interannual to longer time scales is largely driven by ocean dynamics and transport (i.e., western boundary current variability), different from the eastern NP where surface heat flux forcing would be a major source of the variability (Kelly and Dong 2004; Kwon et al. 2010; Minobe et al. 2016). This study presented the decadal variability of the oceanic thermal conditions in the western NP associated with the KE and SAFZ variability. Looking into the subsurface variability, not limited to SST, enabled more in-depth understanding of the spatial structure of the KE and SAFZ variability. The two dominant modes of the upper-ocean heat content variability exhibit different spatial structures; the KE–SAFZ mode shows generally the same sign of anomalies over the KE and SAFZ extended to the downstream region (Fig. 2), whereas the KE mode shows positive and negative anomalies to the south and north of the KE (Fig. 3). Regressed anomalies of temperature and salinity targeting the KE–SAFZ mode (Fig. 7) are larger in the subsurface layer near the latitude of the KE, but are stronger in the surface layer near the latitude of SAFZ. This suggests that the KE–SAFZ mode is associated with both the KE and SAFZ variability. The anomalies targeting the KE mode (Fig. 8) are limited to the southern latitude compared to those targeting the KE–SAFZ mode, which implies that the KE mode is mainly related to the KE variability. Regression of current velocity shows a pair of positive and negative anomalies for the KE–SAFZ mode at the latitude of KE (Figs. 7e and 7f), indicating meridional shift of the jet axis, but the same-signed anomalies are observed for the KE mode (Figs. 8e and 8f), suggesting changes in jet strength.
Analysis of the 45-yr (1964–2008) reanalysis dataset, longer than the satellite-altimetry period, helped us understand the three-dimensional structure of the decadal western NP variability. Some studies noted that the two leading modes (KE–SAFZ and KE modes) are not independent; the northward (southward) shift of the KE axis is accompanied by the strengthening (weakening) of the KE jet (Qiu and Chen 2005, 2010; Sasaki et al. 2013). This study argues that the two characteristics exhibit different spatial and temporal variability possibly based on different mechanisms. Different periods of analysis would have led to different views on the independence of the two modes and longer data may be beneficial to confirm if they are indeed independent of each other or not. It also needs to be pointed out that the linear trend of the upper-ocean heat content is highly sensitive to the period of calculation, and hence natural variability on decadal time scales needs to be carefully interpreted in regard to the linear trend. An attempt to separate the anthropogenic effect from the internal climate variability indeed showed distinct differences between the linear trends of the SSH with and without removing the PDO contribution (Hamlington et al. 2014), although the PDO itself is considered to be a combination of different processes (Newman et al. 2016). The sensitivity of the natural variability contribution to the data period highlights the importance of understanding climate variability by employing sufficiently long data.
The present results also are important in understanding ocean–atmosphere feedback over the western NP. The western NP has attracted much attention because KE and SAFZ is a unique region where SST and ocean–atmosphere heat flux variability are mainly controlled by oceanic circulation (Kelly 2004; Tanimoto et al. 2003), while they are generally controlled by atmosphere over most of the midlatitude NP region (Cayan 1992). Recent observational studies showed that KE and SAFZ could influence interannual to decadal large-scale variability of atmospheric circulation over the NP (Frankignoul et al. 2011; Taguchi et al. 2012). In particular, Taguchi et al. (2012) suggested that SST variability in the SAFZ plays an important role in basin-scale atmospheric circulation variability associated with changes in the strength of the Aleutian low. Furthermore, based on observational analysis, Qiu et al. (2014) suggested that ocean–atmosphere feedback from the KE leads enhanced quasi-decadal variability, resulting in longer predictability. A complete review of major findings on the feedback from the western boundary currents in this region can be found in Kwon et al. (2010) and Minobe et al. (2016).
The presence of the KE–SAFZ and KE modes, thus, has substantial implications on the ocean–atmosphere influence. First, although SST variability in the SAFZ were treated as if they were independent from KE variability in the previous studies (Frankignoul et al. 2011; Taguchi et al. 2012), the spatial structure of the KE–SAFZ mode (Fig. 2) and its regressed temperature anomalies (Figs. 6a and 7a) suggests that a part of SST variability in the SAFZ is closely related to the KE variability in the upstream KE and SAFZ region. The physical mechanism connecting the surface SAFZ and subsurface KE, however, remains to be explored in future studies. Second, KE variability studied by Qiu et al. (2014) is significantly consistent with the KE mode in the present study. The KE index defined by Qiu et al. (2014, see their Fig. 5b) strongly resembles the PC time series of the KE mode (Fig. 3). Also, the spatiotemporal evolution of SSH signals in their paper is generally consistent with the SSH evolution of the KE mode (Fig. 11c). The close similarity between Qiu et al.’s (2014) KE index and the present KE mode is probably because their domain for computing the area-averaged SSH to obtain the index is over and to the south of the mean KE axis, which is where the present KE mode exhibits large amplitude (Fig. 3). Although Qiu et al. (2014) suggested that the long Rossby waves are important agents of transferring the SSH anomalies from the central NP to the western boundary current region, the jet-trapped Rossby waves along the KE are likely to play an essential role in the propagation to the west of the date line as described in the previous sections; a series of regressed SSH anomalies in Fig. 11c are consistent with the jet-trapped SSH anomalies in Figs. 6 and 7 of Sasaki et al. (2013), which are stronger to the south of the KE axis. The PC time series of the KE mode in Fig. 3 also resembles that in Fig. 2b of Sasaki et al. (2013), which is linked with large-scale atmospheric pressure variability in the NP. Therefore, the jet-trapped Rossby waves captured in the KE mode can play an important role in coupled decadal variability of the KE and the atmosphere over the NP.
We thank the reviewers and the editor for their thoughtful comments and helpful suggestions to improve the manuscript. This study was supported by KIOST In-House Grant PE99601 and the project titled “Study on Air-Sea Interaction and Process of Rapidly Intensifying Typhoon in the Northwestern Pacific (2018)” funded by the Ministry of Oceans and Fisheries, Korea. H. Na acknowledges partial support from the F3 project of Hokkaido University. K.-Y. Kim acknowledges support from the National Research Foundation of Korea under Grant NRF-2017R1A2B4003930. S. Minobe is support by the Japan Society for the Promotion of Science (JSPS) KAKENHI Grants 26287110 and 26610146, and Y. N. Sasaki is supported by the JSPS KAKENHI Grant-in-Aid for Young Scientists (B) Grant 16K1780.