Half-precessional cycle of thermocline temperature in the western equatorial Pacific and its bihemispheric dynamics

Significance

The precessional insolation plays a critical role in modulating the low-latitude oceanographic changes through ENSO-like dynamics. While previous studies focused on the precessional insolation-induced low-latitude surface water changes, our well-dated, centennial-resolution proxy data and numerical simulation reveal the half-precession cycle of thermocline water temperature (TWT) in the western equatorial Pacific that directly linked to ENSO-like zonal temperature seesaw over the equatorial Pacific. These half-precessional TWT changes are proposed to originate from the interplay of subtropical-to-tropical thermocline anomalies forced by the antiphased meridional insolation gradients in the two hemispheres at the precessional band. This work illustrates that the tropical thermocline water could bridge the meridional climatic processes and the east−west ENSO-like dynamics at orbital timescales.

Abstract

The El Niño−Southern Oscillation (ENSO), which is tightly coupled to the equatorial thermocline in the Pacific, is the dominant source of interannual climate variability, but its long-term evolution in response to climate change remains highly uncertain. This study uses Mg/Ca in planktonic foraminiferal shells to reconstruct sea surface and thermocline water temperatures (SST and TWT) for the past 142 ky in a western equatorial Pacific (WEP) core MD01-2386. Unlike the dominant 100-ky glacial−interglacial cycle recorded by SST and δ¹⁸O, which echoes the pattern seen in other WEP sites, the upper ocean thermal gradient shows a clear half-precessional (9.4 ky or 12.7 ky) cycle as indicated by the reconstructed and simulated temperature (ΔT) and δ¹⁸O differences between the surface and thermocline waters. This phenomenon is attributed to the interplay of subtropical-to-tropical thermocline anomalies forced by the antiphased meridional insolation gradients in the two hemispheres at the precessional band. In particular, the TWT shows greater variability than SST, and dominates the ΔT changes which couple with the west−east SST difference in the equatorial Pacific at the half-precessional band, implying a decisive role of the tropical thermocline in orbital-scale climate change.

Present-day climate is significantly affected by the tropical Pacific (1). The western Pacific warm pool, which has a permanent sea surface temperature (SST) of >28 °C, is the largest reservoir of warm upper water on Earth and hence a heat engine for global climate. Furthermore, thermocline temperature in the western equatorial Pacific (WEP) contributes significantly to the maintenance of the equatorial SST climatology and acts as a precursor of El Niño−Southern Oscillation (ENSO) variability through the eastward propagation of subsurface temperature anomalies (2, 3). Paleoceanographic records from the equatorial Pacific Ocean provide insight into ENSO behavior when global boundary conditions—ocean and atmosphere circulations, sea level and ice volume, atmospheric greenhouse gas concentration, and albedo—were different from today, and thus help in the projection of trends or variance of ENSO in response to climate change. The ENSO cycle is known to be a feature of Earth’s climate in the geological past (4, 5), and its signal should be more prominent in the thermocline response, as in modern climate (6). However, a lack of tropical thermocline water temperature (TWT) records with high temporal resolution has limited our understanding of how ENSO responds to orbital changes (7, 8). Here we examine thermocline temperature evolution in the WEP over the past 142 ky to investigate long-term ENSO-like changes and three-dimensional ocean dynamics.

We use deep-sea sediment samples from core MD01-2386, which was collected from the WEP during the R/V Marion Dufresne 122 cruise in 2001 (Fig. 1; 01°07.80′N, 129°47.56′E; 2,816 m water depth), for reconstructing records of SST and TWT (SI Appendix, Table S1) based on the Mg/Ca ratios of surface-dwelling Globigerinoides ruber sensu stricto and thermocline-dwelling Pulleniatina obliquiloculata planktonic foraminiferal shells, respectively. About 30 G. ruber and 20 P. obliquiloculata shells were picked from each sample of the upper 2,089 cm in the core, with an average sample interval of 2 cm (1,037 G. ruber samples and 887 P. obliquiloculata samples). Foraminiferal shells were pretreated following the procedure described by Lea et al. (16). Trace elements including Mg/Ca were determined in G. ruber samples from the upper 400 cm by Element2 inductively coupled plasma mass spectrometry (ICP-MS) at the Department of Earth Science, University of California, Santa Barbara, following the procedure in Lea et al. (17); the remaining samples were analyzed at the State Key Laboratory of Marine Geology, Tongji University, China, with an instrumental precision of 1.1% (1 SD). We use the species-specific Mg/Ca SST calibration of G. ruber developed in the study area (18). For TWT, the published calibrations of P. obliquiloculata Mg/Ca and temperature (18–21) have similar exponential constants (∼0.09), which would result in comparable trends and amplitudes of TWT estimates. In order to better compare with previous studies, we apply the equation of Anand et al. (19) to calculate TWT. The error of TWT reconstruction, introduced by the Mg/Ca measurements of 87 replicates for P. obliquiloculata and Mg/Ca temperature calibrations (22), is estimated at ±1.1 °C.

Fig. 1.

Modern features of the thermocline water at site MD01-2386 based on the annual mean SODA reanalysis dataset over the years from 1871 to 2010 (9). (A) Core location (white star) and the tropical Pacific temperature at 112 m water depth—the tenth vertical layer of the SODA dataset—here denoted as TWT. For the STC, the black solid arrows indicate its western boundary and interior paths (10–12), whereas the black dashed arrows indicate its oceanic wave pathways (13–15). EUC, ME, and HE denote the Equatorial Undercurrent, Mindanao Eddy, and Halmahera Eddy, respectively. (B) For the SODA ocean temperature (T) at the location of MD01-2386 (01°07.80′N, 129°47.56′E), its vertical lapse rate (red line) and correlation with Nino3_SST (blue line) both reach maxima near the 120-m water depth. (C) Nino3_SST (blue line) is highly correlated (R = −0.66) to MD2386_TWT (red line) but less correlated (R = −0.14) to MD2386_SST (black line), where “MD2386_” stands for regional averaged values of the nearest 0.5° × 0.5° grid. D and E show spatial correlation coefficients of annual mean SST and TWT relative to the MD01-2386_TWT (red line in C). Dotted regions in D and E are significant at the 99% confidence level.

The age model of this core is established by 17 radiocarbon dates (Fig. 2 and SI Appendix, Table S2) of G. ruber (measured at Leibniz Laboratory, Kiel University, Germany) for the upper 980 cm of the core, as well as the correlations of planktonic G. ruber and benthic Cibicidoides wuellerstorfi δ¹⁸O curves with the benthic δ¹⁸O stack LR04 (24) (SI Appendix, Fig. S1). Radiocarbon dates are converted to calendar ages by program CALIB 7.1, and the depth−age relationship is established by applying the Bayesian statistical methodology of Bacon (25) (SI Appendix, Fig. S2). The stable isotopic analysis of G. ruber (1,226 samples, 2-cm interval), P. obliquiloculata (1,224 samples, 2-cm interval), and C. wuellerstorfi (410 samples, 2-cm interval from 0 cm to 600 cm and from 1,684 cm to 1,984 cm) were carried out using a Finnigan MAT253 mass spectrometer at the State Key Laboratory of Marine Geology, Tongji University. Thus, the average time resolution of samples is ∼120 y but with a higher resolution of ∼60 y for the upper 980 cm in the core.

Fig. 2.

Comparison of (A) G. ruber δ¹⁸O, (B) P. obliquiloculata δ¹⁸O, (C) SST (based on G. ruber Mg/Ca), and (D) TWT (based on P. obliquiloculata Mg/Ca), (E) δ¹⁸O difference between P. obliquiloculata and G. ruber (Δδ¹⁸O_P-G), (F) temperature difference between SST and TWT (ΔT_G-P), and (G) differences between the surface (0 m) and subsurface (100 m) temperatures (ΔT_0–100m) estimated by the transfer function of SIMMAX28 (23). All bold curves are five-point moving average filters. Gray dashed line in D and red dashed lines in E–G denote the Gaussian band-pass filters (center frequency = 0.109 ky⁻¹) of TWT and ΔT_G-P (bandwidth = 0.005), and Δδ¹⁸O_P-G and ΔT_0–100m (bandwidth = 0.015). MIS 1 to ∼6 at the top indicates marine isotopic stages, while red triangles denote the accelerator mass spectrometry (AMS) ¹⁴C age determinations.

Half-Precessional Cycle in the Upper Thermal Structure.

The SST reconstructed from core MD01-2386 varies between 24.4 °C and 29.3 °C, while the TWT fluctuates between 15.5 °C and 26.9 °C over the past 142 ky (Fig. 2). The Mg/Ca-derived SST and TWT averaged 27.5 ± 0.5 °C and 22.7 ± 0.7 °C (all errors quoted are 1 SD) during the Late Holocene (0 ka BP to 4 ka BP), respectively, close to the observed modern annual mean SST of 28.8 °C and TWT of 24.2 °C, as defined by the temperature at a depth of 112 m—the tenth vertical layer of the Simple Ocean Data Assimilation (SODA) dataset (9), and also the habitat depth of P. obliquiloculata [between 100 m and 150 m water depth (upper thermocline) (18, 26)] at this site. SST averaged 25.8 ± 0.6 °C across marine isotope stages (MISs) 2 to 4, and 28.4 ± 0.5 °C during MIS 5e, while the TWT averaged 21.2 ± 1.2 °C across MISs 2 to 4, and 23.5 ± 1.8 °C during MIS 5e. SST and TWT average 25.6 ± 0.5 °C and 21.5 ± 1.2 °C during the Last Glacial Maximum (19 ka BP to 23 ka BP) and thus are lower than the Late Holocene mean by 1.9 ± 0.5 °C and 1.2 ± 0.9 °C, respectively.

The Mg/Ca reconstructed TWT (SD 1.68 °C) is more variable than the SST (SD 0.91 °C) (Fig. 2 C and D), as is also the case for modern observed TWT and SST (Fig. 1C; with SD of 1.14 °C and 0.34 °C, respectively). Especially, the millennial- to centennial-scale variability of the TWT exceeds the estimated error, suggesting greater short-term variability. The millennial-scale variability of TWT during the last glacial interval (SI Appendix, Fig. S3) displays behavior similar to Antarctic air temperature and atmospheric CO₂ (27, 28), implying a greenhouse gas forcing on tropical TWT variability and a potential climate link between the tropical Pacific and Southern Ocean, possibly via ocean tunnel connections (29, 30). Over Termination I, both SST and TWT display a clear mid-Termination plateau centered at ∼12.5 ka BP (13.2 ka BP to 11.8 ka BP). A similar plateau has been identified in atmospheric CO₂ records and other tropical SST records (22, 31–33).

A five-point moving average filter was applied to all SST and TWT (1-σ range of variability is ±0.4 °C for SST and ±1.0 °C for TWT; Fig. 2). Generally, SST changes in parallel with the δ¹⁸O of G. ruber, displaying a typical 100-ky glacial−interglacial cycle with a peak-to-peak amplitude of 3 °C to 5 °C, consistent with previous studies (16). The TWT also exhibits a clear glacial−interglacial cycle, but with a larger peak-to-peak amplitude of 5 °C to 8 °C, which is consistent with a newly published P. obliquiloculata Mg/Ca TWT record of the last 110 ky (ref. 34 and SI Appendix, Fig. S4) from site GeoB17419-1 near site MD01-2386, indicating that TWT is more sensitive to orbital-scale climate change. After removing glacial−interglacial trends, the spectrum of SST is dominated by a strong precessional (22.8 ky) signal (Fig. 3A), similar to that at site MD05-2930 in the Gulf of Papua (37). Interestingly, an obvious half-precessional cycle (9.4 ky) (38) can be identified, even by naked eye, from the difference between SST and TWT (ΔT_G-P, denoting the upper water thermal gradient) over the past 142 ky (Fig. 2F), and is dominant in the spectrum of TWT and ΔT_G-P (Fig. 3 B and C). The obliquity cycle is relatively weak in TWT and ΔT_G-P at site MD01-2386 and in TWT at site MD05-2930 (near 10°S) (37), slightly different from the dominant obliquity cycle in TWT record at site GeoB17419-1 north off Papua New Guinea near 2°S (34).

Fig. 3.

Spectral amplitudes for the (A) SST, (B) TWT, and (C) ΔT_G-P records from core MD01-2386 are compared with those of the CESM modeled (G) SST_CESM, (H) TWT_CESM, and (I) ΔT_CESM at the same location. Also shown are spectral amplitudes of the (D) ΔT_0–100m and (E) Δδ¹⁸O_P-G from core MD01-2386, and (F) the cross-spectrum between ΔT_G-P and ΔT_CESM. Note that, before spectral analysis, glacial−interglacial changes of the Mg/Ca SST and TWT have been removed by a method of quintic polynomial fitting for a better comparison with the CESM simulation. The spectral analyses are performed by Redfit38 with Hanning (in A–E) or rectangular (in G–I) windows (oversample, 4; segment, 1) (35), and the cross-spectrum analysis is performed by Redfit-X with Hanning window (oversample, 4; segment, 3; MC, Monte Carlo; MC-Csig, Monte Carlo significant confidence) (36). Numbers above the spectrum curves in A–I denote those dominant periods (unit: kiloyears) exceeding the 95% confidence level, that is, 23- to 19-ky periods for precession and 12.7- to 8.5-ky periods for half-precession.

Mg/Ca ratios of sedimentary planktonic foraminifera are affected by the deep-water dissolution effect (21, 39, 40). Foraminifera shells from core MD01-2386 (2,816 m) were deposited in undersaturated bottom waters and likely experienced partial dissolution (see discussion in Methods). But, during the last two terminations, when the Mg/Ca temperatures change the most significantly (SI Appendix, Fig. S4), the planktonic foraminiferal fragmentation—the best indicator of dissolution (41)—remains roughly constant. Although the fragmentation suggests relatively intensified deep-water dissolution during MISs 2 to 5d (SI Appendix, Fig. S4), its spectrum exhibits no significant peak at the precessional and half-precessional periods (SI Appendix, Fig. S5). Therefore, while it is possible that some of the TWT variability is due to dissolution, the main features, such as the half-precessional cycle, do not appear related to dissolution.

The half-precession variability of the reconstructed ΔT_G-P changes is confirmed by the temperature difference (ΔT) between the water depths of 0 m and 100 m (ΔT_0–100m; Figs. 2G and 3D) reconstructed from the transfer function of planktonic foraminiferal census data (23). Both the ΔT_G-P and ΔT_0–100m show roughly the same range of variation (∼6 °C) and clear half-precessional cycles, and match well with each other except for the interval of 125 ka BP to 135 ka BP (Fig. 2 F and G). The half-precessional variability is further confirmed in a transient sensitivity experiment in the Community Earth System Model 1.0.4 (CESM) under the orbital forcing since 300 ka and other boundary conditions in AD 1950 (called experiment CESM_transient). In this experiment, the model was integrated for 3,000 model years with the 100-fold accelerated orbital insolation forcing of the past 300 ky (detailed in Methods).

The modeled ΔT at site MD01-2386 (called ΔT_CESM; SI Appendix, Fig. S7) shows a prominent half-precessional change that is statistically significant at 95% level (Fig. 3I), although the orbital insolation forcing only consists of eccentricity, obliquity, and precession (without a half-precession component). Despite the smaller variance in TWT and SST in the model (called TWT_CESM and SST_CESM), which results partly from the absence of other driving factors (i.e., ice sheet, sea level, and atmospheric greenhouse gases), the TWT_CESM also exhibits a larger magnitude of variation than the SST_CESM (with SDs of 0.39 °C and 0.22 °C, respectively; SI Appendix, Fig. S7), and the simulated vertical temperature difference ΔT_CESM is generally comparable to the proxy-reconstructed ΔT_G-P, notably, in periods and phase. This consistency in period and phase can be further seen in the cross-spectrum between ΔT_G-P and ΔT_CESM, which shows significant coherence in the half-precessional band (9.2 ky) at the 95% confidence level (Fig. 3F). The robustness of this half-precessional cycle in the WEP TWT seems to be also confirmed in the same type of transient experiment but in another independent air−sea coupled model (called Community Climate System Model version 3) (detailed in SI Appendix). This robust model−data consistency suggests that some features of the ΔT record from the WEP, and, in particular, its half-precessional cycle, could be forced by orbital insolation.

The difference between the oxygen isotope composition of thermocline-dwelling species P. obliquiloculata and surface-dwelling species G. ruber (Δδ¹⁸O_P-G; Fig. 2E) is widely used to indicate changes in the depth of the thermocline, with larger values representing a shallower thermocline (42, 43). Whereas the δ¹⁸O curves of P. obliquiloculata and G. ruber display a typical glacial−interglacial cycle of 100 ky, the dominant period for the Δδ¹⁸O_P-G is precession and half-precession (12.7 ky; Fig. 3E). The Δδ¹⁸O_P-G is correlated to the ΔT_G-P with a cross-spectrum coherence of 0.85 at the 95% confidence level in the half-precessional band. The half-precession periods (12.7 and 9.4 ky) in this study (Fig. 3) are almost exactly half-periods of the two precessional periods (23 and 19 ky) (38, 44), but the periods are slightly different for TWT and ΔT_G-P (9.4 ± 0.4 ky) and for Δδ¹⁸O_P-G (12.7 ± 0.6 ky) at the 95% confidence level (Fig. 3). A similar half-precessional period (12.0 ky) is also significant in the primary production index of coccolith at site MD01-2386 (SI Appendix, Fig. S4), indicating a nontrivial linkage between thermocline dynamics and phytoplankton productivity. The half-precessional cycle has been found in other records from equatorial regions (45–47), but our results indicate the existence of half-precession cycle in proxies for the upper thermal gradient in the WEP, which is directly linked to ENSO. Our numerical study also provides modeling support of the half-precessional response in the WEP thermocline.

Equator−Subtropical Insolation Gradient Forcing from Two Hemispheres.

The TWT in the WEP is largely maintained by the thermocline waters subducted from the subtropical Pacific in both hemispheres in late local winter; these waters are transmitted to the WEP in the Subtropical−Tropical Cell (STC) through thermocline ventilation in the low-latitude western boundary pathway and oceanic Rossby and Kelvin waves (Fig. 1A and refs. 2, 10–12, 14 and 15, and 48–50). These mechanisms influencing modern TWT at the WEP should have also played an important role in the geological past. Because site MD01-2386 is located in the WEP and right at the crossroads of the western boundary pathways from both hemispheres (Fig. 1A), we hypothesize that the TWT at this site can be determined by the equatorward subsurface temperature anomalies transported by the STCs from both hemispheres.

The CESM_transient simulation shows that the precessional forced TWT changes at site MD01-2386 are alternatively controlled by the contrasting thermocline temperature anomalies from the North and South Pacific (Fig. 4 A and B), indicating an interhemispheric interplay of the STCs from both hemispheres. For example, in boreal summer at precessional minimum, thermocline temperature is warmer (colder) in the subtropical North (South) Pacific (Fig. 4A) when Earth is closer to (farther away from) the sun in the Northern (Southern) Hemisphere, and is reversed at precessional maximum (Fig. 4B) (all of the signs are opposite in boreal winter; Fig. 4 C and D). The thermocline temperature anomaly is then transported into the WEP through the STC, especially in the low-latitude western boundary current (10, 48, 50–52). Therefore, the equatorial thermocline temperature is determined by the contrasting thermocline temperatures and, in turn, insolation in the two hemispheres (53).

Fig. 4.

A and B are composite difference of July−September (JAS; boreal summer) TWT relative to climatological mean from the CESM simulation during precessional (A) minimum and (B) maximum. C and D are composite difference of January−March (JFM; boreal winter) TWT relative to climatological mean during precessional (C) minimum and (D) maximum. Core location of MD01-2386 is shown as the yellow stars in A–D. (E) Five-point smoothed ΔT_G-P of core MD01-2386 is compared with that of the ΔT_CESM. (F) ΔT_G-P is compared with meridional insolation gradients between 0°N and 30°N during November (brown dashed line) and between 0°N and 30°S during May (blue dashed line). (G) The theoretical ΔT from a simplified conceptual model (detailed in Methods) is compared with ΔT_G-P.

The similarity between ΔT_G-P and ΔT_CESM (Fig. 4E), as proved by cross-spectrum coherences of 0.78 at 19 ky and 0.88 at 9.2 ky (Fig. 3F), suggests that the half-precessional signal in the WEP thermocline water may be a “footprint” of the STCs from two hemispheres during different “seasons” in the precessional cycle. Indeed, the dynamic mechanism discussed above suggests that this signal should be the strongest in the western part of the equatorial thermocline, where the equatorward convergence is the strongest. This spatial structure of the half-precessional signal is confirmed by the tongue-shaped distribution of positive regression coefficient values in the ocean temperature field, both at the thermocline depth plane (Fig. 5D) and along the equatorial profile (Fig. 5E).

Fig. 5.

Link between the upper thermal gradient in the WEP and the west−east SST difference in the equatorial Pacific. (A) ΔT_G-P records of core MD01-2386, and (B) the SST difference between the west and east equatorial Pacific (ΔSST_W-E) calculated from the SST records of cores MD01-2386 and TR163-22 (17) (detailed in Methods). Gray lines are the original records, black thick lines are the five-point moving average filters. Red line in A and blue line in B are band-pass filters at the period of 9.4 ky (frequency = 0.106 ky⁻¹, bandwidth = 0.005); comparison of the band-pass filters is shown in C. Associated half-precessional ENSO-like dynamics in the CESM_transient experiment are shown in D and E. (D) Regression coefficients of the annual mean TWT against the half-precessional filter of TWT_CESM at site MD01-2386 (red dashed line in SI Appendix, Fig. S7B). White dotted areas in D are significant at the 99% confidence level for the ocean temperature. (E) Similar to D but for regression coefficients of the annual mean equatorial ocean temperature (shaded; unit is degrees Celsius) and currents (black arrows indicate horizontal velocity [centimeters per second] and vertical velocity [3.33 × 10⁻⁵ cm·s⁻¹]) along the latitudes 5°S to 5°N. Yellow stars in D and E show the position of sites MD01-2386 and TR163-22.

We further propose a conceptual model of theoretical ΔT forced by precessional insolation changes between the equator and subtropics of two hemispheres (Fig. 4 F and G; detailed in Methods and SI Appendix, Fig. S8). Results from the CESM_transient experiment (SI Appendix, Fig. S10) show that the November meridional insolation gradient between 0°N and 30°N and the May insolation gradient between 0°N and 30°S can be taken as the original forcing of the subducted temperature anomaly in the North and South Pacific, respectively. The envelope curves of the two insolation gradients seem to be coincident with the half-precessional fluctuations of ΔT_G-P (Fig. 4 F and G). Two minima or maxima of the envelope curves during each precession cycle [so-called “clipped precession” (45), also shown in SI Appendix, Fig. S8B] may generate a signal of half-precessional period in the theoretical ΔT (Fig. 4F).

Except for Termination II, the half-precessional change of theoretical ΔT generally matches well with the five-point moving averaged ΔT_G-P at site MD01-2386 over the past 142 ky (Fig. 4G), with a cross-spectrum correlation of >0.50 at the 95% confidence level, supporting the hypothesis that the upper water thermal structure in the WEP is forced by the meridional equator-to-subtropical gradients of insolation or temperature in both hemispheres. For each precessional cycle, the ΔT_G-P at the WEP will show two peaks, due to the interplay of the STCs in both hemispheres (Fig. 4F), and may be closely related to the insolation-driven meridional migration of the Intertropical Convergence Zone (ITCZ) (11, 13, 54). Although a previous study found no relationship between equatorial insolation and half-precession (55), subtropical forcing is hypothesized to generate a significant temperature response in the equatorial thermocline for forcing periods longer than decades (48), such as orbital-scale change in the Middle Pleistocene (56).

Role of Thermocline in the Orbital-Scale ENSO-like Changes.

It has been proposed that changes in tropical Pacific SST parallel atmospheric CO₂ concentration and precede the δ¹⁸O changes during glacial terminations (16, 17, 57), supporting greenhouse gas forcing on tropical climate (30, 58). In this study, the SST, TWT, and surface δ¹⁸O change together during the last termination, but the TWT clearly leads the SST and surface δ¹⁸O changes during Termination II (Fig. 2). This leading role of the WEP TWT is consistent with modern observations, in which annual mean ocean temperature near the 120-m water depth (rather than at the surface) at site MD01-2386 exhibits the largest anticorrelation with the sea surface temperature in Nino3 region (5°S-5°N,150°W-90°W) (Nino3_SST) (Fig. 1 B and C). As an indicator of the warm pool heat content (3), modern WEP TWT has been assumed as an effective precursory ENSO index that triggers El Niño events at the sea surface of the eastern equatorial Pacific (EEP) (59). That means, in the precursory stage of El Niño, a warmer WEP TWT accompanies with a La Niña-like equatorial SST pattern simultaneously (or increased west−east SST gradient; Fig. 1D).

While present-day climate is featured by a strong west−east SST gradient in the equatorial Pacific, paleoclimate records suggest that the mean state of the equatorial Pacific has not always been characterized by such sharp zonal SST gradients due to various timescale fluctuations in the equatorial thermocline (8, 29, 60, 61). The half-precessional signal in TWT should also have its imprint in the surface climate through its impact on the zonal SST gradient along the equator (60, 61), because the WEP thermocline anomaly can regulate the EEP SST through multiple external forcing-modulated air−sea feedbacks (62–64), that is, dynamic movements of thermocline tilt due to an eastward equatorial Kelvin wave (59), or upper oceanic upwelling anomalies in the EEP (64). In order to investigate the orbital-scale behavior of the mean state of the equatorial Pacific, the SST differences between sites MD01-2386 in the WEP and TR163-22 in the EEP (17) are calculated to represent the west−east SST gradient in the equatorial Pacific (ΔSST_W-E). The age models of both sites are established using the same method and procedure (detailed in Methods). As shown in Fig. 5, the ΔSST_W-E is well anticorrelated to the ΔT_G-P in the WEP at the half-precessional band, with a high cross-spectrum correlation of 0.89, but the amplitude for the ΔT_G-P changes is greater, consistent with the “ocean dynamical thermostat” theory (2, 52, 62). Particularly, both ΔT_G-P and ΔSST_W-E clearly display half-precessional cycles of 9.4 ky, which is not significant in the SST records of either MD01-2386 or TR163-22, reflecting the variability of zonal SST gradient rather than features of any of the original SST time series. It is noted that the half-precessional filter for ΔT_G-P at the period of 9.4 ky changes almost simultaneously with the reversal of ΔSST_W-E filter (Fig. 5C). This orbital-scale anticorrelation between ΔT_G-P and ΔSST_W-E is almost the same as that in the modern interannual ENSO pattern, implying that the WEP thermocline water also plays an active role in the past ENSO-like changes over the equatorial Pacific. These half-precessional ENSO-like dynamics are further supported by the CESM_transient experiment, in which the TWT at MD01-2386 can represent large parts of the WEP TWT and forms a zonal seesaw with the eastern Pacific upper ocean temperature (Fig. 5 D and E), which has been shown earlier as the footprint of the half-precession signal in WEP.

Both modern observation (Fig. 1) and numerical simulation (12, 59) have shown that the upper water thermal structure in the WEP is tightly coupled with an ENSO-like SST anomaly. Therefore, the thermocline temperature change at site MD01-2386 can be used as a sensitive diagnostic tool to track changes in the mean state and its potential modulation of paleo-ENSO variability. Here, the ΔT_G-P is well correlated to the surface ENSO-like proxy, ΔSST_W-E, indicating that the ENSO-like mean state over the equatorial Pacific also has an obvious half-precessional cycle. As discussed above, the TWT and ΔT_G-P in the WEP are driven by the meridional insolation gradient in the tropical regimes of both hemispheres, which is related to the precession-paced ENSO-like SST anomaly (63, 64) and the south-to-north migration of the ITCZ. The half-precessional cycle has been found in records of ITCZ-driven monsoon rainfall at the equator in the Late Quaternary (47) and even in the deep time of the Devonian through Permian (44, 65). This first discovery and understanding of the half-precessional signal in the WEP thermocline builds a bridge to understand long-term changes of ENSO-like variability and ITCZ migration. Untangling the climatic links between the south-to-north migration (monsoon) and the east−west asymmetry (ENSO-like) of the ITCZ through the thermocline temperature change over the Pacific at long timescales will help improve our understanding and projections of future climate change trends in the tropical−extratropical climate.

Methods

Laboratory Measurements.

The sediments from core MD01-2386 are olive-gray calcareous ooze with slight bioturbation. The upper 2,088.5 cm of the core was sampled at an average interval of 2 cm. For the stable isotope analyses, about 8 to 12 specimens of G. ruber sensu stricto (250 μm to 350 μm) and P. obliquiloculata (350 μm to 450 μm) were picked from each sample, while 3 to 10 specimens of benthic C. wuellerstorfi (>154 μm) were picked from the samples of 0 cm to 600 cm and 1,680 cm to 1,985 cm in the core. The stable isotope composition was measured by Finnigan-MAT253 mass spectrometer equipped with an automatic carbonate preparation device (Kiel III) at the State Key Laboratory of Marine Geology, Tongji University. The isotopic results were calibrated on the Pee Dee Belemnite scale using the China National standard NBS 19. The SD is ±0.07‰ for δ¹⁸O and ±0.05‰ for δ¹³C, respectively.

For the Mg/Ca analyses, foraminiferal specimens were pretreated following the procedure described by Lea et al. (16), but with additional checking and removing of potential contamination performed under a microscope (66), particularly for P. obliquiloculata. The Mg/Ca measurements were done using an ICP-MS at the State Key Laboratory of Marine Geology, Tongji University and the Department of Earth Sciences, University of California, Santa Barbara (SI Appendix, Table S1). Mn/Ca ratios of G. ruber and P. obliquiloculata were generally below 0.5 mmol/mol and had no correlation with Mg/Ca, suggesting the Mg/Ca values were not affected by postdepositional Mn-rich oxide or carbonate coatings. Sample reproducibility based on 119 replicates for G. ruber and 87 replicates for P. obliquiloculata was 2.5% and 4.9%, respectively.

MD01-2386 Age Model.

The upper 9.8 m of core MD01-2386 is dated by 17 radiocarbon measurements of G. ruber performed in the Leibniz Laboratory of Kiel University (SI Appendix, Table S2). The marine reservoir age ΔR is 40 ± 23 y, calculated by the weighted average of eight modern measurements close to the site of core MD01-2386. The top 5 cm of core MD01-2386 appears as a “mudline” (brownish color corresponding to the existence of abundant oxidized particles), but the core-top (0 cm) age is estimated to be 828 y by the linear regression of the ¹⁴C dating results over the Holocene interval (0 ka to 10 ka) (SI Appendix, Fig. S2). Radiocarbon ages are converted to calendar ages via the calibration curve of Marine13 (67). The depth−age model for the radiocarbon-dated interval is established by the Bayesian statistical methodology of Bacon (68), and the age for 0 cm to 50 cm core depth is linearly interpolated by the estimated core-top age and the age of the first ¹⁴C dating point (at 50.5 cm). The settings and results of the Bacon analysis are shown in SI Appendix, Fig. S2. The average sedimentation rate is 26.6 cm/ka. Tie-point age model uncertainties derived from Bacon range from 7,351 +78/−97 y at 140.5 cm to 36,706 +1,531/−1,176 y at 976.5 cm (SI Appendix, Table S2).

The depth−age model for the lower part (9.8 m to 21.0 m) of core MD01-2386 is established using the δ¹⁸O of benthic foraminifer C. wuellerstorfi and planktonic foraminifer G. ruber. The LR04 benthic δ¹⁸O stack (24) is utilized as a tuning target for the oxygen isotopic stratigraphic correlation. The establishment of the age model of core MD01-2386 is performed in three steps. Firstly, we compared the benthic δ¹⁸O record of MD01-2386 with the LR04 benthic stack (SI Appendix, Fig. S1A). The two benthic δ¹⁸O curves are in very good accordance during the period from 20 ka BP to the Late Holocene, the interval where we have plenty of radiocarbon dates. It is inferred that the deep-water δ¹⁸O variation around site MD01-2386 is generally in phase with the LR04 stack during glacial terminations. Thus, we obtained six age ties to synchronize the benthic δ¹⁸O of MD01-2386 to LR04 for the penultimate glacial termination (SI Appendix, Fig. S1A) from 136 ka BP to 106.5 ka BP.

Then, we compared the G. ruber δ¹⁸O of MD01-2386 with LR04 (SI Appendix, Fig. S1B). Using the alignment of C. wuellerstorfi δ¹⁸O, the G. ruber δ¹⁸O lags LR04 by about 1 ka to 2 ka during glacial Termination II and the transition from MIS 5e to MIS 5d. For the rest below 9.8 m in the core (between 43.5 ka BP and 91 ka BP, and between 141 ka BP and 154 ka BP), the age ties were obtained by tuning the G. ruber δ¹⁸O to the LR04 stack (SI Appendix, Fig. S1B and Table S2). All of the defined age ties of the δ¹⁸O correlations are shown in SI Appendix, Table S2. Finally, the ages between the tie points are given by Bayesian statistical distribution (for ¹⁴C dating) or linear interpolation (for δ¹⁸O-stratigraphic ties) of the most nearby two tie points.

Transient Experiments and the Conceptual Model of Theoretical ΔT.

In order to verify the validity of reconstructed ΔT_G-P changes, we use CESM with T31_gx3v7 resolution (3.75° × 3.75° for atmosphere and nominal 3° resolution for ocean) (69) to simulate the SST and TWT at the site of core MD01-2386. As a spin-up, the CESM was first run for 200 model years under orbital parameters and other boundary conditions in AD 1950. Then the model was integrated for 3,000 model years with the transient orbital insolation forcing of the past 300 ky, in which orbital parameters were advanced by 100 y at the end of each model year according to Kutzbach et al. (70). The outputs of this transient experiment (called CESM_transient experiment) in the last 3,000 model years were used in our analysis. Comparisons between the simulated and reconstructed ΔT are detailed in SI Appendix.

Based on results from this CESM_transient experiment and from modern STC studies, here we further presented a conceptual model for the half-precessional changes of ΔT at site MD01-2386. In modern climate, subtropical wind stress forcing and thermal forcing both can affect the WEP TWT through multiple mechanisms (48, 49, 51), including the STC transport volume changes (71, 72), perturbed temperature advections of the mean STC flow (59, 73), and the fast propagation of oceanic Rossby and Kelvin waves (14, 15, 50). At orbital timescales, we assumed that precessional changes of boreal or austral winter insolation (and associated meridional gradients) will not only induce surface heating anomalies over the STC subduction regions but also modify wind stress conditions from the equator to the extratropics of both hemispheres. Associated STC mechanisms will be simplified as three terms (advection term, volume term, and wave term),

$$\text{Theoretical ∆T}\hspace{0.17em}=\hspace{0.17em}\text{Advection}\hspace{0.17em}+\hspace{0.17em}\text{Volume}\hspace{0.17em}+\hspace{0.17em}\text{Wave}.$$

Here, the advection term represents the impact of subtropical temperature anomalies (perturbed temperature) on the WEP thermocline through the advection of mean STC flow. The Volume term stands for the impact of STC transport volume changes due to subtropical zonal wind stress anomalies, which are assumed as responses to the equator-to-subtropical insolation gradient during November or May. For the Wave term, we took into consideration the fast response of WEP thermocline depth or temperature through wind-driven oceanic wave pathways of the STC. Detailed calculations are given in SI Appendix.

Data Availability.

All of the data necessary to assess the validity of this research are presented in the paper and SI Appendix.