Coral Records of Temperature and Salinity in the Tropical Western Pacific Reveal Influence of the Pacific Decadal Oscillation Since the Late Nineteenth Century

The Pacific Decadal Oscillation (PDO) is a complex aggregate of different atmospheric and oceanographic forcings spanning the extratropical and tropical Pacific. The PDO has widespread climatic and societal impacts, thus understanding the processes contributing to PDO variability is critical. Distinguishing PDO‐related variability is particularly challenging in the tropical Pacific due to the dominance of the El Niño–Southern Oscillation and influence of anthropogenic warming signals. Century‐long western Pacific records of subannual sea surface temperature (SST) and sea surface salinity (SSS), derived from coral Sr/Ca and δ18O profiles, respectively, allow for evaluating different climatic sensitivities and identifying PDO‐related variability in the region. The summer Sr/Ca‐SST record provides evidence of a significant SST increase, likely tied to greenhouse gas emissions. Anthropogenic warming is masked in the winter Sr/Ca‐SST record by interannual to multidecadal scale changes driven by the East‐Asian Winter Monsoon and the PDO. Decadal climate variability during winter is strongly correlated to the PDO, in agreement with other PDO records in the region. The PDO also exerts influence on the SSS difference between the dry and wet season coral δ18O (δ18Oc)‐SSS records through water advection. The PDO and El Niño–Southern Oscillation constructively combine to enhance/reduce advection of saline Kuroshio waters at our site. Overall, we are able to demonstrate that climate records from a tropical reef environment significantly capture PDO variability and related changes over the period of a century. This implies that the tropical western Pacific is a key site in understanding multifrequency climate variability, including its impact on tropical climate at longer timescales.

More than just a single mode of climate variability, the PDO is presently considered a combination of extratropical and tropical atmospheric and oceanographic processes operating at varying timescales (Liu & Di Lorenzo, 2018;Newman et al., 2016). The strength of the Aleutian Low, persistent oceanic memory closely dependent on ENSO-SST anomalies, and the Kuroshio Current system act together to drive the spatial pattern of the PDO (Newman et al., 2016). Each PDO shift may then be a result of different combinations of the above processes (Deser et al., 2004;Newman et al., 2016), confounding our understanding of its climatic behavior. Adding more complexity in constraining decadal variability is anthropogenic warming, where the shift of the PDO in 1976-1977 from cold to warm phase is hypothesized to be linked to greenhouse gas emissions Graham, 1995;Miller et al., 1994;Nurhati et al., 2011). One approach to overcome these challenges is to document the different processes that contribute to PDO variability beyond the instrumental period and anthropogenic influence, allowing identification of dependent and independent forcing mechanisms of the PDO and how their interrelationships varied back in time (Newman et al., 2016).
Networks of tree-ring records from continental East Asia and northwestern America provide multicentury to millennial scale histories of PDO variability (e.g., Biondi et al., 2001;D'Arrigo & Wilson, 2006;Gedalof & Smith, 2001;MacDonald & Case, 2005;Minobe, 1997). However, terrestrial data do not fully represent oceanic processes linked to the PDO. Coral-based records documenting interannual to multidecadal SST patterns are concentrated in the central and south Pacific regions (e.g., Cobb et al., 2001;Linsley et al., 2000;Nurhati et al., 2011;Urban et al., 2000) but are extremely rare in the tropical and subtropical western Pacific (e.g., Asami et al., 2005;Felis et al., 2010). Most of these records are based on δ 18 O profiles, preventing independent estimates of SST anomalies in the region. Because the PDO is an aggregate of several processes and remotely influences the ENSO-dominated tropical region, combined records of Sr/Ca and δ 18 O are more advantageous to accurately estimate PDO-related climate variability.
A previous study using corals from Palaui, northeastern Luzon, Philippines (Ramos et al., 2017), shows that interspecies (i.e., averaged Diploastrea heliopora and Porites lobata) wintertime Sr/Ca-SST is sensitive to PDO variability, providing support for PDO reconstruction at longer timescales. It was also shown that the P. lobata dry (winter) season δ 18 O c -SSS record, compared with that of the wet (summer) season, reflects regional changes in salinity, allowing for a reconstruction of ENSO variability. In this paper, we present extended records of Sr/Ca and δ 18 O proxies derived from the same P. lobata core and reconstructed SST and SSS for the past~132 years using previously published P. lobata proxy calibrations (Ramos et al., 2017). We assess these records across different frequencies to examine their relative sensitivity to PDO variability and possible changes in these relationships through time. We also evaluate the potential impacts of the PDO on tropical western Pacific climate, which is considered to weakly correspond to decadal climate variability (Deser et al., 2004;Mantua & Hare, 2002;Newman et al., 2016).

Coral Sampling and Analytical Methods
In May 2012, a 3-m long coral core of Porites lobata was collected from 4-to 5-m water depth off the coast of Palaui Island, Philippines (18.5°N, 122.2°E; Figure 1). The core was drilled using a pneumatic drill with~55mm diameter core bit. Monthly instrumental SST (Integrated Global Ocean Services System, centered at 18.5°N, 122.5°E; Reynolds et al., 2002) and SSS (Simple Ocean Data Assimilation, centered at 18. 8°N, 122.3°E;Carton & Giese, 2008) records from 1982 to 2012 indicate that the warmest and wettest months occur from June to September while the coldest and driest months occur from December to March. Mean annual SST and SSS both exhibit limited variability (i.e., 1.4°C and 0.66 psu, respectively). Annual rainfall is on average 2,000 ± 500 mm/year (1σ; Philippine Atmospheric Geophysical and Astronomical Services Administration).
Palaui Island is located at the northeastern tip of Luzon facing the western Pacific Ocean to its east and the Luzon Strait to its north. It lies near the northern tip of the western Pacific warm pool where the Asian monsoon, ENSO and the PDO exert influences on the regional SST, salinity, and rainfall variability at seasonal to multidecadal timescales (e.g., Delcroix et al., 2007;H. Wang & Mehta, 2008;Yan et al., 1992).

Coral Subsampling
The coral core was cleaned, cut into 7-mm thick slabs, and X-rayed according to methods previously outlined (Ramos et al., 2017). Using a micrometer-controlled drill press, subsamples were drilled along the extending corallites at 0.5-mm increments and at constant sampling depth of 1 mm, yielding approximately biweekly sampling resolution. Each individual sample weighed <300 μg and 40-to 90-μg powder splits were set aside for stable isotope analysis. The remaining powder was used for Sr/Ca measurements.
Prior to 1894, δ 18 O was only measured on dry and wet season samples. The dry (wet) δ 18 O samples were selected on the basis of the winter (summer)-time Sr/Ca depth-and time-series data, in which δ 18 O and Sr/Ca maxima (minima) coincide seasonally. Each season constitutes about four to eight consecutive δ 18 O samples, representing the dry and wet season values of each year. The reduction in sample points and the difference in instrumentation are unlikely to create a bias across time periods as we consistently examine seasonal extremes at interannual to multidecadal timescales. Time assignments for the interannual dry and wet season δ 18 O followed that of the paired Sr/Ca data.

Chronology Development
Age-depth models were developed using the annual density bands (supporting information Figure S1 in the supporting information) and fine-tuned using the seasonal cycles of Sr/Ca. Beyond the instrumental and calibration period (i.e., 1982 to 2012) previously reported (Ramos et al., 2017), ages were assigned using the averaged Integrated Global Ocean Services System SST climatology, aligning Sr/Ca profile minima, maxima, and inflection points with their respective SST points each year using the Analyseries software (Paillard et al., 1996). The Sr/Ca time series was then linearly interpolated into a monthly record. The rough Sr/Ca-SST age-depth model was similarly applied to the corresponding δ 18 O record, which was then interpolated at monthly resolution.

SST and SSS Reconstruction
Records of SST and SSS variability extending back to 1880 were reconstructed using previously published calibration equations for SST:  (Ramos et al., 2017), where RMSR is the root-mean-square of the residual describing the difference between the instrumental and reconstructed records. These equations were applied to interannual reconstructions (i.e., summer/3-year wet and winter/3-year dry SST and SSS) to effectively increase the signal to noise ratio that may otherwise be dampened by the limited SST and SSS variability at our site (Ramos et al., 2017).

Data Analysis
We estimated the power spectral density on our records and climate indices using a simple fast Fourier transform (FFT) algorithm. Significance level (90%) was estimated using 10,000 Monte Carlo simulations. We also performed coherence and phase analysis using a nonparametric multitaper method (MTM) with bias correction (https://www.mathworks.com/matlabcentral/fileexchange/22551-multi-taper-coherencemethod-with-bias-correction; Huybers & Denton, 2008) to evaluate the significant frequencies shared by our reconstructions and climate indices. MTM coherence was estimated with two windows and 50 iterations on Monte Carlo error estimates. To further isolate and observe variability specific to coherent frequencies between records, we applied a Gaussian filter using the Analyseries software (Paillard et al., 1996).

Sr/Ca-SST Reconstruction
The Sr/Ca record shows distinct annual cycles throughout the reconstruction period extending back to 1880 (Data Set S1 and Figure S2 in the supporting information). The summer and winter SST reconstructions exhibit interannual to multidecadal scale variability over a 3.4 and 3.1°C SST range, respectively ( Figure 2a). Both interannual reconstructions show similar trends at the beginning of the record, indicated by their significant positive 15-year running correlation ( Figure 2b). Around the mid-1920s, the records decouple, with negligible relationships recorded during 1925-1935, 1947-1974, 1979-1983, and 1994-2012. This change in decadal behavior between records through time is confirmed by calculating the 50-year running variance for each record (Figure 2c). Summer SST exhibits a gradual decrease in variance that reverses in direction in the early 1970s. Winter SST shows similar variance to that of the summer record until the late 1920s when it began to steeply increase, reaching a peak around 1950 ( Figure 2c). The winter SST variance subsequently decreased until the mid-1960s when variance increased to the previous magnitude ( Figure 2c). There is no discernible long-term trend in either SST record (m summerSST = 0.0029°C/year and m winterSST = 0.0014°C /year; r < 0.2, p < 0.05), indicating minimal mean SST change over the past century. However, from 1970 to present, the summer record shows an SST increase of 0.03°C/year or 1.23°C over the past~4 decades (r = 0.66, p < 0.0001), indicating significant recent summer warming.

Evaluating Drivers of SST Variability
The PDO has previously been shown to exert an influence on Palaui winter SST variability, with warm (positive) PDO events characterized by anomalously cool SSTs in the western Pacific and central North Pacific Ocean (Felis et al., 2010;Gedalof & Smith, 2001;Mantua et al., 1997;Ramos et al., 2017). The extended winter SST record reported here exhibits significant correlation with the PDO at both interannual (r = −0.34, p < 0.0001, n = 112) and 3-year binned timescales (r = −0.55, p < 0.001, n = 38) from 1880 to 2012. Compared with other modes of Pacific decadal variability (i.e., Interdecadal Pacific Oscillation, IPO (Power et al., 1999), and North Pacific Index, NPI (Trenberth & Hurrell, 1994)), our winter SST record is weakly correlated (r IPO = −0.01 and r NPI = 0.10, p >0.30). This suggests that interannual to multidecadal SST changes in our winter SST reconstruction may strongly be linked to the PDO.
To further evaluate the robustness of this relationship and quantify the significant frequencies shared between our coral records and climate indices, we generated MTM power spectral densities and coherence between our summer and winter SST reconstructions and the PDO index over the same time period between 1901 and 2010. We similarly performed this analysis using solar irradiance (Lean, 2000), ENSO (i.e., Niño 3.4) and EAWM (D'Arrigo et al., 2005) indices over the same time period for comparison to the SST records.
Results of the FFT spectral analysis reveal significant variance at multidecadal periods of 55 years/cycle and interannual periods of 3.1-3.2 and 2.1-2.3 years/cycle in the winter SST record and the PDO index (Figures 3a and 3b.1). Further evidence that our record is sensitive to the PDO is spectral coherence (>90%) at the previously mentioned frequencies and at periods of~10 years/cycle (Figure 3b.2). Between the winter SST and the EAWM, the records show significant power and coherence at a period of 2.8 years/cycle (Figure 3c.2), indicating the additional influence of the EAWM on the winter SST variability at high frequencies. The winter SST also reveals significant spectral power at interannual periods of 2.6 and 3.7 years/cycle (Figure 3a) but is incoherent with the Niño 3.4 index and solar irradiance (supporting information Figure S2).
Summer SST, on the other hand, exhibits significant power at 2.3, 2.5, 3.7, and 5.2 years/cycle (supporting information Figure S3). Compared with the PDO, EAWM, and Niño 3.4 indices and solar irradiance (supporting information Figure S3), summer SST is incoherent with these records at the previously mentioned time frequencies, indicating that the impact of these climatic forces on the summer SST record is limited or absent. Therefore, we further investigate the drivers of multifrequency SST variability focusing on the winter SST reconstruction and the PDO and EAWM indices.

Palaui Winter SST Variability
The winter SST record and EAWM indices were filtered to 0.359 ± 0.0045 cycles/year (2.8 year/cycle). The SST and PDO records were filtered to 0.018 ± 0.0045 (55 years/cycle), 0.105 ± 0.0045 (10 years/cycle) and 0.473 ± 0.0045 (2.1 years/cycle) cycles/year. The 2.8-year filtered winter SST and EAWM records show a significant negative correlation (r = −0.43, p < 0.0001; Figures 4a and 4b), consistent with regional cooling during strong EAWM periods. The relatively low correlation strength between records is likely due to the change in sign through time, only showing significant negative relationships around the mid-1960s toward the present (Figure 4b). The observed gradual increase in amplitude in the EAWM record reaching maximum values in the 1960s is not evident in our winter SST reconstruction but may explain the change in phase and correlation strength between records, which occurred about the same time.
The 2.1-year filtered winter SST and PDO records show a significant but low negative correlation (r = −0.30, p = 0.002; Figures 4c and 4d), likely resulting from the observed change in phase over time. Changes in phase are not observed in the 10-and 55-year filtered records, which in contrast exhibit strong antiphase relationships (r 10-year = −0.96 and r 55-year = −0.90, p < 0.0001; Figures 5a and 5b), consistent with the known spatial structure of the PDO-related SST anomalies across the Pacific. The 55-year filtered timeseries reveal consistent amplitudes throughout the records (Figure 5a), whereas in the 10-year filtered timeseries, the amplitudes were attenuated between 1920 and 1940 and became gradually amplified toward the present (Figure 5b), suggesting a more active PDO in recent years. On the basis of these results, we applied a 10-year moving average on the winter SST record to reconstruct the winter PDO (Figure 5c). Compared with the similarly averaged PDO index, the records show a strong negative relationship (r = −0.75, p < 0.0001, n = 101; Figure 5c), highlighting that SST variability at this site reliably records the PDO at decadal scale.
Our winter SST record exhibits known regime shifts since the start of the twentieth century: shifts from cold to warm phase in 1924 and 1976 and shifts from warm to cold phase in 1945 and 1998 (Figure 5c). Beyond 1900, when the PDO index is no longer available, we find good agreement (r = 0.77, p <0.0001) in the SST fluctuations between our record and the only available coral Sr/Ca-based PDO reconstruction in the western Pacific, Ogasawara coral from southern Japan (Felis et al., 2010;Figure 5d). Both coral records exhibit warm winter SST anomalies between 1885 and 1895 (Figures 5c and 5d), indicative of a cold (negative) PDO phase during this period. While the climatological winter SST averages were slightly different for Palaui and Ogasawara (i.e., December to March and November to February, respectively), the two coral records exhibit good overall reproducibility (r = 0.50, p < 0.0001, n = 106) over 10-year timescales, suggesting that the two proxy reconstructions from different geographical locations are responding to the same decadal forcing. Since the late nineteenth century, the dry season SSS record exhibits a long-term decreasing trend of 0.003 psu/year (r 2 = 0.52, p < 0.0001; Figure 6a), indicative of freshening. Wet season SSS shows no observable long-term trend (slope, m wet(1880-2012) = −0.0020 psu/year), but a "drier" anomaly is observed during the late nineteenth century and during a prolonged period between 1950 and 1980. Around 1976 to present, the wet season SSS record decreased to similar magnitude prior to the prolonged drier period, indicative of recent freshening.

Dry and Wet season δ 18 O c -SSS Reconstruction
While the seasonal salinity variations appear independent, there are periods when the salinity values in both seasons are statistically equivalent. Between 1958 and 1976, the salinity difference between the dry and wet season SSS records is small to negligible (<0.15 psu; Figure 6b), indicating the loss of SSS seasonality. Before and after this time period, the salinity difference is greater than 0.15psu, reaching a maximum of 0.72psu in the early 1930s (Figure 6b). The similarity in SSS values between 1958 and 1976 could be due to seasonal excursions resulting from offsets in timing between ENSO or local rainfall at our site as previously reported (Ramos et al., 2017). However, at longer timescales the SSS difference shows distinct decadal variability that we hypothesize is additionally controlled by large-scale processes.

Evaluating Drivers of SSS Variability
To evaluate the controlling factors in the seasonal SSS reconstructions at longer time periods, we compared the SSS difference to the PDO index. The PDO is significantly correlated to the dry-wet season SSS difference (r = 0.39, p = 0.02, n = 37; Figure 6b) such that the period of small SSS seasonality coincides with the predominant cold (negative) PDO phase, while periods of larger SSS differences coincide with the predominant warm (positive) PDO phases. The correlation becomes stronger and more significant with application of a 3-and 6-year lag on the PDO time series (r 3yr_lag = 0.57, p < 0.0001, n = 36 and r 6yr_lag = 0.72, p < 0.0001, n = 35), providing further support that the PDO exerts influence on the SSS variability at our site. Results of the spectral analysis reveal significant power at periods of 6.8 and 6.4 years/cycle in the SSS difference and PDO records, respectively (Figure 7a). In addition, both records are significantly coherent at periods >27 years/cycle (Figure 7b). Because we are interested in low frequency variability, we filtered our data to 0.023 ± 0.014 cycle/year. The >27-year filtered records show a strong positive relationship, despite the PDO leading by 6 years (r = 0.70, p < 0.0001, n = 37; Figure 7c). Accounting for this lead results in a stronger positive relationship (r = 0.96, p < 0.0001, n =35), suggesting a significant PDO-related control on salinity variability at longer timescales.

Climatic Controls on SST Variability
Both the summer and winter Sr/Ca-SST records capture SST variability significantly, each providing records of different SST behaviors. Our SST reconstructions are significantly correlated at the beginning of the records until the mid-1940s (Figure 2b), suggesting that seasonal SST changes are governed by similar factors. However, toward the present, the summer SST record indicates a significant warming trend that started in the 1970s (Figures 2a and 8a), whereas the winter SST record shows a different response (Figure 2a). Beginning in the 1970s, both records exhibits increasing variance (Figure 2c). However, the winter SST variance is consistent with decadal scale changes in variance throughout the record, whereas summer SST variance only had a decreasing trend earlier in the record before 1970. The onset of increasing summer SSTs (Figure 8a) is coincident with the onset of regional warming (e.g., Tierney et al., 2015 and references therein) likely tied to increasing anthropogenic greenhouse gas emissions during this period. A record of well-mixed greenhouse gas-radiative forcing (Schmidt et al., 2012) indicates that its magnitude over the most recent 50 years increased thrice as fast compared to the preceding 50 years (i.e., 1950 to 2000: 0.0270 W m −2 year −1 [r 2 = 0.99, p < 0.0001] vs. 1900 to 1950: 0.0094 W m −2 year −1 [r 2 = 0.99, p <0.0001]; Figure 8b). However, compared to the magnitude of a 1°by 1°gridded instrumental summer SST record at our site (i.e., 0.17°C per decade; Rayner et al., 2003), our warming estimate is twice as high. This disagreement in magnitude may be due to differences in spatial coverage of the instrumental and proxy records where the latter better represents summer SST variability within a shallow reef environment. The presence of anthropogenic warming only in the more recent summer records, and not in the winter (as discussed below), may indicate the power of decadal scale variability in masking or slowing down anthropogenic signals in our SST records. One possible implication for this is that nonwarming winters may lessen the impacts of increased thermal stress in warming seas.
Winter SST exhibits greater sensitivity to the PDO-and EAWM-related SST changes. Interannual to multidecadal scale variability (i.e., 50 to 70 years) is a known feature of climatic oscillations in the North Pacific during the eighteenth and nineteenth centuries (Mantua & Hare, 2002). Multidecadal scale oscillations are present in coral records from Ogasawara (Felis et al., 2010) and Koshiki (Watanabe et al., 2014) islands in Japan, reflecting PDO variability in their SST (Sr/Ca) and summer monsoon (δ 18 O) reconstructions, respectively. Other marine-based North Pacific SST reconstructions including those extracted from coralline algae proxies (e.g., δ 18 O (Halfar et al., 2007) and Mg/Ca (Williams et al., 2017)) and geoduck clam growth indices (e.g., Black et al., 2009;Strom et al., 2004;supporting information Figure S6) also exhibit multidecadal scale variability. Tree-ring climate reconstructions from continental East Asia, west Canada, and northwest America (e.g., Black et al., 2009;D'Arrigo & Wilson, 2006;MacDonald & Case, 2005;Minobe, 1997) also exhibit strong periodicities of 50 to 70 years. Causes of such multidecadal scale variability are still unclear (MacDonald & Case, 2005;Mantua & Hare, 2002) but are likely forced by ocean-atmosphere interaction and modulated by solar radiation (Minobe, 1997). To compare the spatiotemporal behavior of the PDO, we applied a 55-year Gaussian filter (0.018 ± 0.0045 cycle/year) on the above PDO reconstructions (excluding the non-SST-related record from Koshiki). Our winter SST record shows strong correlations with the PDO index and both marine-and terrestrial-based reconstructions across the Pacific, showing the expected phase relationship (Table 1 and Figure 9). Significant but relatively weak correlations, on the other hand, are observed between the western and northern Pacific records and the strongly correlated records from northwest America and Canada (Table 1 and Figure 9). Due to the clustering of strong relationships between different proxies within a geographic area, but reduced correlations across regions, a change in the spatial pattern of the PDO over time is a likely cause. Because the PDO is a sum of several atmospheric and oceanic processes that are spatiotemporally dynamic (Newman et al., 2016), the observed inconsistency between regions can be expected as proxy sensitivity changes with location (D'Arrigo & Wilson, 2006;Newman et al., 2016). Reduced correlations to the PDO may also lead to increased proxy sensitivity to other local climatic or environmental influences such as ENSO, monsoon failures, volcanic eruptions, storm activity, and other extreme events, resulting in disagreement between records (e.g., D' Arrigo & Wilson, 2006;Williams et al., 2017). Nonetheless, good reproducibility between proxies within each region is valuable for understanding the processes that contribute to PDO variability.
Interannual oscillations observed in the western Pacific PDO records likely originate from ENSO signals known to imprint PDO variability (D'Arrigo & Wilson, 2006;Felis et al., 2010;Watanabe et al., 2014). However, coherence between our SST records and Niño 3.4 is absent. The interannual frequency of the EAWM is evident in our record but changes in its phase and amplitude are not reflected in our reconstruction, resulting in an obscured winter SST and EAWM relationship. This may be due to the strong multidecadal component driving our SST variability and masking the interannual signals. Comparing the high-frequency filtered winter SST, PDO, and EAWM records shows that the period of negligible relationship between winter SST and PDO coincides with the period when the EAWM influence on winter SST is more significant (Figure 4). Because the PDO and EAWM impact SSTs at our site in the same direction (i.e., positive PDO and positive EAWM leads to cooler SST anomalies at our site), it can be hypothesized that at different time periods, these climate systems may show dominance over the other in modulating SST variability at higher frequencies. Longer records are needed to verify these changes and formulate mechanisms describing the inter-relationship between the PDO and EAWM back in time.

Climate Controls of SSS Variability
The PDO impacts surface salinity at our site by modifying the hydrological balance (i.e., evaporation/precipitation) and/or water advection (Delcroix et al., 2007). During warm (cold) PDO phases, in situ salinity observations over the tropical Pacific indicate higher (lower) salinity values in the western Pacific due to decreased (increased) rainfall (Delcroix et al., 2007). However, we found the opposite relationship between the PDO and salinity reconstructions during the wet season at our site when maximum annual rainfall occurs. This suggests that the PDO dominantly impacts the seasonal SSS difference through water advection instead of precipitation. The Kuroshio current, a northward flowing western boundary current in the Pacific, transports high salinity western Pacific water to our site as it impinges on the coast and intrudes into the Luzon Strait (Hu et al., 2015;Nan et al., 2015;Qu et al., 1998). During warm (positive) PDO phases, the Kuroshio transport is slower, increasing Kuroshio intrusion into the Luzon Strait (Wu, 2013) and advecting more saline western Pacific waters to Palaui. Similarly, El Niño/La Niña events enhance/reduce Kuroshio intrusion into the strait (Qu et al., 2004;Ramos et al., 2019). Changes in Kuroshio transport are more pronounced during the winter (dry) season when ENSO events mature and prominent PDOs are detected (Deser et al., 2004;Trenberth et al., 1998). Therefore, the significant difference between the dry and wet season salinity records during predominantly positive PDO phases is likely due to the advection of high salinity waters to our site in the dry season ( Figure 6b).
Between 1958 and 1976 when SSS difference is negligible, the dry season SSS reconstruction indicates predominance of La Niña events as previously reported (Ramos et al., 2017). La Niña-related increases in rainfall and reduced Kuroshio intrusions positively combine to decrease dry season salinity at Palaui. This period also coincides with the predominant cold PDO phase (Figure 6b), similarly reducing the Kuroshio intrusion and hindering advection of more saline waters to our site. Thus, the predominant cold PDO phase between 1958 and 1976, combined with ENSO-related freshening, lead to insignificant changes in salinity between seasons.
Overall, salinity variability at our site implies interaction between the PDO and ENSO. Warm/cold PDO and El Niño/La Niña phases constructively combine to enhance/reduce advection of more saline waters at our site. Because the PDO is viewed as a "reddened" ENSO reaction (Newman et al., 2016) where warm/cold phases are analogous to El Niño/La Niña episodes, salinity as a function of PDO-related Kuroshio water advection shows a response consistent with the   Note.
All numbers are significant at 95% level except those italicized. Number superscripts indicate the location of these records in supporting information Figure S6.
Kuroshio transport and drought/above-normal rainfall anomalies associated with El Niño/La Niña events. This suggests that the mechanisms driving ENSO-related salinity changes at our site similarly contribute to the PDO-related SSS variability.
We also compared our salinity records with previously published SSS reconstructions in close proximity to our study site-Ogasawara Island (Felis et al., 2009) in the Western Pacific and Hainan Island (Deng et al., 2013) in the northern South China Sea. These records show significant freshening trends since the late nineteenth century, though with variable magnitude and timing. In Ogasawara, the abrupt freshening in the early 1990s by as much as 0.35‰ or 0.82psu is attributed to a combined forcing of atmospheric changes and oceanic advection (Felis et al., 2009). Weakened westerlies together with reduced Kuroshio transport lead to pronounced freshening during this period. In contrast, a gradual freshening trend of~0.4‰ since 1900 is observed in Hainan, northern South China Sea, which is attributed to PDO-related changes in rainfall over South China and terrestrial runoff (Deng et al., 2013). In Palaui, our dry season salinity record shows a similar freshening trend of 0.4‰ (0.39 psu) since the late 1800s. While the PDO impact on Hainan and Palaui records is evident, the mechanisms by which the PDO affects climate patterns at these sites are variable even within the same regional geographic location. Thus, isolating the PDO's role in the overall freshening trend remains complex. This is expected due to the numerous atmospheric and oceanic processes involved in the PDO; hence, distinguishing PDO-related impacts is critical. In Palaui, reconstructing salinity variability in different seasons presented a unique opportunity to distinguish the PDO's specific influence on changing salinity through time.

Conclusions
Century-long records of Sr/Ca and δ 18 O from Palaui allow reconstruction of SST and SSS variability and evaluation of their climatic drivers. Our SST records capture significant variability, with the summer and winter records exhibiting different climate sensitivities. The summer SST record provides evidence of significant warming that commenced in the 1970s. This warming is likely tied to the rise in western Pacific SSTs associated with greenhouse warming. Our winter SST, on the other hand, is sensitive to interannual to multidecadal scale changes driven by the EAWM and the PDO, which potentially slows down the effects of anthropogenic warming. Similar to other terrestrial and marine-based PDO records across the Pacific, our record exhibits decadal climate variability strongly correlated to the PDO. Comparing PDO records from the western, northern, and eastern Pacific yields reduced correlation across basins, which may indicate a change in the spatial pattern of the PDO. Because the PDO and EAWM affect SST at our site in a similar direction, the dominance of PDO-related changes in our winter SST record obscures the monsoonal impact. However, these interrelationships through time may provide clues on the reversing roles of the PDO and EAWM in modulating winter SSTs at high frequencies.
Reconstructing salinity variability at different seasons similarly allowed for identifying distinct salinity controls between the wet and dry seasons. Both ENSO and non-ENSO-related rainfall patterns are evident in our salinity records. By determining the seasonal salinity difference through time, the PDO's control on salinity in this region is additionally revealed. The PDO and ENSO constructively combine to enhance/reduce advection of more saline Kuroshio waters to our site.
Overall, this study demonstrates that climate records from a tropical reef environment like Palaui robustly capture PDO variability and related changes over a century. The strong imprint of decadal variability potentially masks the effects of anthropogenic warming and maintains a stable-appearing long-term mean SST. This also implies that the tropical western Pacific is a key region for understanding multi-frequency climate variability throughout the Pacific basin, including changing patterns of impact on tropical climate at longer timescales.