Millennial-scale variability of East Asian summer monsoon inferred from sea surface salinity in the northern East China Sea (ECS) and its impact on the Japan Sea during Marine Isotope Stage (MIS) 3

Abstract

Color alternations in deep-sea sediment in the Japan Sea have been thought to be linked to millennial-scale variations in the East Asian summer monsoon (EASM), associated with the Dansgaard-Oeschger (D-O) cycles and Heinrich events in the high-latitude North Atlantic during Marine Isotope Stage 3 (MIS 3). In this study, we investigate the variability of sea surface salinity (SSS) in the northern East China Sea (ECS) to evaluate the EASM precipitation in South China and its linkage to the sediment color of the Japan Sea during MIS 3. High time resolution (< 100 years) SSS along with sea surface temperature (SST) records were reconstructed using paired Mg/Ca and the oxygen isotope of planktic foraminifera Globigerionoides ruber sensu stricto from core KR07-12 PC-01 recovered from the northern ECS. The results indicate that millennial-scale variability of the SSS is observed with the amplitude of ~ ± 1 during MIS 3. The variations in SSS are well correlated to D-O cycles and Heinrichs. The EASM precipitation decreases in association with the southward shift of the westerly jet in D-O stadials and Heinrichs, suggesting suppressed moisture convergence along the EASM front associated with weakened North Pacific subtropical high in response to the slow-down of the Atlantic Meridional Overturning Circulation. In a comparison between the SSS in the ECS and the color alternation in the Japan Sea, closely correlated variations between the two records in the interval 44–34 ka indicate that the SSS in the ECS plays a crucial role in regulating nutrient and salinity inflow into the Japan Sea. However, the linkage becomes ambiguous, especially after ~ 30 ka, when the sea level falls toward the level of the last glacial maximum. This shift is associated with changes in sediment facies, confirming that the underlying mechanism in regulating the sedimentary change in the Japan Sea depends on the sea level.

Background

Since high-resolution, continuous oxygen isotope (δ¹⁸O) records of the Greenland ice cores covering the past > 100 ka have demonstrated millennial-scale climate changes during Marine Isotope Stage (MIS) 3 (59–29 ka), the so-called Dansgaard-Oeschger (D-O) cycle. The D-O cycle attracted considerable attention due to its abruptness, large amplitude, high frequency, and potential impact on global climate (e.g., Dansgaard et al. 1993). In the Asian monsoon region, the Quaternary hemipelagic sediments of the Japan Sea are characterized by centimeter- to decimeter-scale alternation of dark and light clay to silty clay, which are bio-siliceous and/or bio-calcareous to a various degree (Tada et al. 1999; Tada et al. 2018; Irino et al. 2018). Tada et al. (1999) demonstrated the occurrence of the basin-wide dark-light color alternations in the sediments of the Japan Sea, which are well correlated with D-O cycles. They hypothesized that these sediment color alternations could be attributed to changes in the nutrient and freshwater influx through the Tsushima Strait due to changes in the Changjiang (Yangtze) River discharge (i.e., summer precipitation in South China). Namely, increased summer precipitation in the Changjiang Basin caused the expansion of the low-salinity and nutrient-enriched coastal waters in the northern part of the East China Sea (ECS) during the interstadials of the D-O cycle, which flowed into the Japan Sea and resulted in the accumulation of organic carbon-rich dark layers (Tada et al. 1999).

Summer precipitation in East Asia is known as Meiyu (China), Baiu (Japan), and Changma (Korea), and is one of the crucial aspects of the East Asian summer monsoon (EASM) system. A series of high-resolution speleothem δ¹⁸O (δ¹⁸O_sp) records in southeastern and central China, which had been interpreted as reflecting EASM intensity/precipitation, demonstrated millennial-scale variations in association with D-O cycles (Wang et al. 2001; Cheng et al. 2016). However, the interpretation of the δ¹⁸O_sp is controversial as studies using analyses of δ¹⁸O mass balance and climate models suggested that the δ¹⁸O_sp record reflects neither local precipitation nor EASM circulation intensity, but rather changes in the moisture source or seasonality of the precipitation (Clemens et al. 2010; Dayem et al. 2010; Pausata et al. 2011; Maher and Thompson 2012). More recently, a trace element study on one of the Chinese speleothems has claimed that millennial-scale δ¹⁸O_sp variations during the last deglaciation were better interpreted as changes in the EASM circulation, instead of the local precipitation (Zhang et al. 2018). By contrast, a proxy of sea surface salinity (SSS) from a marine core provides reliable information on the changes in regional precipitation, through a freshwater inflow from rivers, which can further serve to test the hypothesis by Tada et al. (1999).

The northern ECS is a crucial area, vis-à-vis its position where the freshwater from the Changjiang (Yangtze) River reaches it and the waters from the ECS inflow to the Japan Sea (Fig. 1). Ijiri et al. (2005) reported millennial-scale light δ¹⁸O peaks of planktic foraminifera Globigerinoides ruber sensu stricto (δ¹⁸O_p) in the northern ECS during Marine Isotope Stage (MIS) 3 and stated that these peaks might capture larger freshwater discharge events associated with D-O interstadials. However, their age model did not allow for determination of these timings precisely enough compared to other records, and the effect of temperature changes on foraminiferal δ¹⁸O was not adequately evaluated because they used an alkenone temperature proxy, which may have been recording sea surface temperature (SST) during a different season and at different depth from those of G. ruber s.s. (Kim et al. 2015).

Fig. 1

a Locality map showing site of KR07-12 PC-01 and reference sites. b Site location in the ECS during MIS 3 and the shore area (shaded area) at − 80 m sea level during MIS 3. c SST (Locarnini et al. 2010) and d SSS in July (Antonov et al. 2010) in the ECS

Thus, in this study, we reconstructed the local δ¹⁸O of seawater (δ¹⁸O_w-local), an indicator of SSS, and SST in the northern ECS during MIS 3 using paired δ¹⁸O_p and Mg/Ca of G. ruber s.s., and estimated variability of SSS to test the presence of millennial-scale EASM precipitation oscillations in South China associated with D-O cycles. We used a marine core KR07-12 PC-01 recovered from the northern ECS that covers the time interval since 44 ka, thus covering the late half of MIS 3.

The ECS is a marginal sea in the northwestern Pacific Ocean, and the East Asian monsoons control its seasonal ocean current system (e.g., Ichikawa and Beardsley 2002). The SSS in the ECS changes drastically throughout a year, particularly in the northern part, due to the considerable influence of river discharge caused by the EASM precipitation (Ichikawa and Beardsley 2002). The Changjiang River is the most significant contributor of freshwater to the ECS, accounting for ~ 90% of the total freshwater supplied to the ECS from rivers, and its discharge is characterized by a remarkable seasonal cycle with its zenith in July and nadir in January (Chen et al. 1994; Isobe et al. 2002). Accordingly, the SSS in the northern ECS decreases in summer and increases in winter (Ichikawa and Beardsley 2002). Changjiang Diluted Water (CDW) is formed by Changjiang freshwater mixing with the ambient seawater around the estuary, which then flows eastward into the Japan Sea during the summer (e.g., Ichikawa and Beardsley 2002; Isobe et al. 2002).

Based on instrumental observations, the maximum SST near the studied site is 28.4 °C in August, and the minimum is 17.7 °C in February (Japan Oceanographic Data Center 2004). In contrast, SSS reaches a maximum of 34.7 in February and decreases to a minimum of 33.0 in July, when the maximum discharge from the Changjiang River occurs. The spatial patterns of SST and SSS during summer are characterized by lower SST and SSS in the northwest and higher SST and SSS in the southeast part of the northern ECS (Fig. 1). The lower SSS during the summer suggests that river discharge due to the EASM dominates the seasonal changes in the Kuroshio (Sun et al. 2005); otherwise, the strengthened Kuroshio results in increased SSS during summer, when its volume transport is the largest (Ichikawa and Beardsley 1993; Andres et al. 2008). It was demonstrated that the CDW reaches the easternmost part of the northern ECS across the shelf break during summer, based on ²²⁶Ra and ²²⁸Ra measurements of surface waters (Inoue et al. 2012). Inoue et al. (2012) calculated the relative contribution of the Changjiang freshwater in this area as approximately 2–3% in July and October, based on mass balance calculation of the SSS and ²²⁸Ra concentration of surface water samples collected between 2008 and 2010. Instrumental SSS data around the core site shows a negative correlation with the Changjiang River discharge during the wet season for the past 50 years (Kubota et al. 2015), confirming that the studied site is an appropriate location to reconstruct SSS.

Methods/Experimental

The piston core KR07-12 PC-01 (31° 40.63′ N, 129° 01.98′ E, 736 m water depth) was composed of olive-black or olive-gray, homogenous, well-bioturbated clay. Scattered shell fragments and mottled textures caused by burrows were occasionally identified (Fig. 2). Ash layers were identified at depth intervals of 98.0–118.0 cm and 947.5–990.3 cm, which correspond to the K-Ah and Aira (AT) tephras, respectively, based on the morphology and refractive index of glass shards in each tephra, and heavy mineral assemblages (Table 1). An ash pocket was found at 170.0–172.0 cm, identified as Sz-S (Table 1). The refractive index of each volcanic glass shard was measured using the RIMS-2002 analyzer (Danhara et al. 1992) at Kyoto Fission Track Co. Ltd. The other ashes found at 75.0–84.0 cm and 188.0–190.0 cm are unknown.

Fig. 2

Age-depth relationship of core KR07-12 PC-01 together with 2σ age uncertainties and columnar section. The black circles and lines represent ¹⁴C-based datum and its age model. The red line indicates the fine-tuned age model. The blue crosses represent the tie points in the fine-tuned age model

Table 1 Results of tephra analyses

Approximately 10 cm³ of bulk sediment samples was subsampled at 2.5 cm intervals and washed using a 63 μm mesh to concentrate foraminiferal tests for ¹⁴C dating, δ¹⁸O_p, and Mg/Ca measurements. For ¹⁴C dating, approximately 3–5 mg of the planktic foraminifera Neogloboquadrina dutertrei, from the size fraction > 250 μm, was collected from 15 horizons (Table 2). All of the ¹⁴C data were acquired at the NIES-TERRA AMS facility, National Institute for Environmental Studies (Tanaka et al. 2000; Yoneda et al. 2004; Uchida et al. 2004). We analyzed 244 samples in total for paired δ¹⁸O_p and Mg/Ca. The time resolution of the analyses was ~ 80 years for MIS 3, ~ 130 years for early MIS 2 and Last Glacial Maximum (LGM), and > 300 years for the last deglaciation. For δ¹⁸O_p and trace element analyses, approximately 30–40 individual G. ruber s.s. was picked from 250 to 355 μm size fractions and crushed to homogenize. After being cleaned by milli-Q water and methanol, approximately 50–70 μg of foraminiferal tests was separated and used for the oxygen isotope analysis. The δ¹⁸O_p was measured by two Finnigan MAT 252 Stable Isotope Ratio Mass Spectrometers with a Keil III carbonate device installed at the Mutsu Institute for Oceanography in the Japan Agency for Marine-Earth Science and Technology and at the University of Tokyo. The reproducibility of the measurement was better than ± 0.05‰ (1σ) for δ¹⁸O_p, as determined by replicate measurements of international standards NBS-19 (RM8544 Limestone) and JCp-1 (Coral Porites sp.), provided by the Geological Survey of Japan (GSJ/AIST).

Table 2 Results of AMS radiocarbon and calendar ages

Additional cleaning steps were performed for trace element analyses. The samples were cleaned using the reductive method, including reductive and oxidative cleaning steps published by Boyle and Keigwin (1985), with a slight modification (Kubota et al. 2010). Mg/Ca analysis was performed using a Thermo Finnigan ELEMENT 2 High-Resolution Multi-Sector Inductively Coupled Plasma Mass Spectrometer (HR-ICP-MS) at the Mutsu Institute for Oceanography in the Japan Agency for Marine-Earth Science and Technology. Isotopes of four elements (²⁴Mg, ⁴⁴Ca, ⁴⁸Ca, and ⁵⁵Mn) were analyzed using Sc as an internal standard (Uchida et al. 2008). Four working standards were prepared by successive dilutions of the stock standard solutions to match the concentrations of Ca (approximately 100 ppb, 500 ppb, 2 ppm, and 5 ppm) and Mg (0.05 ppb, 0.2 ppb, 1.0 ppb, and 5 ppb), respectively, which covered the Ca and Mg concentration ranges of all samples (Kubota et al. 2010, 2015). The precision of the replicate analysis of the working standard for Mg/Ca was better than ± 0.09 mmol/mol, corresponding to ± 0.3 °C in temperature scale. Mn/Ca was analyzed to monitor Mn-Fe oxide contamination. Mn/Ca was under 200 μmol/mol in most of the samples (Additional file 1: Figure S1), but exceeded 400 μmol/mol in 5 samples. However, these values are not excluded because there is no correlation between Mg/Ca and Mn/Ca (Additional file 1: Figure S1). In order to examine the homogeneity of the sample, G. ruber s.s. were repicked from 66 randomly selected horizons of core KR07-12 PC-01 and run for Mg/Ca analyses with the same cleaning protocol. The reanalyzed Mg/Ca value at 1392.1–1389.6 cm was 5.0 °C lower than the first one and was discarded. An average of the differences in Mg/Ca values of the replicate analyses for the rest of 65 horizons was 0.16 mmol/mol (~ 0.7 °C). Compared with the Mg/Ca values of U1429 for the same period in the previous study (Clemens et al. 2018), an average of KR07-12 PC-01 was slightly lower by 0.157 mmol/mol than those of U1429. Since the Mg/Ca values of U1429 were confirmed by an international CaCO₃ reference standard, coral Porites standard material JCp-1 (4.199 ± 0.065 mmol/mol, Hathorne et al. 2013), we added 0.157 mmol/mol (~ 0.6 °C) to the original Mg/Ca values of KR07-12 PC-01 to adjust to the value of the U1429.

Although Mg/Ca of foraminiferal calcite is primarily controlled by temperature, a secondary effect of the salinity has been pointed out; Mg/Ca increases when salinity increases (e.g., Lea et al. 1999). In this study, in order to examine the salinity effect on the results to better estimate the SSS variability in the northern ECS, we tested three calibration methods to obtain SST and δ¹⁸O_w-local as follows: (1) a conventional Mg/Ca calibration without salinity correction (referred to “no salinity correction”), (2) an Mg/Ca calibration with salinity correction (“salinity correction”), (3) an Mg/Ca calibration with salinity correction in addition to incorporating the effect of changes in endmember δ¹⁸O_w (“endmember correction”). The SSS is reconstructed in the methods of “salinity correction” and “endmember correction.” Here, we refer to “endmember” as physical values such as SSS and δ¹⁸O_w in the source regions. The purpose of the “endmember correction” is to examine whether or not changes in the δ¹⁸O_w in the source region account for changes in the δ¹⁸O_w in the northern ECS. The results are compared in Fig. 3.

Fig. 3

Comparison among the results with three different calibrations for KR07-12 PC-01: “No salinity correction” (gray), “salinity correction” (red), and “endmember correction” (blue). a δ¹⁸O of planktic foraminifer (δ¹⁸O_p). b SST. c δ¹⁸O_w-local. d SST vs. δ¹⁸O_w-local derived from “no salinity correction” calibration method. e SST vs. δ¹⁸O_w-local derived from “salinity correction” calibration method

First, for the “no salinity correction” calculation, we used an Mg/Ca calibration (Eq. 1) by Dekens et al. (2002), without correction for sediment water depth, and Bemis et al.’s (1998) δ¹⁸O–temperature relationship (low light) (Eq. 2). The δ¹⁸O_w-local was obtained after removing the global sea level effect from total δ¹⁸O_w using Waelbroeck et al.’s (2002) curve (Eq. 3; Thirumalai et al. 2016).

$$ \mathrm{Mg}/\mathrm{Ca}=\exp\ \left(0.09\times T\right)/0.38 $$

(1)

$$ T=16.5-4.8\times \left({\updelta}^{18}{\mathrm{O}}_{\mathrm{p}}-{\updelta}^{18}{\mathrm{O}}_{\mathrm{w}}+0.27\right) $$

(2)

$$ {\updelta}^{18}{\mathrm{O}}_{\mathrm{w}}={\updelta}^{18}{\mathrm{O}}_{\mathrm{w}-\mathrm{local}}-0.008\times \mathrm{sea}\ \mathrm{level} $$

(3)

Here, T, δ¹⁸O_p, and δ¹⁸O_w denote temperature, δ¹⁸O of planktic foraminiferal calcite, and δ¹⁸O of water. δ¹⁸O_w in Eq. 3 includes local and global (sea level) components. Second, for the “salinity correction” calculation, we used an Mg/Ca calibration with salinity correction (Eq. 4; Tierney et al. 2015) instead of Eq. 1. The SST and δ¹⁸O_w-local were derived using a MATLAB script, Paleo-Seawater Uncertainty Solver (PSU Solver) (Thirumalai et al. 2016) with the methodology of Clemens et al. (2018) as explained below. As no regional Mg/Ca calibration exists for the ECS, we employed the Mg/Ca calibration of Tierney et al. (2015) that utilized all available culture data in a multivariate calibration that accounts for both salinity and temperature. We utilized the δ¹⁸O-temperature relationship of Eq. 2 and the ECS seawater δ¹⁸O_w-local-salinity relationship of Horikawa et al. (2015) (Eq. 5), as well as the global sea-level curve of Waelbroeck et al. (2002) (Eq. 3).

$$ \mathrm{Mg}/\mathrm{Ca}=\exp\ \left(0.084\times T+0.051\times S-2.54\right) $$

(4)

$$ {\updelta}^{18}{\mathrm{O}}_{\mathrm{w}-\mathrm{local}}=0.23\times S-7.74 $$

(5)

Here, S represents salinity.

Third, for the “endmember correction” calculation, we replaced Eq. 5 to 6 (LeGrande and Schmidt 2011). This model (Eq. 6) allows slope (a) and intercept (b) to vary through time so that the effect of temporal changes in endmembers is captured. The following two steps were conducted to obtain “a” and “b” in Eq. 6: step 1—estimate of the freshwater δ¹⁸O to obtain intercept “b,” step 2—estimate of the Kuroshio water δ¹⁸O and salinity to obtain slope “a.” Eventually, errors are given based on the Monte Carlo simulation, as will be explained later in this section. Before the calculation, each data set was linearly interpolated (100-year step).

$$ {\updelta}^{18}{\mathrm{O}}_{\mathrm{w}}=a\times S+b $$

(6)

In step 1, we regard the freshwater δ¹⁸O of the Changjiang River as intercept “b” (Kubota et al. 2015). Following Kubota et al. (2015), we utilized a composite record of Cheng et al.’s (2016) δ¹⁸O_sp to estimate the freshwater δ¹⁸O_w (Kubota et al. 2015), on the assumption that the temporal variations in δ¹⁸O_sp reflect the drip water δ¹⁸O in the caves, hence, the river water δ¹⁸O. We used the total δ¹⁸O_w, which includes local and global components, in Eq. 6 instead of δ¹⁸O_w-local to avoid correcting a global component for δ¹⁸O_sp. Here, we define that the variability in intercept “b” is equal to that in δ¹⁸O_sp (Eq. 7). Both sides of Eq. 7 represent subtracting a modern value from values at a given time in the past. We employed − 7.74‰, which is derived from Eq. 5 (Horikawa et al. 2015), as the modern intercept (left side of Eq. 7). For the modern δ¹⁸O_sp value (right side of Eq. 7), an average for the last 1 ka of the Chinese speleothem record (Cheng et al. 2016) was employed. The time series of the endmember δ¹⁸O_w is shown in Additional file 2: Figure S2. The adequacy of these parameters will be discussed in the “Results and Discussion” section.

$$ b-\left(-7.74\right)={\updelta}^{18}{\mathrm{O}}_{\mathrm{sp}}-\left(-9.0\right) $$

(7)

In step 2, we obtained another set of δ¹⁸O_w and S as input data in Eq. 6 to determine slope “a” using a high-time resolution paired Mg/Ca and δ¹⁸O_p data of G. ruber from core MD06-3067 (6° 31′ N, 126° 30′ E, 1575 m water depth) in the western tropical Pacific (Additional file 2: Figure S2; Bolliet et al. 2011). We regard this site as a reference location for the endmember of the Kuroshio water, since the site of core MD06-3067 is affected by the Mindanao Current that shares the same origin of waters, North Equatorial Current, with the Kuroshio. A strong upwelling affects intermediate and subsurface waters in this region, but the upwelled water does not reach the surface upper 75 m (Udarbe-Walker and Villanoy 2001) where G. ruber dwells. Therefore, it is inferred that surface water in this region still holds the origin’s water properties. In fact, the SSS at nearby site MD06-3067 in the western tropical Pacific is 34.0–34.3 during July to September (World Ocean Atlas 2009; Antonov et al. 2010), which is similar to the SSS in the Kuroshio region in the ECS. Although low-resolution Mg/Ca and δ¹⁸O data exists in the north of the bifurcation latitude (~ 14° N) of the Kuroshio at Benham Rise (MD06-3047B; Jia et al. 2018), the site MD06-3067 is a better site in terms of the time resolution, as a major process of this “endmember correction” is to evaluate the influence of the millennial-scale variability in the source region. The δ¹⁸O_w and SSS of core MD06–3067 are derived from PSU Solver with Eqs. 2–4 and δ¹⁸O_w-salinity equation for Palau (Conroy et al. 2017).

In the final step, Eqs. 2–4, 6, and 7 were solved with the input data derived from steps 1 and 2. We performed the Monte Carlo simulation to constrain errors associated with the age uncertainty among the records and temperature, salinity, and δ¹⁸O_w reconstructions propagated from analytical errors for measurements of Mg/Ca and δ¹⁸O_p. In the Monte Carlo simulation, temperature, salinity, and δ¹⁸O_w at a given time “t” were calculated with randomly selected input data (Mg/Ca and δ¹⁸O_p of KR07-12 PC-01, δ¹⁸O_sp, and δ¹⁸O_w and salinity of MD06-3067) in the range of t – 0.5 ka < t < t + 0.5 ka. The analytical error was prescribed to the Mg/Ca and δ¹⁸O_p data, incorporated as a normal distribution (Thirumalai et al. 2016). We set 0.16 mmol/mol and 0.05‰ for the Mg/Ca and δ¹⁸O uncertainties, respectively. Median value and standard deviation (1σ) are yielded at each time step after 500 repetitions of the simulation (Fig. 3). All of the calculations were performed on MATLAB.

Results and Discussion

Age model

Conventional ¹⁴C ages were calibrated to calendar ages using CALIB 7.1 (Stuiver et al. 2016) with Marine13 (Reimer et al. 2013). For the local reservoir correction (ΔR), − 93 ± 69 years, a weighted mean of the 10 nearest locations to the core site (available at http://calib.org/marine/) was applied. Constant linear sedimentation rates were assumed between two adjacent ¹⁴C-based age-controlling points (listed in Table 2) when constructing the age model (Fig. 2). In addition to the ¹⁴C-based datums, published calendar ages of the K-Ah (7.165–7.303 ka) and AT (30.009 ± 0.189 ka) tephras, dated on Lake Suigetsu sediments in SG06 core (Smith et al. 2013), were used as the age datums. In this age model, instantaneous sedimentation was assumed between the top and bottom of the two thick ash layers (K-Ah 98.0–118.0 cm, AT 947.5–990.3 cm). A published ¹⁴C age of S-Sz is 11.295 ± 0.30 ka derived from soil materials below the tephra layer (Okuno et al. 1997), which was calibrated to a calendar age of 12.631–13.768 ka (2σ) in this study using CALIB 7.1 (Stuiver et al. 2016) with IntCal13 (Reimer et al. 2013). We did not apply the recalibrated age of S-Sz in our model since its uncertainty is large. The S-Sz age with our age model, which is 12.4 ka, is younger than the 2σ probability range of the soil samples, probably because the soil materials were sampled below the tephra layer (Okuno et al. 1997).

The age model was further fine-tuned to a composite record of the Chinese δ¹⁸O_sp (Additional file 3: Figure S3 and Additional file 4: Table S1), assuming an in-phase relationship between the climate in the northern ECS and hydroclimate variability over China as mentioned in Clemens et al. (2018). This assumption is justified in the following discussion by correlating the AT tephra layer in our core and its age with others. AT tephra is most precisely dated using the Suigetsu Lake sediment, whose microfossil ¹⁴C ages are generally consistent with the IntCal13 curve for a 31–29 ka interval (Reimer et al. 2013). Although the speleothem ¹⁴C data from Hulu Cave, China, was available back to 27 ka in 2013, a recently published ¹⁴C data set from the same cave confirmed the consistency with IntCal13 between 10 and 33 ka within the uncertainty range of less than ± 0.5 ka (Cheng et al. 2018 and included figures). Correlating the AT tephra with the speleothem record is possible through IntCal13 without consideration of the marine reservoir, since the ages of our core around 30 ka rely on the age determined by the Suigetsu Lake sediment. We expect that the age uncertainty range of this correlation is probably less than 0.5 ka. Compared with the time series of δ¹⁸O_sp, a positive peak in δ¹⁸O_p in KR07-12 PC-01 at 30 ka is highly likely correlated to a positive peak in δ¹⁸O_sp. This agreement allows us to conclude that this assumption is plausible and further infer that in-phase correlation is highly likely for other climatic events during MIS 3. An out-of-phase relationship is especially unlikely, as the time offset between negative and positive peaks is beyond the age uncertainty.

The age-depth curve of the fine-tuned age model was compared with the age model based only on ¹⁴C datums in Additional file 3: Figure S3. The fine-tuned age model did not contradict the ¹⁴C-based ages as the obtained ages with the fine-tuned age model at the horizons of the ¹⁴C datums were within the 2σ uncertainties (Fig. 2 and Additional file 3: Figure S3). The average linear sedimentation rates were approximately 40 cm/ka during MIS 3 and 15 cm/ka during the Holocene. The highest one (~ 50 cm/ka) was found at LGM to deglaciation. Our sampling interval is frequent enough to satisfy the required Nyquist frequency (1/500 year⁻¹ in this case), which is 1/2 of the expected frequency to obtain accurate peaks in a time series data. The time resolution of the ¹⁴C-based datums is 3–4 ka as listed in Table 2. We assume that the age uncertainty of the sampling horizons between the datums is similar to those at the datums based on the linear age-depth relationship in our age model (Fig. 2).

Salinity effect on Mg/Ca calibration

This section discusses the results derived from “no salinity correction” and “salinity correction.” The temporal patterns of the variations in SST and δ¹⁸O_w-local with the two calibrations agree well. However, the result with salinity correction for KR07-12 PC-01 shows less amplitude than that without salinity correction for both SST and δ¹⁸O_w-local (Fig. 3), indicating that salinity effect on Mg/Ca amplifies the variability of SST. This is indicative in Fig. 3, showing that the positive correlation between SST and δ¹⁸O_w-local is stronger in “no salinity correction” than in one with salinity correction for MIS 3. Furthermore, these relationships depend on the sea level for the case without salinity correction, and the positive correlation becomes stronger as the sea level lowers (Fig. 3d, e). Given that the river mouth of the Changjiang River approaches as the sea level lowers, the sea-level dependence of the SST-δ¹⁸O_w-local relationship reflects a change in salinity effect on the studied site in association with sea-level lowering. Alternatively, one would argue that the low sea-level stand prevents water inflow from the ECS to the Japan Sea, resulting in reduced freshwater transport into the Japan Sea and increased salinity variability in the northern ECS. Therefore, we conclude that SST and δ¹⁸O_w-local reconstructions with salinity correction are a better estimate than the calibration without salinity correction.

Calibrating the local salinity variations

We interpret the δ¹⁸O_w-local as a proxy for SSS and will refer to it in the discussion as such. In the modern ECS, a linear salinity-δ¹⁸O_w relationship is confirmed (e.g., Horikawa et al. 2015), reflecting mixing between the freshwater from the Changjiang River and seawater (Kuroshio Water) (e.g., Horikawa et al. 2015; Kubota et al. 2015). Namely, the salinity and δ¹⁸O_w at the studied site can be interpreted in the context of a simplified two endmember mixing model of the freshwater from the Changjiang River and Kuroshio Water (Kubota et al. 2015). By contrast, the following factors are involved in controlling the ECS δ¹⁸O_w in the past: (1) changes in δ¹⁸O_w and salinity in the source region and (2) changes in mixing ratio between the Changjiang River freshwater and Kuroshio Water (≒salinity at the studied site). An effect of the global ice volume is also involved in the changes in endmember δ¹⁸O_w and salinity. As the calibration method of “endmember correction” described in the “Methods/Experimental” section incorporates the changes in endmember δ¹⁸O_w and salinity in the source region through time, the result derived from these calculations gives a better estimate, especially for the salinity. As a result, neither SST nor δ¹⁸O_w with “endmember correction” has a large difference from those with “salinity correction” (Fig. 3), indicating that the “salinity correction” is a reasonably close approximation of those variabilities.

We utilized the δ¹⁸O_sp data as the input parameter to determine the freshwater δ¹⁸O (Eq. 7) in the “endmember correction” method. In this method, we assume that the variability of the freshwater δ¹⁸O is the same as that of the δ¹⁸O_sp. In general, δ¹⁸O_sp can be regarded as a function of drip water δ¹⁸O and cave temperature, unless precipitation of the speleothems is at equilibrium (Hendy 1971). When drip water δ¹⁸O increases and/or cave temperature decreases, the δ¹⁸O_sp increases. The absence of any identification of the kinetic effect on the published δ¹⁸O_sp from the Chinese caves enables us to interpret the δ¹⁸O_sp as reflecting drip water δ¹⁸O and cave temperature (Wang et al. 2008). The assumption we apply to Eq. 7 does not incorporate the temperature component in estimating the endmember freshwater δ¹⁸O. However, the cave temperature variability, if it is deduced from the SST in the ECS, would be correlated with the temporal variation in δ¹⁸O_sp (Additional file 3: Figure S3). Thus, incorporating the temperature component is supposed to suppress the variability of the endmember freshwater δ¹⁸O and the calculation without the temperature component results in giving a maximum range of the variability of the freshwater δ¹⁸O_w. Moreover, high variability in the freshwater δ¹⁸O is expected to decrease the salinity variability or freshwater contribution in the northern ECS (Eqs. 6 and 7). Nevertheless, the SSS reconstruction with “endmember correction” still shows high variability in KR07-12 PC-01 (ca. ± 1) during MIS 3 (Fig. 3), indicating that the changes in endmember δ¹⁸O_w have little impact in the northern ECS. Even though the freshwater δ¹⁸O is highly variable with the amplitude of ± 1‰, its effect is diminished at the studied site due to a smaller contribution of the freshwater to the northern ECS compared with the seawater. Furthermore, neither the variability in δ¹⁸O_w nor the salinity in the western Pacific warm pool has a significant impact on the SSS reconstruction in the ECS.

Changes in SST, local seawater δ¹⁸O, and SSS

The output data from “endmember correction” will be used for discussion hereafter. As planktic foraminifer G. ruber s.s. is abundant in warm seasons in the northern ECS (Yamasaki et al. 2010), we interpret that our results reflect the signals of the warm months (Kubota et al. 2015). The δ¹⁸O, SST, and δ¹⁸O_w-local of KR07-12 PC-01 replicate those of U1429 (Additional file 3: Figure S3), but have higher time resolution. The SST of core KR07-12 PC-01 ranges from 20.2 °C to 24.4 °C, with an average of 22.3 ± 0.7 °C (1σ) during MIS 3 (Figs. 3 and 4), characterized by 1–3 °C amplitude variations at the millennial scale. The SST increases at D-O interstadials 10 and 9 in Greenland and decreases in association with Heinrich event 4 (H4). Four high SST events from 38 to 32 ka can be correlated to D-O interstadials 8–5. Heinrich event 3 (H3) is recognizable, but the magnitude of the decrease in SST is similar to other SST minima. While D-O interstadial 3 is evident in the SST records, D-O interstadial 4 is less distinct. Heinrich event 2 (H2) is characterized by a pronounced decrease in SST by ~ 3 °C. The average SST in LGM is 22.3 °C, which is the same as that in MIS 3. Thus, the LGM is not the coldest period in the last 44 ka. This is a regional phenomenon as a similar trend is found in the middle Okinawa Trough (Chen et al. 2010) and western Pacific warm pool (Stott et al. 2002) based on Mg/Ca-based SST from G. ruber. However, alkenone-based SST, which is interpreted as reflecting an annual mean SST, shows the lowest values during LGM in the northern ECS (Ijiri et al. 2005; Clemens et al. 2018). The proxy-dependent results suggest seasonality in the SST evolution since the last glacial period.

Fig. 4

Comparison of the ECS records and other regions. a δ¹⁸O from NGRIP with age model GICC05 (Svensson et al. 2008). b SST (black) and c SSS (black) of KR07–12 PC-01 with 1σ uncertainties superimposed on δ¹⁸O_sp (orange) from China (Cheng et al. 2016). d L* (purple; Bassinot and Blatzer 2002) and ESR (green; Nagashima et al. 2011) of MD01-2407 in the Japan Sea, recalibrated to IntCal13. e Gulang mean grain size, Loess Plateau, China (Sun et al. 2012). Heinrich events 2 (H2), 3 (H3), and 4 (H4) are highlighted in blue. D-O cycle numbers are shown in the NGRIP δ¹⁸O. The high SSS events in the ECS are highlighted in yellow. The ¹⁴C-based datums (circles) and tie points (crosses) are shown in b for KR07-12 PC-01 and in d for MD01-2407. Red triangles (AT) represent AT tephra

The SSS in the northern ECS is highly variable, ranging from 31.5–34.0 during MIS 3. Millennial-scale variation is recognized in SSS with major negative peaks in association with D-O interstadials 10–5 and 3 (Fig. 4). A predominant positive shift is found at H4, while H3 is less distinct. SST and SSS tend to covary on the millennial scale (H4, D-O 8, D-O 5, and H3). Contrary to the other high SSS events in association with H4 and H3, H2 is characterized by a negative shift, which suggests an alternative underlying mechanism of the EASM precipitation.

Regional and global factors affecting the Japan Sea records

Temporal variation in lightness (L*) of the hemipelagic sediments in the Japan Sea shows a millennial-scale temporal change similar to δ¹⁸O of the Greenland ice cores (Tada et al. 1995; Wang and Oba 1998; Tada et al. 1999). The L* of the sediment principally reflects organic carbon content (C_org), which is controlled by primary production and burial rate, except for during the lowest sea-level stand, such as LGM, when L* is affected by pyrite content (Tada et al. 1999). Tada et al. (1999) suggested that changes in freshwater discharge from the Changjiang and Yellow Rivers, and consequent changes in the spatial extent of the nutrient-rich and low-salinity ECS coastal water, were the main cause of changes in a nutrient influx into the Japan Sea, and primary production there during MIS 3. The increased influence of the ECS coastal water not only enhanced primary productivity, but also reduced deep-water ventilation, leading to the development of the anoxic bottom waters and deposition of the dark layers (Tada et al. 1999; Irino et al. 2018). This hypothesis can be tested by comparing the temporal variations in the SSS in the northern ECS with L* of the hemipelagic sediments in the Japan Sea (Watanabe et al. 2007). While the nutrient and salinity flux into the Japan Sea is regulated by the EASM, the strength of the East Asian winter monsoon (EAWM) is a crucial factor to form a high-oxygenated deep-water mass called the Japan Sea Proper Water (JSPW) (Suda 1932; Nitani 1972; Ikehara and Fujine 2012; Gamo et al. 2014). The cold winds of the winter monsoon cool the surface water along the far eastern coast of Russia and promote the formation of sea ice, which increases the sea surface density to form the JSPW (Suda 1932; Nitani 1972). In addition to the factors described above, sea-level change controls the influx of the ECS waters into the Japan Sea as the narrow and shallow Tsushima Strait prevents the inflow of the ECS waters at a low sea-level stand. As the Japan Sea is connected to other ocean basins only through shallow (< 130 m in sill depth) and narrow (< 90 km in width) straits in addition to the Tsushima Strait, the sea-level change and resulting restriction of the seawater inflow from other basins are crucial to control the SSS in the Japan Sea. Thus, the following three climatic/oceanic factors that potentially control the sediment color of the Japan Sea are discussed in this section: (1) SSS in the northern ECS, (2) EAWM strength, and (3) global sea-level change. Factors 1 and 3 regulate the nutrient influx into the Japan Sea, while all three factors are involved in the deep-water formation and decomposition of the organic material in the sediment-water interface on the sea floor.

In this study, we used a high time resolution (1 cm interval) L* record of core MD01-2407 in the Japan Sea (Bassinot and Blatzer 2002), which was recalibrated to calendar ages using Marine13 (Additional file 5, Additional file 6: Figure S4, and Additional file 7: Table S2), for comparison among the records (Fig. 4). The temporal resolution of the 1 cm interval data is approximately 120 years. In the intervals 44–34 ka (D-O11–7) and 30–28 ka (H3–D-O 4), the Japan Sea L* profile varies in phase with the northern ECS SSS. The dark (light) layers are correlated to the low (high) SSS in the northern ECS. Since AT tephra is found in both cores, the timing of each environmental change in both seas can be constrained (Fig. 2). In the Japan Sea, AT tephra is found at the base of the light later (H3), which is correlated with one of the high SSS events in the ECS. Based on the comparison above, the changes in the SSS in the northern ECS and consequent changes in nutrient and salinity input into the Japan Sea had been a dominant factor altering the sediment color in the Japan Sea in these intervals. The variations in SSS in the northern ECS and associating surface water density in the Japan Sea had been amplifying the variability of the sediment color through the process of production of deep water (Tada et al. 1999). The variations in the EAWM probably had also enhanced the variability of the sediment color during this time period. In fact, a high-resolution quartz grain size record of loess from Gulang section in the Chinese Loess Plateau, which is an indicator of EAWM, shows a good correlation with the L* during the interval 44–34 ka. By contrast, the millennial-scale linkage between the northern ECS and Japan Sea became less clear after 34 ka, which is probably linked to the sea-level lowering. Lambeck et al. (2014) have reported a rapid sea-level fall (by ~ 40 m in less than 2000 years) at ~ 30 ka, following the gradual fall starting from ~ 35 ka. The rapid and substantial fall in sea level from ~ − 80 to ~ − 120 m at 30 ka would have reduced the water exchange between the northern ECS and Japan Sea from 20% of today to almost zero (Additional file 8: Figure S5). Previous studies have revealed that the SSS in the Japan Sea had been lowering toward LGM, which was inferred from a shift to low values in planktic foraminiferal δ¹⁸O in the Japan Sea after 30 ka (Oba et al. 1991; Kido et al. 2007; Sagawa et al. 2018). The sediment facies of the dark layers also shifted from bioturbated layers interbedded by fine laminations to finely laminated layers in association with the decrease in the SSS in the Japan Sea around 28 ka (Watanabe et al. 2007). This facies change is accompanied by an increase in S/C_org, suggesting the euxinic condition in the bottom water (Watanabe et al. 2007), likely caused by the poor ventilation of the bottom water and not by the increase in the primary productivity, as the nutrient input from the ECS was reduced significantly (Watanabe et al. 2007). The comparison of the SSS in the ECS and L* in this study indicates that they are not correlated well in this interval, suggesting the millennial-scale variations in the ECS SSS had almost no impact on the Japan Sea at the lower sea level. Thus, there seems to be a threshold around the sea level of − 80 to − 120 m from a mode where the Japan Sea responds well to the ECS variation to the other mode where it does not. Meanwhile, one still can find interbedded light layers, such as H2, after 28 ka (Watanabe et al. 2007). These light layers might be caused by a different mechanism from those during MIS 3. We speculate that deposition of the light layers might be involved in a millennial-scale sea-level rise that increases the ECS water influx to increase the density of the Japan Sea. In this case, the deep water ventilation plays a more significant role in controlling the L*. Although the timing of the rapid sea-level rise is an issue (Yokoyama and Esat 2011), based on studies on an oxygen isotope record from the Red Sea and isotopes in corals from Papua New Guinea, it is now accepted that 15–20 m, and possibly up to 30 m, of sea level increases is associated with Heinrich events (Broecker 1994; Siddall et al. 2003; Yokoyama and Esat 2011), or with the warm period after the termination of a Heinrich event (Arz et al. 2007). A change in sea level of 15–20 m corresponds to over 10% change in the cross section in the Tsushima Strait, potentially leading a massive change in the water volume exchange between the two ocean basins.

Summer precipitation variability and westerly jet in East Asia

The most prominent feature in our new records is that the SSS in the ECS varies in association with the D-O cycles, which can provide fundamental information on summer precipitation changes under the EASM system. Although the Chinese δ¹⁸O_sp is one of the most well-known proxies from the EASM region, the interpretation of the Chinese δ¹⁸O_sp is still controversial on a millennial-scale (e.g., Zhang et al. 2018). In the earlier stage of the studies on δ¹⁸O_sp, the δ¹⁸O_sp were interpreted as the amount of summer precipitation at a given cave location (Wang et al. 2001). Subsequently, other interpretations were proposed: the fraction of water vapor removed from air masses along the moisture trajectory between the tropical Indo-Pacific and the cave sites (Hu et al. 2008) or isotopic fluctuation in the moisture source region (Pausata et al. 2011). By contrast, the most recent study argued that the δ¹⁸O_sp is better understood as a measure of the large-scale monsoon circulation, not reflecting the local precipitation (Liu et al. 2014). This interpretation is also put forward in a study on trace elements of speleothems, which suggests a wetter condition in H1 and Younger Dryas during the last deglaciation (Zhang et al. 2018). In their discussion, when the δ¹⁸O_sp becomes heavier, which is traditionally interpreted as “weak monsoon,” the local summer rainfall increases in the Changjiang River catchment basin (Zhang et al. 2018). Our new SSS result indicates that most D-O stadials and H4 and H3 intervals correspond to less EASM precipitation and D-O interstadials correspond to more EASM precipitation. Our result contradicts what was claimed in Zhang et al. (2018), suggesting that the millennial-scale summer precipitation response might depend on a global climate setting, such as sea level. In fact, H2 in the ECS, characterized by the decreased SSS, is different from other stadials and Heinrichs. A complex response in the summer precipitation was also suggested based on pollen record from Lake Suigetsu, central Japan, which indicates a wetter condition during H1 compared to the following Bølling-Allerød (Nakagawa et al. 2006).

The SSS variation in the ECS and its relationship to the westerly jet will give us a new insight into the variability of summer precipitation and its relationship to monsoon circulation itself. Today, the EASM front migrates northward from South China from May and reaches the North China-Inner Mongolia regions in August. From the meteorological aspect, the northward migration of the EASM front follows the northward shift of the westerly jet over East Asia on a seasonal scale (Sampe and Xie 2010). Therefore, a proxy of the position of the westerly jet is regarded as an indicator of the EASM circulation. On an interannual time scale, the negative correlation between EASM rainfall in the Changjiang River basin and the summer southerly wind intensity in the northernmost region of the EASM was observed (Jiang et al. 2008); the EASM circulation was stronger, and the Changjiang River basin had less precipitation. A similar relationship was observed on a millennial-scale during the Holocene, based on a proxy of the westerly jet and summer precipitation records from inland China and ECS (Nagashima et al. 2013). According to Nagashima et al. (2011), the position of the westerly jet has varied in harmony with D-O cycles in MIS 3, based on the electron spin resonance (ESR) signal intensity of quartz in the fine silt fraction of the Japan Sea sediments, which is interpreted as a proxy for dust provenance, sourced either from the Taklimakan Desert or the Mongolian Gobi (Fig. 4). The higher (lower) ESR signal intensity during stadials (interstadials) suggests the southern (northern) position of the westerly jet and weaker (stronger) EASM circulation (Nagashima et al. 2011). In the interval 44–34 ka, the increased EASM precipitation in the Changjiang River basin followed the northward shift of the westerly jet position (EASM front), which was opposite to what was observed on a millennial-scale during the Holocene (Nagashima et al. 2013) and on an interannual to decadal scale today (e.g., Jiang et al. 2008).

Global context of East Asian monsoon variations

A comparison of the Chinese δ¹⁸O_sp record and loess grain size indicates that the EASM and EAWM are coupled on a millennial time scale during the interval 60–34 ka (Sun et al. 2012). Their variations are well correlated to the North Atlantic climate variability as well (Sun et al. 2012). In the study of Sun et al. (2012), North Atlantic water-hosing experiments, using the Community Climate System Model Version 3 to mimic the Heinrich or stadial events, suggest a dynamical response of the EASM and EAWM systems to the suppressed Atlantic Meridional Overturning Circulation (AMOC) and cooling in the North Atlantic. The freshwater forcing increases the latitudinal temperature gradient and induces stronger EAWM during winter (Sun et al. 2012). During summer, cooling in North Atlantic results in a southward shift of the Intertropical Convergence Zone, followed by a stronger Walker circulation and weaker EASM (Zhang and Delworth 2005). The moisture convergence along the EASM front and precipitation in the Changjiang River basin is suppressed as the North Pacific subtropical high weakens in response to the freshwater forcing during summer (Sun et al. 2012). However, an opposite summer precipitation response is suggested in a numerical simulation using the Community Earth System Model (Zhang et al. 2018). In this simulation, a stronger meridional temperature gradient, and resultant southward-shifted and strengthened westerly jet, leads to enhanced convection along the slope of the Tibetan Plateau, increased precipitation in southern China, and decreased precipitation in northern China. Thus, the precipitation response in the East Asian monsoon region depends on the climate models, being not constrained by the numerical models as depicted in Kageyama et al. (2013). Conversely, our records give a constraint on the mechanism of the EASM response to AMOC variability during MIS 3, suggesting that the decreased precipitation in Heinrichs is caused by the reduction of the moisture supply from the surrounding oceans, probably linked to the weakening of the subtropical high in the North Pacific (Sun et al. 2012). We infer that, under this mechanism, the migration pattern of the westerly jet and associated position of the EASM front do not regulate the special summer precipitation distribution in the same manner as today, but moisture budget in the atmosphere plays a dominant role. However, an alternative explanation is required for the opposing precipitation trend, such as H2 and H1, when the summer precipitation pattern is opposite to what is observed for MIS 3. A recent study points out that the internal climate variability is important in the rainfall variation in East Asia’s mid-latitude on an interannual and decadal scale today (Ueda et al. 2015). Internal variability is the natural variability of the climate system that occurs in the absence of external forcing, and includes processes intrinsic to the atmosphere, the ocean, and the coupled ocean-atmosphere system (Deser et al. 2010; Deser et al. 2012). Interannual and decadal scale variabilities arise from the intrinsic variability of the system due to a random stochastic process and results from dynamic and thermodynamic interactions of the coupled ocean-atmosphere system (Deser et al. 2012). Ueda et al. (2015) argue that today’s interannual and interdecadal precipitation pattern over mid-latitude East Asia is caused by a combination of the atmospheric internal variability and SST in the western Pacific and Indian Ocean. On a millennial time scale, the EAWM variability after 30 ka, decoupled to the North Atlantic climate (Sun et al. 2012), appears to lack the forcing outside of the East Asian climate system, perhaps resulting from the increased intrinsic variability of the EAWM. One would argue that the increased summer precipitation in southern China in H2 and H1 is interpreted as reflecting the increased convection along the slope of the Tibetan Plateau, induced by the longer placement of the EASM front in the southern position (Zhang et al. 2018). Alternatively, however, the summer precipitation response in H2 and H1 may be explained in the context of the random stochastic processes in the EASM system as well as the EAWM. In the latter case, the overall monsoon system in East Asia would randomly vary, rather than responding to the North Atlantic forcing during MIS 2. Either way, the opposite millennial-scale EASM precipitation pattern during MIS 2 leads to a conclusion that the regional precipitation is not a dominant factor controlling the millennial-scale Chinese δ¹⁸O_sp in this period.

Conclusion

We reconstructed SST, δ¹⁸O_{w -local}, and SSS using the paired Mg/Ca and δ¹⁸O_p in the northern ECS. To better estimate the SSS variability, we tested the three calibration methods to reconstruct these records. The SST record derived from the Mg/Ca calibration without salinity correction shows higher variability than the one with salinity correction, and its variability depends on the sea level. The SSS reconstruction is not significantly affected by the endmember δ¹⁸O_w and salinity variability in the source regions, indicating that the modern local δ¹⁸O_w-salinity calibration is applicable to the MIS 3 case. The most prominent feature of our SSS record is the millennial-scale variations in association with the D-O cycle in Greenland and the Chinese δ¹⁸O_sp, where high SSS events coincide with D-O stadials and Heinrichs, while low SSS events coincide with D-O interstadials. The comparison of the SSS in the northern ECS, the proxy of the EAWM, and the sediment color in the Japan Sea reveal that the changes in nutrient and salinity flux in the Japan Sea induced by the Changjiang River discharge in addition to the strength of the EAWM are likely the primary factors determining the surface productivity changes manifesting as color changes in the sediments of the Japan Sea when the sea level was higher than ca. − 80 m. The glacioeustatic sea-level changes reduced the influx of the ECS coastal water entering into the Japan Sea, especially after ~ 30 ka, at which the linkage between the SSS in the northern ECS and the sediment color of the Japan Sea became less clear.

The SSS variation in the ECS and its relationship to the westerly jet indicates that the decreased EASM precipitation in the Changjiang River basin followed the southward shift of the westerly jet position, which is opposite to what was observed on a millennial-scale during the Holocene and interannual time scale today. We infer that the moisture convergence along the EASM front and overall moisture budget in the atmosphere, which would be altered in association with the D-O cycles and Heinrichs, plays a dominant role in controlling the EASM precipitation in southern China during MIS 3. The precipitation response during MIS 2 is opposite to MIS 3, suggesting the alternative mechanism involved at different global climate backgrounds.