The far-reaching impacts of central Pacific El Niño events on global climate differ appreciably from those associated with eastern Pacific El Niño events. Central Pacific El Niño events may become more frequent in coming decades as atmospheric greenhouse gas concentrations rise, but the instrumental record of central Pacific sea-surface temperatures is too short to detect potential trends. Here we present an annually resolved reconstruction of NIÑO4 sea-surface temperature, located in the central equatorial Pacific, based on oxygen isotopic time series from Taiwan tree cellulose that span from 1190 AD to 2007 AD. Our reconstruction indicates that relatively warm Niño4 sea-surface temperature values over the late twentieth century are accompanied by higher levels of interannual variability than observed in other intervals of the 818-year-long reconstruction. Our results imply that anthropogenic greenhouse forcing may be driving an increase in central Pacific El Niño-Southern Oscillation variability and/or its hydrological impacts, consistent with recent modelling studies.
Sea-surface temperature (SST) variations in the tropical Pacific and associated changes in global atmospheric circulation dominate global climate variability on interannual timescales. Recent studies distinguish between canonical El Niño events, with warming located in the central to eastern tropical Pacific, and central Pacific (CP) El Niño events, wherein warming is confined to the central tropical Pacific1,2,3. Both CP and the East Pacific El Niño-Southern Oscillation (ENSO) events have widespread impacts on global climate4,5,6,7, including influences on western North American drought8, East Asian monsoons9 and hurricane properties6. Climate model simulations suggest that CP ENSO variability may increase under greenhouse forcing1,10,11, but instrumental records of tropical Pacific SSTs are too short to provide robust constraints on recent trends in ENSO variability.
Instrumental and modelling studies use indices of large-scale SST variability in the central to western tropical Pacific—the NIÑO3.4 and NIÑO4 indices, respectively—to quantify CP ENSO variability through time1,2. While there are several high-resolution, multi-century reconstructions of NIÑO3.4 SST12,13,14,15, only two such reconstructions of NIÑO4 SST exist—one derived from an ice core in Peru16 and another based on tree-rings in southwest America17. Additional multi-century reconstructions of CP SSTs are required to improve quantification of the response of CP ENSO variability to both natural and anthropogenic climate forcings.
Here we present an 818-year-long, annually resolved record of tree-ring cellulosic oxygen isotopic (δ18O) composition from Taiwan, a region where CP ENSO-related changes in atmospheric circulation and hydroclimate are large9,18,19. Tree-ring cellulosic δ18O is a well-established proxy sensitive to large-scale hydrological conditions20. Our results show that the warm phases of our reconstruction correspond to strong El Niño years. The annual variation and variance of NIÑO4 SST are relatively high during the late twentieth century likely due to anthropogenic global warming.
Taiwan tree-ring δ18O chronology
In total, 50 Chamaecyparis formosensis Matsum tree-ring cores from 29 individual trees were collected from Mt. Daxue, Taiwan (∼24° N, 121° E) at an elevation of 2,000–2,200 m above sea level (Fig. 1a). After cross-dating the ring-width time series across every core (Supplementary Table 1), 16 cores were selected for cellulose δ18O analysis following standard protocols (Methods).
To avoid potential δ18O artefacts associated with juvenile isotope effects21, we excluded the first 20 years (rings)21,22,23,24,25 from those cores with no visible rot in the pith, following standard procedures. The resulting 16 annually resolved time series contain overlapping segments of 67–408 years in length (Fig. 2a–d,f), and have significant common variability where they do overlap (r=0.51–0.89, P<0.001; Supplementary Table 2). In generating a composite tree-ring δ18O record from this ensemble, we arithmetically averaged the 16 individual δ18O time series into a single δ18O time series spanning 1190–2007 AD (Fig. 2e). Aside from the high degree of between-sample reproducibility of cellulosic δ18O, a variety of standard statistical metrics confirms that the composite Taiwan tree-ring δ18O time series is robust (Fig. 2f; Methods).
Climate signals of Taiwan tree-ring composite δ18O record
The Taiwan tree-ring composite δ18O record is a sensitive indicator of regional hydroclimate, as indicated by significant correlations with regional precipitation δ18O time series (Methods; Fig. 3). At the local scale, the record is significantly correlated to temperature, precipitation and relative humidity (RH), such that higher tree δ18O values reflect warm, dry conditions (Supplementary Fig. 1a). Indeed, the Taiwan tree δ18O record is significantly correlated to the mean value of May–September Palmer Drought Severity Index26 (at the grid point 24.75° N, 121.75° E, r=−0.45, P<0.0001), consistent with previous studies of tree δ18O time series from southeast Asia27,28,29. Correlation analysis also showed significant relationship between Taiwan Palmer Drought Severity Index and regional precipitation (Supplementary Fig. 2).
Relationship between NIÑO4 SST and tree-ring δ18O
The new Taiwan tree δ18O record is significantly correlated with NIÑO4 SST, located in the heart of the CP ENSO region (Fig. 1; Supplementary Fig. 3). Correlations are highest after 1950, when the quality of tropical Pacific SST data is highest30 (Supplementary Fig. 4a). We also computed correlations between our Taiwan tree-ring δ18O series and other proxy-based reconstructions of NIÑO3.4 and NIÑO4 SST, and find that proxy–proxy correlations are higher in the early 20th century than the corresponding proxy-SST correlations (Supplementary Table 3). Our analyses suggest that the quality of instrumental data during the early 20th century may be somewhat reduced relative to the late twentieth century. Given that Taiwan δ18O-NIÑO4 SST correlations are highest during March–May (Supplementary Fig. 1b; Supplementary Table 4), we used the tree δ18O record to reconstruct the March–May NIÑO4 SST from 1190 to 2007 AD (Fig. 4a). Calibration and verification metrics (Methods; Supplementary Fig. 4b; Supplementary Table 5) confirm that our Taiwan tree δ18O-based reconstruction of NIÑO4 SST is robust throughout its length.
The positive correlation between Taiwan tree δ18O and the boreal spring NIÑO3.4 index reflects drier conditions across the western Pacific region during El Niño events. Time series of rainfall δ18O from the Global Network of Isotopes in Precipitation database confirm that during CP El Niño events, the δ18O of rainfall increases across the western Pacific as the locus of large-scale convection shifts away from the maritime continent to the CP (Methods; Fig. 3). These results suggest that the Taiwan tree δ18O record tracks local changes in rainfall δ18O, in line with findings from forward modelling studies of tree cellulose δ18O variability20. Previous studies have documented that local rainfall δ18O is more sensitive to ENSO variability than local rainfall amount in the western tropical Pacific27,28,29,31, given the appreciable spatial and temporal averaging inherent in the rainfall δ18O variations. Further support for the CP El Niño-Taiwan hydrological link comes from positive 850 hPa geopotential height anomalies in March–May in South Asia and the western Pacific during CP El Niño events32 (Fig. 5). The associated large-scale anticyclonic flow during CP El Niño extremes causes subsidence and the weakening of the prevailing southwest winds over Taiwan, both of which contribute to precipitation decreases throughout the region of interest. Collectively, these analyses provide a dynamical context for the observed correlations between the Taiwan tree δ18O record and the NIÑO4 SST index—the target of our reconstruction.
Characteristics of the reconstructed NIÑO4 SST
The Taiwan tree δ18O-based SST reconstruction contains a rich spectrum of variability spanning interannual to centennial timescales (Fig. 4; Supplementary Figs 5 and 6), similar to other multi-century reconstructions of tropical Pacific SSTs14. Several individual years stand out as exceptionally warm, allowing for the potential identification of strong El Niño years over the last millennium. On multi-decadal to century timescales, reconstructed SST values during the late 20th century are significantly higher than during any previous interval (Fig. 4d; Supplementary Table 6), consistent with anthropogenic warming of the NIÑO4 region. Indeed, anomalous late twentieth century warming in the central tropical Pacific is also inferred from coral δ18O time series from Maiana33 and Palmyra12 (Supplementary Fig. 7), as well as from a tree-ring-based reconstruction of NIÑO3.4 (ref. 14; Supplementary Fig. 8), all of which exhibit significant correlations with the Taiwan tree δ18O record over their periods of overlap (Supplementary Table 7).
Prominent interannual variability dominates the Taiwan tree δ18O record (Supplementary Figs 5 and 6), with a spectral signature similar to that observed in instrumental time series of the ENSO phenomenon. Before the twentieth century, the largest interannual excursions occur during the early to mid-seventeenth century, in line with previous observations of enhanced ENSO activity during this time12, possibly related to enhanced volcanic activity34. Indeed, the highest single tree-ring δ18O value of the entire reconstruction corresponds to 1651 AD, and may be linked to an exceptionally large El Niño event documented in historical records from the Paraná River region in South America35 as well as in coral δ18O records from the central tropical Pacific12,33 (Fig. 4a; Supplementary Fig. 7a,b). Interannual SST variance reaches a relative maximum during the late twentieth century (Fig. 4d), although appreciable spread in the individual tree-ring series precludes a finding of significance at the 95% confidence level. Taken at face value, our results provide empirical support for model projections of increased CP ENSO activity under continued anthropogenic climate change2,10. As such, the fact that the record-breaking 2015/2016 El Niño event was characterized by maximum warming in the CP, as opposed to the eastern Pacific, is consistent with a growing body of observational and modelling evidence for a prevalence of CP ENSO under greenhouse forcing.
Taken together, our results suggest that anthropogenic climate change has had a profound effect on SSTs in the CP, whereby anomalous warming over the last decades is accompanied by an increase in interannual variance. NIÑO4 SST values over the last two decades are likely higher than natural variations over the last 818 years, owing to a combination of relatively high CP ENSO activity and a late 20th century warming trend. In light of our results, it seems plausible that the dominance of CP ENSO extremes in the first two decades of the twenty-first century may continue, albeit with some important caveats. First, the global climate impacts of future CP ENSO extremes will critically depend on the evolution of the mean climate state in the tropical Pacific36,37, which itself is poorly constrained at present. Second, the new Taiwan tree δ18O record is the newest addition to growing archive of high-resolution paleo-data sets that can be used to probe the sensitivity of tropical Pacific climate to a variety of external climatic forcings over the recent past. One such example comes from the early- to mid-Holocene, when some models and data suggest that processional insolation forcing may have driven a shift towards greater CP ENSO activity and less East Pacific ENSO activity37. Should the dominance of CP ENSO extremes continue in the coming decades, investigations of the causes, and consequences, of any past shifts towards CP ENSO may provide some clues about future tropical Pacific climate trends and their global impacts.
Sample selection and cellulose extraction
Fifty tree-ring cores from 29 Chamaecyparis formosensis Matsum trees were collected from Taiwan in 2008. According to the standard dendrochronologial techniques, samples were polished and cross-dated. The quality of cross-dating was validated by program COFECHA38. Each individual tree-ring was identified as a calendar year. After accurate cross-dating, we selected 16 cores from each of 16 individual trees to carry out stable oxygen isotope analysis on the principle of ensuring at least 4 cores were present over the whole record.
The tree-ring cellulose was extracted as follows: materials of whole annual ring were separated and sliced into thin sections by a razor under microscope; the thin sections cut with a razor knife allow the chemical processing to proceed completely and rapidly; the sliced samples were chemically treated by acetone, a mixture of toluene and ethanol, acidified sodium hydrochlorite and 17.5% solution of sodium hydroxide in successive steps39,40; the cellulose of annual rings was transferred to a small bottle and homogenized with an ultrasonic cell crusher (JY92-2D, Ningbo Scientz Biotechnology Co., Ningbo, China); and the cellulose samples were dried overnight.
Stable isotopic analysis
We loaded each 130–170 μg homogenized cellulose sample into a silver capsule. Each silver capsule with an annual sample was sealed and packed. The samples were converted to CO at 1,350 °C using pyrolysis-type elemental analyser (TC/EA, Thermo Fisher, Germany) interfaced to an isotope ratio mass spectrometer (Delta V Advantage, Thermo Fisher, Germany). The 18O/16O ratio was expressed in delta (δ18O) notation with reference to a standard material for which the isotopic ratio is known (equation (1)). The δ18O was determined from the following equation:
where Rsample and Rstandard are the 18O/16O ratios for the sample and standard cellulose, respectively. Values of δ18O were reported with respect to the Vienna Standard Mean Ocean Water. The analytical reproducibility by analysing Merck cellulose (Merck KGaA, Darmstadt, Germany) was ±0.2‰.
Tree-ring δ18O chronology development
We used a Numerical Mix Method41 to establish an accurate and reliable Taiwan tree-ring δ18O chronology. The idea behind Numerical Mix Method is that several individual δ18O series are measured first, and then the mean values are calculated using an arithmetic average to produce a single isotope chronology. This method treats the stable isotope series such as a tree-ring index, akin to ring width. It follows the standard procedure of tree-ring width chronology development by measuring individual tree-rings and creating a mean site value42,43,44.
The individual δ18O series were combined into a single chronology by computing arithmetical mean. There is no reason to suspect that the Taiwan tree-ring δ18O series should not preserve low-frequency climate signals. EPS, the expressed population signal45,46, is used to evaluate the agreement between the δ18O series (or the common variance relative to the total variance). Generally, that an EPS value is >0.85 is considered to be an acceptable threshold for a reliable chronology45,46. The Rbar45 parameter indicates the average correlation between the δ18O series for each year over the sequential time periods. In this study, EPS and Rbar were calculated for Taiwan tree δ18O chronology by using a 50-year window that lags by 25 years.
Tree-ring δ18O responses to local climate parameters
The Yilan meteorological station (1936–2007 AD, 24° 46′ N, 121° 45′ E; 8 m above sea level) is the closest meteorological station to the sampling site with sufficiently complete records for climate response analysis. Thus, the observed precipitation, temperature and RH records from Yilan were used to identify the tree-ring δ18O climatic response. Considering the possibility that the climate of the current year not only affects tree growth in the current year but also in subsequent years47, we incorporated meteorological data from November of the prior year to October of the current year into our model. The monthly mean temperature, mean precipitation and mean RH of Yilan station are shown in Supplementary Fig. 9.
Tree-ring δ18O responses to precipitation δ18O time series
Generally, the amount of precipitation shows a negative correlation with δ18O of precipitation at lower latitudes, which is referred as the ‘Amount Effect’48. Decreasing amount of precipitation in the western Pacific region during El Niño enriches the ratio of 18O/16O in precipitation. Tree physiology has demonstrated that δ18O of tree-rings was positively correlated with δ18O of precipitation and negatively correlated with RH18. However, there was no strong correlation between the local RH and the tree-ring δ18O (r=−0.21, n=72, P>0.01) at Taiwan sampling site, which implies that the δ18O of precipitation was the most important factor for determining the tree-ring δ18O value.
On the other hand, the Taiwan tree-ring δ18O series was significantly correlated with the precipitation δ18O records from the adjacent western Pacific region obtained from Global Network of Isotopes in Precipitation: r was 0.42 with Bangkok precipitation δ18O (n=40, P<0.0001), 0.68 with Kunming (n=16, P<0.001) and 0.57 with Hong Kong (n=35, P<0.0001; Fig. 3). Therefore, enriched 18O/16O of precipitation resulted in high δ18O value of tree-rings in Taiwan (including the western Pacific) during El Niño events and vice versa. It suggests that the Taiwan tree-ring δ18O contained large-scale precipitation δ18O signals in the northwestern Pacific sector.
Split calibration-verification method
Analysis revealed that the Taiwan tree-ring δ18O is highly correlated with NIÑO4 SST from March to May during 1950–2007 (SSTMAM, Supplementary Fig. 1b). A transfer function was then designed to reconstruct the central Pacific NIÑO4 SST using Taiwan tree-ring δ18O:
(n=58, r=0.734, R2=0.539, R2adj=0.531, F=65.546, P<0.0001, D/W=1.82).
As shown in Fig. 1b, the reconstructed SST matched the observed Kaplan NIÑO4 SST pretty well. The r was 0.64 after first difference (1951–2007, P<0.0001, Supplementary Fig. 4b), indicating their significant and stable relationships in high frequency.
The stability and reliability of the regression equation were evaluated using the split calibration-verification method45,49. It was performed by calibrating the NIÑO4 SST data from a subperiod (the data set was divided into two parts, 1950–1979 and 1978–2007) and verifying the reconstruction using the remaining data. The results were evaluated by the correlation coefficient (r), the sign test (ST), the reduction of error test (RE), the coefficient of efficiency (CE) and the product means test (t) during the verification period. Generally, RE and CE values greater than zero indicate a rigorous model skill49. Larger values of the RE and CE indicate better results. Moreover, the values of CE are more rigorous and are typically lower than those of RE (Supplementary Table 5).
The effective number of degree of freedom estimation
Since there are autocorrelations in the data used in this study and their corresponding number of degree of freedom is reduced, the effective number of degree of freedom (EDOF) is estimated to test the significance level of correlations for each pair of time series. In this paper, we used the method described by Bretherton et al.50 for estimating EDOF. The EDOF is estimated by:
where N denotes the length of the time series, and r1 and r2 refer to the lag-one autocorrelation of each series, respectively.
The data at the ends of the time series before smoothing were padded by using the mean value of the remaining data. When there were 30 remaining data, their mean value was used as the 31th data. So that there were enough 31 data used to calculate the 31-year low-pass filter value. And, this process was repeated until the last year51.
Data that have contributed to the reported results are available from the corresponding author on request.
How to cite this article: Liu, Y. et al. Recent enhancement of central Pacific El Niño variability relative to last eight centuries. Nat. Commun. 8, 15386 doi: 10.1038/ncomms15386 (2017).
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This study was jointly supported by grants from the QYZDJ–SSW–DQC021, NSFC41371221, NSFC41630531, CASXDA05080202, 973 Program 2013CB955900 and the Key Project of IEECAS. K.M.C. acknowledges support from NOAA C2C2 Award # NA11OAR4310165, and NSF Awards OCE–1446274 and OCE–1446343. T.N. acknowledges support from Grant-in-Aid for Scientific Research from the Japanese Society for the Promotion of Science (23242047 and 26244049). C.-C.S. thanks grant support (104–2119–M–002–003) and the National Taiwan University (105R7625).
Supplementary Figures, Supplementary Tables and Supplementary References
This work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/