Drivers of East Asian millennial-scale climate over the past 400,000 years

Yoshimi Kubota (  yoshimi@kahaku.go.jp ) National Museum of Nature and Science https://orcid.org/0000-0003-2793-6776 Steven Clemens Brown University https://orcid.org/0000-0002-1136-7815 Kyung Eun Lee Korea Maritime and Ocean University https://orcid.org/0000-0002-1548-9991 Ann Holbourn Christian-Albrechts-University Etsuko Wakisaka University of Toyama Keiji Horikawa University of Toyama https://orcid.org/0000-0003-0337-0413 Ryuji Tada University of Tokyo Katsunori Kimoto Research Institute for Global Change

When speleothem CaCO 3 is precipitated under equilibrium conditions, δ 18 O sp is a function of drip water (rainfall) δ 18 O and cave temperature 10 . Rainfall δ 18 O itself is a complicated function of source area, dynamics along the transport path, local rainfall amount, and cloud condensation parameters 11 . When using these records, a primary impediment to interpreting millennial-scale variability is the lack of a means to decompose water isotope proxies into the key constituents of rainfall amount and surface temperature. However, the δ 18 O signal of calcite from planktonic foraminifera (δ 18 O pf ) preserved in nearshore marine sediments can be quantitatively partitioned into sea surface temperature (SST) and the δ 18 O of seawater (δ 18 O w ) using well-established methods of paired Mg/Ca and δ 18 O pf [12][13][14] . Here, we The ECS is a marginal basin located at the rim of the western North Paci c and receives 90% of its uvial input from Yangtze (Changjiang) River, which results in a strong correlation between the runoff and the summer sea surface salinity (SSS) in the northern ECS 16  increasing salinity) is also evident in a direct comparison between the SST and δ 18 O w time-series ( Fig. 1C and S4-S7); near-simultaneous millennial-scale responses were observed during Heinrich events 5, 6, 7a, 8, 9, 10, and 11 during MIS 3 through to late MIS 5 (~40-110 ka), and abrupt climate events between the middle MIS 6 to late MIS 7 (~160-210 ka) and between early MIS 8 and late MIS 9 (~290-320 ka) ( Fig. 1). A relatively broad interval of high coherence with high con dence was also found in early MIS 7 (~230-250 ka), where the phase relationship is ~180° opposite to the other intervals, suggesting a different forcing mechanism.
We infer that the ECS SST and δ 18 O w signals re ect different climate forcings or internal feedbacks during intervals of low coherence but respond to similar mechanisms during times of high coherence. The apparent link between the degree of coupling and glacial state implies that aspects of orbital-scale forcing, such as insolation, atmospheric CO 2 concentrations, and ice-sheet extent, might be involved.
However, a comparison between Northern Hemisphere local insolation, wherein the precession cycle is the most prominent, and the ECS wavelet coherence does not show similarity (Fig. 1B, 1C). This implies that orbital con guration and the associated local insolation (31 °N) changes have little effect on the coherence pattern. For atmospheric CO 2 and ice volume, the high coherence times correspond to 'intermediate' atmospheric CO 2 levels and ice volume conditions, which we de ne as an ice volume equivalent to a −10 to −90 m reduction in the sea level, including both middle-late interglacial and earlymiddle glacial periods.
To consider the possible linkage between the intermediate glacial condition and the high coherence between δ 18 O w and temperature in the ECS, we examined the changes in the spectral power of millennialscale variance in the North Atlantic palaeoceanographic record as a measure of frequency and magnitude of abrupt hemispheric climate changes. A comparison between coherence in the ECS record and the wavelet spectrum of the relative abundance of Neogloboquadrina pachyderma sinistral at ODP Site 983 23 (see Methods) revealed some similarities; periods of high coherence corresponded to a high spectral power in the North Atlantic record (Fig. 1D, 1E). The high spectral power in the North Atlantic wavelet re ects strong millennial-scale climate events, such as Heinrich events and DO oscillations 24  These ndings indicate that abrupt climate change in East Asia is driven by two primary mechanisms: Atlantic perturbations and regional variability (feedbacks) primarily dependent on the magnitude of North Atlantic events, and the mean global climate state. To further examine whether the amplitude of East Asian climate variability is dependent on these forcings, we calculated the standard deviation for each glacial condition (Methods, Fig. S8 increasing salinity. This positive correlation can be explained by the changes in the extent of mixing between cold, diluted water formed near Yangtze River mouth and warm, saline water from the Kuroshio Current. A positive correlation between SST and δ 18 O w is also observed during the Holocene, the early stage of the last deglaciation 25 , and at the maxima of MIS 5 (Fig. 1D), suggesting that ocean mixing has dominated during some full interglacial and glacial periods. In contrast, the periods with a negative correlation between SST and δ 18 O w cannot be reasonably explained by ocean mixing. Rather, we interpret this relationship as re ecting the dominance of the atmospheric signal over ocean mixing.
The Heinrich events that occurred during full glacial stages (e.g., H-1 and H-2) are not correlated with the low SST and high δ 18 O w signals in the ECS; however, MIS 6 is an exception, when SST and δ 18 O w are negatively correlated with high coherence, especially around H-11. Thus, we infer that the mean temperature state during full glacial stages is critical for explaining millennial-scale variability in East Asia, as the full glacial period of MIS 6 was slightly (~1°C) warmer than other glacial stages based on the ECS Mg/Ca-SST record (Figs. S4-S7).
We directly compared our ECS data with the Greenland (NGRIP) record over the last 100 ky to elucidate the mechanism of signal propagation from the North Atlantic to East Asia. Speci cally, we performed wavelet coherence analysis to clarify parameters in the ECS record that are more coherent with the Greenland isotope record (Fig. 2) (Fig. 2). The high level of coherence between the ECS SST and Greenland δ 18 O likely stems from the faster propagation of the temperature signal from the North Atlantic (via the Westerlies) relative to a more local and complex response to rainfall. In this regard, rainfall variability re ects inherent natural variability more so than temperature, which is dominated by external signals (from the North Atlantic) only during high-magnitude events, such as Heinrich events. This hypothesis and more in-depth investigation of the underlying mechanisms could be tested using a coupled climate model simulation that embeds water δ 18 O changes with different greenhouse gas, ice-volume, and sea-level boundary conditions. High pass lter and wavelet coherence Prior to the process of high pass lter, each data set were linearly interpolated at every 100-year interval.
The built-in function of high pass lter in MATLAB ® R2020a was used to remove the orbital scale cyclicity (> 10 ky). Wavelet coherence was obtained using the function 'wtc' on MATLAB ® R2020a 19 . The statistical signi cance level of the wavelet coherence is estimated using Monte Carlo methods 19 . For the comparison between the ECS and Chinese cave δ 18 O, we used the composite δ 18 O record in 22 . For the comparison between the ECS and Greenland, we used the North GRIP (NGRIP) ice core results 29,30 .

Standard deviation for each glacial state
We performed F-test to test the null hypothesis that the two data set comes from normal distributions with the same variance against the alternative that the population variance of one sample set is greater than that of the other. The F-test was conducted using the function 'vartest2' on MATLAB ® R2020a. For all combination between full interglacial (> −10 m of sea level), intermediate glacial (−10 m-−90 m) and full glacial (< −90 m), the null hypothesis was rejected at the 5% signi cance level (p < 0.0001). These results indicate that the variance in intermediate glacial state is the greatest for both SST and δ 18 O w. The standard deviations for each glacial state were illustrated in Fig. S9. We also calculated the standard deviation for each glacial-interglacial cycle (Fig. S8), where the standard deviation was calculated every 20 m of sea level.
Data and materials availability: All data necessary to assess the validity of this research will be presented in the World Data Service for Paleoclimatology, National Centers for Environment Information, National Oceanic and Atmospheric Administration.

Methods References
This is a list of supplementary les associated with this preprint. Click to download.