Corals Evidence an Underestimation of the 20th Century Warming in the Eastern Pacific Cold Tongue

The trade winds cause strong upwelling in the eastern equatorial Pacific, and create the eastern Pacific Cold Tongue (EPCT) that has far‐reaching impacts on global climate. However, large discrepancies persist in quantifying 20th‐century EPCT sea surface temperature (SST) changes across different instrumental data sets. Here we synthesize four coral Sr/Ca‐SST records from the tropical central‐eastern Pacific to develop a Cold Tongue Index (CTI) reconstruction for 1887–1997. The coral CTI record shows a rapid 20th century warming of the EPCT, suggesting an underestimation of warming trends in instrumental CTI records. The decadal to multidecadal changes in reconstructed EPCT SST show an association with the Walker Circulation. Our reconstruction indicates that recent EPCT cooling during the global warming hiatus is not unusual in the context of the 20th century. Our results provide new evidence for 20th century EPCT SST changes and an observational constraint for predicting future tropical climate changes.


Introduction
The tropical Pacific Ocean is known for its distinct pattern of sea surface temperature (SST) distribution around the equator, featuring a warm pool in the western Pacific and a cold tongue in the eastern Pacific (Bjerknes, 1969;Cai et al., 2021;Philander, 1990).The eastern Pacific Cold Tongue (EPCT) extends from the coast of the Americas into the central Pacific (6°N-6°S, 180°W-90°W), exerting influence on global climate through its impact on atmospheric heat and moisture transport (Cai et al., 2021;Trenberth, 1997).The interannual variability of the EPCT SST has been widely utilized as an index for the El Niño-Southern Oscillation (ENSO) activity (Trenberth & Caron, 2000).The strong upwelling of cold water in the tropical Pacific is driven by easterly trade winds, and it regulates heat distribution within the tropical Pacific (Bjerknes, 1969;Cai et al., 2021;Philander, 1990).The EPCT region is characterized by a strong atmospheric subsidence which modulates global circulation patterns and the position of the intertropical convergence zone (Koutavas & Lynch-Stieglitz, 2004).As the decadal cooling of the EPCT in the early 21st century has been suggested to be a driver of the global warming hiatus (England et al., 2014;Kosaka & Xie, 2013, 2016;Power et al., 2021), a great need exists for an improved understanding of the long-term variations of EPCT SST in order to enhance climate predictions.However, considerable uncertainties persist regarding changes of the EPCT SST and their cause during the 20th century.
Observations from the HadISST and Kaplan SST (v2) data sets indicate a long-term cooling trend, while the ERSST (v5) and COBE SST (v1) data sets suggest a 20th century warming (Text S3 and Figure S1 in Supporting Information S1) (Bunge & Clarke, 2009;Cane et al., 1997;Deser et al., 2010;Solomon & Newman, 2012).These discrepancies are attributable to data scarcity prior to the ∼1960 s, changing measurement techniques and differences in analytical procedures to interpolate missing or doubtful data in gapping time series (Bunge & Clarke, 2009;Deser et al., 2010).In addition, model simulations show contradictory responses of EPCT SST to external radiative forcing (Cane et al., 1997;Clement et al., 1996;Held & Soden, 2006;Seager et al., 2019;Vecchi et al., 2006).The "ocean dynamical thermostat" model predicts a cooler EPCT in response to positive radiative forcing (Cane et al., 1997;Clement et al., 1996).Eastern upwelling brings cold waters up and counteracts the warming tendency from anthropogenic global warming, so the positive radiative forcing initially warms the western Pacific more than the EPCT (Cane et al., 1997;Clement et al., 1996).The enhanced zonal SST gradient across the tropical Pacific strengthens the equatorial trade winds and drives a cooler EPCT through the Bjerknes feedback (Cane et al., 1997;Clement et al., 1996).In contrast, another theory suggests that global warming increases water vapor in the low troposphere and weakens the Walker Circulation, which has been predicted in numerous coupled general circulation models (CGCMs) (Held & Soden, 2006;Vecchi et al., 2006).This would lead to a weakening of the upwelling in the eastern Pacific via the Bjerknes feedback and leads to rapid warming of the EPCT (Held & Soden, 2006;Vecchi et al., 2006).Given these large discrepancies in both observations and models, high-resolution reconstructions of EPCT SST are urgently needed to improve our understanding of tropical climate change over the 20th century.
Corals, as monthly-resolved and (multi-)century-long proxies, are a unique archive for reconstructing natural climate variability in tropical oceans (Corrège, 2006).Estimating the 20th century trend in EPCT SST using coral δ 18 O records shows large uncertainties due to the combined influence of thermal and hydrological changes on coral δ 18 O (Thompson et al., 2011;Tierney et al., 2015).In particular, a notable overestimation of the warming trend in the coral δ 18 O records is thought to have occurred in the post-1960 period, when instrumental SST data become more reliable (Tierney et al., 2015).Coral Sr/Ca is thought to only respond to changes in SST rather than the combined influence of SST and seawater δ 18 O (δ 18 O sw ) (Beck et al., 1992), and has been used to reconstruct SST variability in the tropical central-eastern Pacific (Carilli et al., 2014;Jimenez et al., 2018;Nurhati et al., 2011;Wu et al., 2014).For example, Jimenez et al. (2018) conducted a trend analysis on the coral Sr/Ca records from the tropical central-eastern Pacific and found a warming trend over the 20th century.Nonetheless, there are substantial differences in the magnitude of the warming among different records (∼0.48-1.76°C/100yr) (Carilli et al., 2014;Jimenez et al., 2018;Nurhati et al., 2011;Wu et al., 2014), leaving large uncertainties in the assessment of 20th century warming trend over the EPCT region (Alpert et al., 2016;Karnauskas et al., 2015).
Here we synthesize four published coral Sr/Ca-SST records from the tropical central-eastern Pacific (Table S1 in Supporting Information S1) using a nested principal component regression (PCR) method to reconstruct EPCT SST changes over the period 1887-1997 (Carilli et al., 2014;Jimenez et al., 2018;Nurhati et al., 2011;Wu et al., 2014).The reconstruction indicates a rapid warming trend over the 20th century, and points to an underestimation of warming trends in instrumental CTI records.

Data and Methods
In this study, four published coral Sr/Ca-SST records from the tropical central-eastern Pacific were selected and subsampled to a seasonal resolution (DJF, MAM, JJA, SON) (Figure 1; Text S1 and Table S1 in Supporting Geophysical Research Letters 10.1029/2024GL108954 Information S1) (Carilli et al., 2014;Jimenez et al., 2018;Nurhati et al., 2011;Wu et al., 2014).The CTI used as the instrumental target of our reconstruction was calculated as the area-averaged SST anomalies over the equatorial central-eastern Pacific (6°N-6°S, 180°W-90°W) (Text S4 in Supporting Information S1) (Deser & Wallace, 1990).The post-1960 mean of the CTI (CTI M ) derived from the ERSST v5 (2°× 2°), HadISST (1°× 1°), Kaplan v2 (5°× 5°) and COBE v1 (1°× 1°) data sets (CTI ERSST , CTI HadISST , CTI Kaplan and CTI COBE ) is used as instrumental target (Texts S3, S4 and Figure S4 in Supporting Information S1).These four instrumental SST data sets differ in their analysis procedures, data sources, and bias corrections (Texts S3 in Supporting Information S1) (Deser et al., 2010;Vecchi & Soden, 2007).The CTI was also constructed from the ICOADS SST (only bucketsampling; CTI bucket ) to further validate the post-1960 EPCT SST changes (Texts S3, S4 and Figure S5 in Supporting Information S1) (Tokinaga et al., 2012).The seasonally resolved coral CTI (CTI coral ) record was computed with a nested PCR procedure based on the selected coral records (Text S5 in Supporting Information S1) (Cook et al., 2002;Wang et al., 2017).PCR is applied in a stepped nested fashion to allow the reconstruction to be extended back in time as shorter coral records drop out of the model.The skills of the reconstruction were assessed using calibration-verification statistics, computed 1000 times by a bootstrap procedure (Text S5 in Supporting Information S1) (Stoffel et al., 2015).Details of this section can be found in Supporting Information S1.

EPCT SST Signals in Coral Sr/Ca Records
To ensure that the selected records capture large-scale climate signals, we correlated the coral Sr/Ca-SST records with instrumental ENSO indices since 1960.Significant correlations (with p value adjusted for autocorrelation p adj < 0.05; Text S8 in Supporting Information S1) show that all the coral Sr/Ca-SST records are sensitive to ENSO variability (Table S3 in Supporting Information S1).Spatial correlation fields with OISST further demonstrate that the coral records reflect SST changes at local-to-regional scales (Figures 1a-1d) (Reynolds  (Reynolds et al., 2007).(a) Coral record from Butaritari Atoll (BA) (Carilli et al., 2014).(b) Coral record from Palmyra Island (PI) (Nurhati et al., 2011).(c) Coral record from Clipperton Atoll (CA) (Wu et al., 2014).(d) Coral record from Wolf Island (WI) (Jimenez et al., 2018).(e) The instrumental CTI M (Deser & Wallace, 1990).(f) The PC1 of coral records.The stippling indicates correlations are significant at the 95% confidence level.All records were converted to anomalies prior to analysis.The black circles indicate the location of coral Sr/Ca-SST records, the rectangles frame the area of instrumental CTI.(g) Comparison of coral PC1 (black) with CTI M (red) and other ENSO indices (other colored lines; Table S3 in Supporting Information S1).(h) Interannual (black) and decadal (11-year lowpass filtered, blue) variations of CTI coral record.The CTI coral was calibrated/validated against the CTI M (red) over the period 1960-1992 using a nested PCR approach.Gray shading shows the uncertainties of CTI coral (±1 RMSE).(i) Calibration R 2 (red), verification R 2 (blue), the reduction of error (RE, green) and the coefficient of efficiency (CE, purple) computed for each nest.( j) Number of coral records included in each nest.The blue shadings highlight the unreliable nests (i.e., nests with the lowest replication, weak statistical skills and large discrepancies with CTI M ; Figure S18 and Table S4 in Supporting Information S1).et al., 2007).Spatial correlation patterns with SST across the tropical Pacific correspond to the SST pattern induced by ENSO (Cai et al., 2021).Yet, obvious differences exist between the coral records attributed to local factors (Figure 1).That is, the Butaritari Atoll and Palmyra Island records are best correlated with SST in the central Pacific (Carilli et al., 2014;Nurhati et al., 2011), whereas the Clipperton Atoll and Wolf Island records reflect SST signals in the tropical northeastern Pacific and the equatorial eastern Pacific, respectively (Figures 1a-1d) (Jimenez et al., 2018;Wu et al., 2014).
To reduce the influence of local factors while isolating the large-scale climate signal, we extracted the common signal of all coral records using a principal component analysis.The first principal component (PC1) of the coral records is significantly correlated with instrumental CTI M (r = 0.72, p adj < 0.01; Figure 1g; Table S3 in Supporting Information S1).Similarities between the spatial correlation patterns of the CTI M (Figure 1e) and PC1 (Figure 1f) with the OISST data set confirm the strong link between PC1 and EPCT SST variability.This means that the combination of individual coral records can isolate the large-scale climate signal and allow the development of a more robust index representing EPCT SST variability.
To this end, we developed a CTI coral record using a nested PCR approach in which a nest is created as the number of available coral record changes over time (Text S5 in Supporting Information S1).The regression models between the CTI M for 1960-1992 and principal components of the coral network were calculated for each of the nested subset of variables and the CTI coral record combines PCR results of the seven nests (Figure 1h; Figures S18a and S8b in Supporting Information S1).Depending on the nest, the shared variance between CTI coral and CTI M is ∼13%-64% for the period 1875-2010, as assessed using the calibration statistics in the nested PCR procedure (Figure 1i; Figure S18 and Table S4 in Supporting Information S1).And, on average, more than ∼56% of the shared variance was observed between 1887 and 1997 in nests 3-6 (Figure 1i; Figure S18 and Table S4 in Supporting Information S1).Consistent variations between the CTI coral reconstruction and instrumental CTI M at annual resolution and for different seasons, indicate that our reconstruction is free of seasonal biases for the post-1960 period (Figure S7 in Supporting Information S1).To further ensure the robustness of our reconstruction, we performed sensitivity tests by producing alternative reconstructions with different reconstruction methods, different instrumental data sets (ERSST, HadISST, Kaplan and COBE) as targets, different number of retained principal components (accounting for 70%-90% of the variance), different calibration periods, and without using a nesting approach (a fixed-nest reconstruction).At the level of the ensemble median, we find high similarity between alternative reconstructions and the final reconstruction (Figures S8-S16 in Supporting Information S1).
Considering that the reconstruction skills were estimated during the post-1960 period, a pseudo-proxy experiment was performed to further assess the methodological robustness of the CTI coral record over the 20th century (Figure S19 and Text S6 in Supporting Information S1).High consistency was observed between the pseudoreconstructions and the instrumental CTI for the period 1887-1997 (Figure S19 and Text S6 in Supporting Information S1).Specifically, the pseudo-reconstructions are consistent with the instrumental CTI even prior to ∼1940 when the (pseudo-)reconstruction relies solely on two records from Palmyra and Clipperton (Figure S19 and Text S6 in Supporting Information S1).These results indicate that changes in the number of coral records have a limited influence on the overall results, and thus further confirm the robustness of our reconstruction.

Rapid EPCT Warming Over the 20th Century
Based on these results, we argue that the CTI coral record provides a reliable basis for investigating EPCT SST variations over the course of the 20th century.The newly developed reconstruction shows a significant warming (0.62 ± 0.16°C/100 yr, p adj < 0.05) of the EPCT SST between 1887 and 1997.For the period 1960-1997, the CTI coral and instrumental CTI show consistent decadal variations and warming trends ranging between 1.36 and 1.96 ± 0.95°C/100 yr (p adj < 0.05) (Figures 2a-2d; Figure S20 in Supporting Information S1).Higher consistencies are found with the CTI coral , CTI ERSST and CTI bucket records, which has smaller biases related to SST measurement techniques (Figure S20 in Supporting Information S1) (Tokinaga et al., 2012).Larger proxyinstrument and inter-instrument discrepancies are found before the ∼1960s (Figure 2).In fact, the CTI coral record suggests cooler conditions for the period ∼1910-1935, resulting in a larger overall 20th century warming trend than found in the instrumental CTI records.The fixed-nest reconstruction and pseudo-proxy experiment support the reliability of CTI coral record during that period, even though only two coral records are available (Figures S16 and S19 in Supporting Information S1).However, the assessment of CTI warming trends over the period 1887-1997 based on instrumental data sets shows large discrepancies across observational data products (Figures 2a-2d; Figure S4 in Supporting Information S1).The CTI ERSST (0.47 ± 0.16°C/100 yr (p adj < 0.05)) and CTI COBE records (0.34 ± 0.14°C/100 yr (p adj < 0.05)) show significant warming trends for the period 1887-1997, while the CTI HadISST (0.21 ± 0.15°C/100 yr (p adj = 0.14)) and the CTI Kaplan records (0.21 ± 0.14°C/100 yr (p adj = 0.12)) rather point to nonsignificant SST trends.Analysis of the seasonal warming trends in the CTI coral shows significant warming trends (p adj < 0.05) across all seasons, with trends ranging from ∼0.5°C/100 yr (JJA) to ∼0.7°C/100 yr (DJF) (Figure S21 in Supporting Information S1).The instrumental CTI exhibits large discrepancy in the seasonal warming trends across different data sets.For example, the CTI ERSST shows significant warming trends (p adj < 0.05) in MAM, JJA and SON, whereas the CTI HadISST only exhibits a warming trend (p adj < 0.10) in MAM (Figure S21 in Supporting Information S1).Highest consistency is found between the warming trends in CTI coral and CTI ERSST across all seasons, while the trends in CTI HadISST and CTI Kaplan show poor agreement with the CTI coral record (Figure S21 in Supporting Information S1).Under the assumption of  (Conroy et al., 2009), the foraminifera Mg/Ca-SST record (pink, f) (Rustic et al., 2015), the tropical eastern Pacific SST derived from the coral network which consist primarily of coral δ 18 O records (cyan, g) (Tierney et al., 2015), and the central equatorial Pacific SST derived from the composite coral δ 18 O record, which was scaled into the expected SST using a slope of 0.22‰/°C (brown, h) (Epstein et al., 1953;Sanchez et al., 2020).The dashed lines represent the linear trends of each record computed over the period 1887-1997.In order to highlight the decadal variability of the Cold Tongue Index, all series were smoothed with a 11-year lowpassed filter.The sediment records in panels (e), (f) were interpolated to annual resolution before smoothing.constant quality in the coral Sr/Ca record over time, the CTI coral record shows a significant 20th century warming trend in the EPCT, potentially indicating an underestimation of warming trends in the instrumental CTI records.
To validate this hypothesis, we compared the secular warming trend in the CTI coral and in sediment-based SST reconstructions from the eastern equatorial Pacific, including a diatom record sensitive to rainfall influenced by SST (temporal resolution of ∼3 years) and a foraminifera Mg/Ca-SST record (temporal resolution of ∼30 years) from the Galápagos archipelago (Conroy et al., 2009;Rustic et al., 2015), with our reconstruction.The sediment records are significantly correlated with long-term changes in CTI coral (CTI coral -diatom: r = 0.59, p adj < 0.01; CTI coral -foraminifera: r = 0.56, p adj < 0.01 over the period 1887-1997) and show a 20th century warming around the Galápagos (Figures 2e and 2f).However, there are large uncertainties in quantifying the warming in the sediment records, in consideration of the nonlinear SST-rainfall relationship and the coarse temporal resolution.

Decadal to Multidecadal Linkage Between Walker Circulation and EPCT SST
The relationship between EPCT SST changes and the Walker Circulation have been determined on seasonal to interannual timescales, with the EPCT warming associated with a weakened Walker Circulation (Philander, 1990;Trenberth & Caron, 2000).By contrast, the linkage of EPCT SST to the (multi-)decadal changes in the Walker Circulation suffers from large uncertainties, due to the lack of reliable (multi-)centennial EPCT SST records (Bunge & Clarke, 2009;Deser et al., 2010).Our reconstruction therefore provides a unique opportunity to fill this knowledge gap.
Correlation (r = 0.71, p adj < 0.01, for the period 1887-1997; Figure S24 in Supporting Information S1) between the 11-year low-pass filtered CTI coral and instrumental SOI shows a significant relationship between decadal changes in EPCT SST and the Walker Circulation (Trenberth & Caron, 2000).The decrease of SOI and the warming of EPCT SST in the 20th century are consistent with CGCMs simulations which suggest that the weakening of the atmospheric circulation within the tropics in response to emission-driven global warming led to the warming of the EPCT (Held & Soden, 2006;Liu et al., 2019;Vecchi et al., 2006).Our results provide evidence for a linkage of the Walker Circulation on EPCT SST over the 20th century.However, the CTI coral and SOI poorly synchronize before ∼1935 (Figure S24 in Supporting Information S1).Ropelewski and Jones (1987) suggested a low quality of sea level pressure (SLP) data before ∼1935.Coincidentally, we also observed a breakdown in the inverse relationship between Tahiti and Darwin SLP during that period (Figure S24c in Supporting Information S1).These, together with the results of fixed-nest reconstruction and pseudo-proxy experiment (Figures S16 and S19 in Supporting Information S1), indicate that the disagreement before ∼1935 could be attributed to the low quality of SLP data, provided that the quality of coral Sr/Ca record remained constant over time.
Considering that the Tahiti is located in close proximity to the EPCT region and experiences a climate that is more closely linked to the EPCT SST than that of Darwin, which is also influenced by the monsoon and the Indian Ocean Dipole (Abram et al., 2020;Patterson et al., 2023), we further compared the CTI coral with the Tahiti SLP at decadal to multidecadal timescales.For the period 1887-1997, we observed that Tahiti SLP decreased with warming EPCT at both decadal (r = 0.56, p adj < 0.01; Figure 3a) and multidecadal (r = 0.88, p adj < 0.01; Figure 3f) timescales, indicating strengthened convection activity over the EPCT area and a weakened Walker Circulation.However, disagreement was still found before ∼1935 (Figure 3a), which could be associated with the low quality of SLP data during that period.In a similar vein, the instrumental CTI records were compared with the Tahiti SLP on decadal to multidecadal timescales .Unlike the CTI coral record, poor agreement between instrumental CTI and Tahiti SLP was observed before ∼1960, again demonstrating the limited reliability of early instrumental SST data over the EPCT region.SST (green), and COBE SST (purple) data sets with the SLP at Tahiti (asparagus green).The series in panels (a-e) and (f-j) are smoothed with 11-and 30-year lowpassed filters, respectively.Decreasing correlation between CTI coral and Tahiti SLP before ∼1935 is related to the poor quality of early instrumental data (Ropelewski & Jones, 1987).

Recent EPCT Cooling Not Unusual in the Context of the 20th Century
The global mean temperatures have experienced a short-lived, yet marked warming hiatus from the late 20th century to the early 21st century (England et al., 2014;Kosaka & Xie, 2013).This hiatus is associated with a "La Niña-like" condition, characterized by a pronounced EPCT cooling (England et al., 2014;Power et al., 2021).
Here we examined whether the recent EPCT cooling during the global warming hiatus stands out as anomalous in the context of the 20th century.The moving 20-year trends in CTI coral and instrumental CTI records are consistent with the phases of Interdecadal Pacific Oscillation (IPO), with a cooler EPCT corresponds to the negative phase of IPO (Figure 4a) (Kosaka & Xie, 2016;Meehl et al., 2011).This underscores the importance of internal variability in modulating EPCT SST changes (Meehl et al., 2011(Meehl et al., , 2013)).The trend observed from 1992 to 2011 emerges as the most negative within the post-1960 period across various instrumental SST data sets (Figure 4a).However, comparison of the 1992-2011 trend in annually averaged CTI M with the moving 20-year trends in annually averaged CTI coral reveals that the 1992-2011 trend is not unusual (14th percentile) in the context of the 20th century, as it does not fall below the 5th percentile in CTI coral (Figure 4b).The 1992-2011 trends in instrumental CTI from different data sets produce results similar to the CTI M , with the instrumental CTI records showing trends ranging from the 9th to 14th percentiles of CTI coral (Figure 4b).Previous studies suggested that the recent EPCT cooling shows a marked seasonality, with the most prominent cooling occurring in DJF (Jimenez et al., 2018;Trenberth et al., 2014).Hence, we further analyzed the CTI coral and instrumental CTI trends across different seasons.Similar to the annual mean trends, the 1992-2011 trend in all seasons consistently remains above the 5th percentile in CTI coral , indicating that they are not unusually negative (Figures 4c-4f).The most substantial 1992-2011 cooling in DJF only reaches the ∼9th percentile of CTI coral , while the EPCT cooling in JJA merely reaches the ∼20th percentile in the moving trends of CTI coral , appearing to be the mildest among all seasons (Figures 4c-4e).Overall, our results indicate that the recent EPCT cooling during the global warming hiatus is not anomalous in the context of the 20th century.Similar cooling trends of comparable magnitude have occurred in the past during the negative phase of the IPO.

Conclusion
Our CTI coral reconstruction provides a long-term context to observational studies and adds substantial new evidence on changes of EPCT SST since 1887.Importantly, the rapid 20th century warming in the reconstruction indicates an underestimation of warming trends in the instrumental CTI records.Among the CTI derived from four instrumental SST data sets, only the warming trend of CTI ERSST falls within the error of that of the CTI coral .The decadal to multidecadal changes in CTI coral show better association with Walker Circulation than instrumental SST records, supporting the robustness of our reconstruction.The CTI coral record indicates that the recent EPCT cooling during the global warming hiatus (1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011) is not unusual in the context of the 20th century.Our findings have implications for prediction of future EPCT SST changes by offering a constraint for model simulations.By comparing our reconstruction with model simulations, we could assess the accuracy and reliability of the models in capturing the observed EPCT warming trend during the 20th century.Nonetheless, future research is needed to assess whether the 20th century EPCT warming exceeds the range of natural variability, and to develop (multi-)century-long coral Sr/Ca records from the EPCT area.

Figure 1 .
Figure 1.EPCT SST signal in the coral Sr/Ca-SST records and coral-based reconstruction of the seasonal Cold Tongue Index (CTI).(a-f) Spatial correlations of coral and instrumental records with the OISST data set since 1982(Reynolds et al., 2007).(a) Coral record from Butaritari Atoll (BA)(Carilli et al., 2014).(b) Coral record from Palmyra Island (PI)(Nurhati et al., 2011).(c) Coral record from Clipperton Atoll (CA)(Wu et al., 2014).(d) Coral record from Wolf Island (WI)(Jimenez et al., 2018).(e) The instrumental CTI M(Deser & Wallace, 1990).(f) The PC1 of coral records.The stippling indicates correlations are significant at the 95% confidence level.All records were converted to anomalies prior to analysis.The black circles indicate the location of coral Sr/Ca-SST records, the rectangles frame the area of instrumental CTI.(g) Comparison of coral PC1 (black) with CTI M (red) and other ENSO indices (other colored lines; TableS3in Supporting Information S1).(h) Interannual (black) and decadal (11-year lowpass filtered, blue) variations of CTI coral record.The CTI coral was calibrated/validated against the CTI M (red) over the period 1960-1992 using a nested PCR approach.Gray shading shows the uncertainties of CTI coral (±1 RMSE).(i) Calibration R 2 (red), verification R 2 (blue), the reduction of error (RE, green) and the coefficient of efficiency (CE, purple) computed for each nest.( j) Number of coral records included in each nest.The blue shadings highlight the unreliable nests (i.e., nests with the lowest replication, weak statistical skills and large discrepancies with CTI M ; FigureS18and TableS4in Supporting Information S1).

Figure 2 .
Figure 2. Comparison of the coral reconstruction with instrumental and proxy records.(a-d) Comparison of CTI coral (black) with instrumental CTI ERSST (red, a), the CTI HadISST (blue, b), the CTI Kaplan (green, c), and CTI COBE (purple, d) and CTI bucket(gray, a-d).(e-h) Comparisons of the CTI coral (black) with the diatom record (orange, e)(Conroy et al., 2009), the foraminifera Mg/Ca-SST record (pink, f)(Rustic et al., 2015), the tropical eastern Pacific SST derived from the coral network which consist primarily of coral δ 18 O records (cyan, g)(Tierney et al., 2015), and the central equatorial Pacific SST derived from the composite coral δ 18 O record, which was scaled into the expected SST using a slope of 0.22‰/°C (brown, h)(Epstein et al., 1953;Sanchez et al., 2020).The dashed lines represent the linear trends of each record computed over the period 1887-1997.In order to highlight the decadal variability of the Cold Tongue Index, all series were smoothed with a 11-year lowpassed filter.The sediment records in panels (e), (f) were interpolated to annual resolution before smoothing.

Figure 3 .
Figure 3. Comparisons of the CTI coral and instrumental Cold Tongue Index (CTI) records with the sea level pressure (SLP) at Tahiti.(a) Comparison of the CTI coral (black) reconstruction with SLP at Tahiti (asparagus green).(b-e) Comparison of instrumental CTI records extracted from the ERSST (red), HadISST (blue), Kaplan SST (green), and COBE SST (purple) data sets with the SLP at Tahiti (asparagus green).The series in panels (a-e) and (f-j) are smoothed with 11-and 30-year lowpassed filters, respectively.Decreasing correlation between CTI coral and Tahiti SLP before ∼1935 is related to the poor quality of early instrumental data(Ropelewski & Jones, 1987).

Figure 4 .
Figure 4. 20-year trends in CTI coral and instrumental Cold Tongue Index (CTI) records.(a) Timeseries of moving 20-year trends in CTI coral (black; ±1 standard error (SE), gray shading), CTI M (brown; ±1 SE, brown shading), CTI ERSST (red), CTI HadISST (blue), CTI Kaplan (green) and CTI COBE (purple), with each point showing the 20year trend centering in that year.The phases of IPO index are indicated in the panel, with negative phases shaded in blue.The black dashed line represents a magnitude of 0, while the brown dashed line indicates the magnitude of the 1992-2011 trend in annual mean CTI M record.(b-f) Histogram of moving 20-year trends in the CTI coral in annual means, DJF, MAM, JJA and SON seasons throughout the 1887-1997 reconstruction interval, with black dashed line indicates the 5th percentile.Colored lines indicate the magnitude of the 1992-2011 trend in instrumental CTI records derived from different instrumental data sets.