Introduction

Dimethylsulphide (DMS) emitted from oceanic phytoplankton plays an important role in controlling concentrations of sulphate (SO42−) aerosols, which can act as cloud condensation nuclei (CCN)1,2,3. Changes in CCN would influence cloud albedo, a key parameter of radiative forcing1,2,3. Increased SO42− can thus cool the Earth by indirect forcing, in addition to direct forcing owing to increased scattering of solar radiation3. To understand these effects, DMS emissions and their links to climate should be evaluated in a pristine environment2. DMS and its oxidation products, SO42− and methanesulphonate (CH3SO3−, hereafter MSA), are also indicators of primary productivity in the Southern Ocean (SO), which is important because they are closely related to atmospheric CO2 variability through the biological pump4. SO42− and MSA in Antarctic ice cores are therefore useful tools for investigating links between the sulphur cycle and climate.

High concentrations of non-sea-salt (nss) SO42− measured in glacial samples from Vostok ice core drilled in East Antarctica5 (Supplementary Fig. 1) have been interpreted as evidence of enhanced oceanic DMS emissions during glacials, assuming that nssSO42− is mainly of marine biogenic DMS origin. A subsequent study on the same ice core6 reports increased MSA concentrations in addition to nssSO42−, further supporting the interpretation of [5] because MSA originates solely from DMS, whereas nssSO42− can come from other sources7,8. Based on these results, DMS emissions and hence nssSO42− have been believed to exert positive feedback on climate. However, more recent studies refute the positive feedback hypothesis7,8, showing that MSA is modified post-depositionally in the Antarctic interior where accumulation rates are low and does not represent DMS production around Antarctica. Furthermore, two deep ice cores drilled at Dome C (EDC) and Dronning Maud Land (EDML) in East Antarctica (Supplementary Fig. 1) show little change in nssSO42− flux over glacial/interglacial cycles7,8,9, while concentrations increase during glacials. Wolff et al.7 point out that increased nssSO42− concentrations in ice cores from sites with low accumulation rates (e.g., Vostok, EDC, EDML) are mainly caused by decreased accumulation rates in glacials, and can therefore not be interpreted as evidence of increased atmospheric nssSO42−. The nearly constant nssSO42− fluxes at EDC and EDML, which face the Indian and Atlantic Ocean sectors of the SO, respectively, have been interpreted to reflect stable DMS emissions and hence stable marine biogenic productivity in the Antarctic Zone (AZ) of the SO over glacial cycles, assuming that the major source of nssSO42− is DMS7,8,9. In contrast, marine sediment records show that export production decreases in the AZ during glacials but increases further north in the Sub-Antarctic Zone (SAZ) of the SO4. This implies reduced primary productivity in the AZ but increased primary productivity in the SAZ during glacials. The disparity between ice and marine core records has been attributed to differences in marine organisms that contribute to these records7.

The stable sulphur isotopic composition of SO42− (δ34S) provides a useful signature of its origins10,11,12. The δ34S data measured from EDC and Vostok ice cores suggest 4–6‰ lower δ34S for the last glacial than for the Holocene and last interglacials, although the data are scattered and sparse11. This has been attributed to isotopic fractionation during transport, as terrestrial contribution of SO42− has been assumed to be small11. However, surface snow samples from a latitudinal transect between a coastal station (Syowa) and an interior site (Dome Fuji, hereafter DF, Supplementary Fig. 1) show remarkably uniform δ34S in East Antarctica13. The results suggest that net isotopic fractionation during long-range transport is insignificant in East Antarctica and thus δ34S in the ice cores from the East Antarctic interior can be used to infer source contributions. Lower δ34S values in the last glacial11 might be due to an increased contribution of terrestrial SO42− originating from increased terrestrial dust7,8. Consequently, little change in the nssSO42− flux over glacial/interglacial cycles7,8,9 can be caused by increased terrestrial sulphate and decreased marine biogenic sulphate.

In this study, we propose this alternative interpretation of nssSO42− flux and make a revised calculation of DMS-derived sulphate, using new ice core records obtained at DF, spanning the last 720,000 years14,15. On the basis of the revised calculation, we compare the DMS-derived sulphate record from DF with those from EDC and EDML. We find that DMS-derived sulphate fluxes decrease in glacials, which indicates reduced DMS emissions in the AZ of the SO. This suggests that primary production, as well as export production, decreases during glacials, which is consistent with marine sediment records4.

Results

Flux variability and potential sources of nssSO4 2−

We calculated nssCa2+ and nssSO42− from Ca2+, Na+, and SO42− concentrations7,8,9,16 (Supplementary Figs. 2a–c). Low accumulation rates (<30 kg m−2 yr−1 in the present day and <50 kg m−2 yr−1 throughout the last 720,000 years) (Supplementary Fig. 2d) at DF14 indicate that the dominant process for aerosol deposition is dry deposition and that the flux, rather than the concentration in ice, better represents the changes in atmospheric aerosol concentration7,16. The flux of nssCa2+ at DF covaries with that at EDC7,8 and EDML16, indicating high and low values during glacials and interglacials, respectively (Fig. 1, Supplementary Fig. 3); fluxes at DF are 2.0 and 0.6 times those at EDC and EDML, respectively. The dominant source of nssCa2+ is terrestrial dust7,8,9,16 and South America is a major source region for dust deposited in the Antarctic interior17,18,19. Different nssCa2+ fluxes in three records can be explained by their different distances from the South American source region16. Contrary to previous studies on EDC and EDML cores, the nssSO42− flux at DF is not constant (Figs. 1 and 2). The flux increases as δ18O (a proxy for temperature at DF14,15) decreases below approximately −58‰, and increases when δ18O is above approximately −57‰.

Fig. 1
figure 1

Temperature proxies and ion fluxes at Dome Fuji (DF) and Dome C (EDC). a The δ18O (DF)14 and δD (EDC)7,8 records averaged over 1000 years. Marine isotope stage numbers for interglacials are also shown. b Fluxes of nssCa2+ at DF and EDC averaged over 1000 years. c Fluxes of nssSO42− at DF and EDC averaged over 1000 years. See Methods for DF chronology and flux calculations. The EDC fluxes are plotted on the AICC12 timescale53,54 using previously published ion data7,8,9,16 and accumulation rates53,54

Full size image
Fig. 2
figure 2

Variability of nssSO42− flux at Dome Fuji (DF). a DF nssSO42− flux plotted against DF δ18O14,15. Data points represent 1000-year averages. Before averaging, the δ18O depths that differ from the ion data depths have been interpolated to match. Gray bar indicates the lower threshold of δ18O (−58‰), below which the nssSO42− flux decreases with δ18O, and the upper threshold (−57‰), above which the nssSO42− flux increases with δ18O. b DF nssSO42− flux plotted against DF nssCa2+ flux. Data points represent 1000-year averages. The slope of the solid red line (m = 2.4) represents the stoichiometric mass ratio of Ca/SO4 as CaSO4. The dashed red line shows the lower bound of the nssSO42− flux data with m = 2.4. Red and blue dots represent the data for warm and cold periods, respectively, corresponding to the δ18O values above and below the thresholds

Full size image

Potential sources of nssSO42− are marine biogenic DMS, volcanic sulphate, and terrestrial dust7,8. With the exception of a few years following large volcanic eruptions, the volcanic input is estimated to be less than 10% of the present-day and Holocene sulphate budgets7. Oceanic DMS was previously regarded as the dominant nssSO42− source over glacial/interglacial cycles, with only a small input from terrestrial sources7,8,9. However, terrestrial sulphate can be a major source in glacials when the amount of dust increases. Fluxes of nssCa2+ and nssSO42− at DF are correlated during cold periods when δ18O is below approximately −58‰. The scatter plot of nssSO42− against nssCa2+ (Fig. 2b) shows a lower bound whose slope is close to the stoichiometric ratio for CaSO4, indicating that a large proportion of nssSO42− during cold periods exists as CaSO4. This same feature is reported for EDML and a similar but weaker correlation between nssCa2+ and nssSO42− fluxes is reported for EDC20. These observations suggest that a large proportion of nssSO42− exists as CaSO4 during cold periods at EDC and EDML20, as well as DF. Micro-Raman spectroscopic analysis of DF samples from the Last Glacial Maximum (LGM) suggests that a large proportion of Ca2+ exists as gypsum (CaSO4.2H2O)21. Furthermore, analyses of DF samples from the LGM using scanning electron microscopy/energy-dispersive X-ray spectroscopy (SEM-EDS) show that the majority of Ca2+ originates from CaSO422. Although the SEM-EDS analyses indicate that a large fraction of the particles consists of silicate minerals containing Ca, they do not dissolve in water. The majority of Ca2+ measured in this study using ion chromatography (see Methods) should therefore originate from CaSO4.

CaSO4 in DF core could originate from two potential sources. First is primary gypsum, i.e., terrestrial gypsum transported from arid source regions as dust8,9,20,23. Second is secondary gypsum formed by the reaction of CaCO3, one of the major components of terrestrial dust, with marine biogenic H2SO4 or SO220,24,25 during dust transport. If primary gypsum is dominant, the major source of nssSO42− during cold periods should be dust, not marine biogenic sulphate. But if secondary gypsum is dominant, the major source of nssSO4 should be marine biogenic sulphate. So far, secondary gypsum has been considered dominant, assuming limited fractions of terrestrial sulphate7,9,11. A mean sediment SO42−/Ca2+ ratio of 0.19 or 0.18 observed for soils10,12 is often referred to as a basis of a small terrestrial contribution. To define an uppermost limit,9 uses a ratio of 0.5 observed in Sharan dust plumes and suggests a maximum terrestrial contribution of only 16%. However, ratios are highly variable and source-dependent8.

Although the compositions of South American minerals are poorly documented, there is some evidence supporting the hypothesis that SO42−/Ca2+ ratios could be much higher than previously assumed. Soil samples from northeastern and central Patagonia, one of the potential source regions, show SO42−/Ca2+ ratios close to or larger than the stoichiometric ratio of CaSO426,27. Because some of these soil samples have high Na+/Ca2+ and Mg2+/Ca2+ ratios (often much higher than 1), which is not the case for nssNa+/nssCa2+ and nssMg2+/nssCa2+ in the Antarctic ice samples (see [28] for nssNa+/nssCa2+ and Supplementary Discussion for nssMg2+/nssCa2+), such soils may not be the source for CaSO4 in the Antarctic ice cores. However, high SO42−/Ca2+ ratios reported in Patagonia cast doubt on the assumption that the SO42−/Ca2+ ratio of 0.5 observed in Sharan dust plumes gives an uppermost limit9.

Gypsum is a major mineral in evaporites29. A distribution of evaporites has been reported in wide regions of South America30. Large areas of the Puna-Altiplano, one of the potential source regions of dust deposited in the Antarctic interior, are covered by salt-lake beds19, which could be sources of gypsum-rich evaporites, although the compositions of these lakes are poorly documented and could vary substantially29. Giant evaporite belts dominated by halite and gypsum are also found in this region31. Puna evaporites are uniquely characterized by scarce carbonates, whereas sulphates and chlorides are abundant31. Furthermore, ion ratios nssCl−/nssNa+ and nssNa+/nssCa2+ estimated from EDC core suggest a significant contribution of halides mobilized from continental evaporite deposits28. The reaction of CaCO3 with H2SO4 and SO2 is slow20,25,32,33, and only a partial neutralization of clay or carbonate particles has been observed even in areas where SO2 and H2SO4 concentrations are greatly enhanced by volcanic contributions20,34. If this can also be extended to the different conditions over the Southern Ocean, then terrestrial gypsum would be needed to explain the relationship between nssCa2+ and nssSO42− in Antarctic ice cores and may be a major CaSO4 source. To validate our idea, it will be important to establish what source areas could provide such a gypsum-rich source of dust.

Revised calculations of DMS-derived sulphate

To calculate the flux of DMS-derived nssSO42−, the contribution of terrestrial sulphate should be removed. We first subtract the terrestrial nssSO42− fraction as a case for a maximum contribution of terrestrial gypsum to nssSO42−flux. Assuming that the majority of nssCa2+ originates from terrestrial sulphate and that nssCa2+ is a major terrestrial cation, we make a first-order estimate of the marine biogenic sulphate flux by subtracting the CaSO4 contribution (nssCa2+ multiplied by 2.4, the stoichiometric mass ratio of SO4/Ca for CaSO4) from the total nssSO42− flux. The residual nssSO42− is thus dominated by sulphate in the form of H2SO4 and/or Na2SO4. Both H2SO4 and Na2SO4 originate from DMS; the former is directly produced from DMS, whereas the latter is produced by the reaction between DMS-derived H2SO4 and NaCl35 (sea salt and/or terrestrial). The residual nssSO42− flux, a revised marine biogenic sulphate flux, co-varies with the temperature proxy δ18O at DF (Fig. 3) and displays high and low values during interglacials and cold periods in glacials, respectively. Similarly, residual nssSO42− fluxes calculated for EDC and EDML show variability consistent with DF (Fig. 3). Opposite behaviors of marine biogenic and terrestrial sulphate would have led to small variability of the nssSO42− flux over glacial/interglacial cycles. Larger dust input at DF and EDML owing to their proximity to the South American source regions relative to EDC (Fig. 1b, Supplementary Fig. 3a) would have resulted in greater variability in the nssSO42− flux at DF and EDML compared with EDC (Fig. 1c, Supplementary Fig. 3b).

Fig. 3
figure 3

Variability of residual nssSO42− flux at Dome Fuji (DF), Dome C (EDC), and Dronning Maud Land (EDML). a Residual nssSO42− flux at DF and EDC for the past 720,000 years, calculated by subtracting the terrestrial CaSO4 contribution from the nssSO42− flux. The EDC flux is plotted on the AICC12 timescale53,54 using previously published ion data7,8,9,16 and accumulation rates53,54. b δ18O at DF over the past 720,000 years14,15. Gray bar indicates the thresholds (Fig. 2a). Marine isotope stage numbers for interglacials are also shown. c Residual nssSO42− at DF, EDC, and EDML calculated by subtracting the terrestrial CaSO4 contribution from the nssSO42− flux. The EDC and EDML fluxes are plotted on the AICC12 timescale53,54 using previously published ion data7,8,9,16 and accumulation rates53,54. d The δ18O values at DF14 for the past 150,000 years. All ion and δ18O values are averages over 1000 years

Full size image

As stated above, we first subtract the terrestrial gypsum contribution to calculate the marine biogenic nssSO42− flux. However, because nssSO42− could have other sources, we perform a sensitivity test as follows. If a major fraction of nssSO42− originates from evaporites, then other minerals commonly contained in evaporites could also contribute to the nssSO42− flux. We take into account Mg2+ and K+, which could originate from evaporites and exist as sulphates29, as well as the contribution of CaCO3, a major mineral in many of the dust source regions26,27 that likely reacts with HNO3 or NOx instead of H2SO4 or SO2 due to faster reactions20,25,32,33,36 (Supplementary Discussion and Supplementary Fig. 4). Our conclusion that the residual nssSO42− flux decreases during cold periods does not change, although the correlation between residual sulphate and temperature proxy changes slightly (Fig. 4a, b, Supplementary Fig. 5). We also change the nssSO42−/nssCa2+ ratio (R1) assuming that part of CaSO4 originates from the reaction of CaCO3 with marine biogenic sulphate. When we change R1 values, we consider only CaSO4 and ignore other minerals. The same conclusion remains if R1 > 1.2, but fails if R1 < 1.2. For EDC and EDML cores, we consider only the CaSO4 contribution because neither Mg2+ nor K+ data are available.

Fig. 4
figure 4

Relationship between residual nssSO42− flux and δ18O at Dome Fuji (DF). a Residual nssSO42− flux, considering only the terrestrial CaSO4 contribution, plotted against δ18O14,15. b Residual nssSO42− flux considering the contributions of CaSO4, MgSO4, Ca(NO3)2, and Mg(NO3)2 plotted against δ18O14,15. Residual nssSO42− flux and δ18O are averages over 1000 years. Before averaging, the δ18O depths that differ from the ion data depths have been interpolated to match. Straight lines in a and b display results of linear regressions. Correlation coefficients (r) were calculated with sample size (n) = 681 and for significance level (α) = 0.05. c Normalized power spectra of residual nssSO42− flux and δ18O at DF. The residual nssSO42− flux was calculated in the same manner as b. Power spectra were calculated with the Blackman-Tukey method (30% lag) using the Analyseries software package55 (see Methods). To use the software, the raw data were resampled to a 200-yr interval using linear interpolation

Full size image

We calculate the marine biogenic/total nss SO42− ratio (R2) for different terrestrial gypsum contributions (Supplementary Discussion and Supplementary Fig. 6). If we assume that Ca2+ and Mg2+ are major evaporite-originated cations that form sulphate in DF core and that the carbonate hosts of these ions react with HNO3 or NOx rather than H2SO4 or SO2, R2 for the LGM is 0.46. This value is consistent with that estimated from sulphur isotopes (R2 ~ 0.5) assuming no isotopic fractionation11. However, R2 values for interglacials exceed 1, which is implausible. This most likely suggests an overestimation of the NO3− derived from the reaction between carbonates and HNO3 or NOx, because NO3− can also exist as HNO3. If we consider only terrestrial gypsum as a major contributor to terrestrial nssSO42−, R1 = 2.4, 1.5, and 1.3 yield R2 = 0.24, 0.52, and 0.59, respectively (Supplementary Fig. 6) for the LGM. The same R1 values yield R2 = 0.36, 0.60, and 0.65, respectively, for EDC core, and R2 = 0.24, 0.56, and 0.60, respectively, for EDML core. Larger dust input at DF and EDML likely yields smaller R2 values compared with EDC. R2 values for the LGM might be underestimated for large R1 values (~2.4) owing to contribution of marine biogenic sulphate. In any case, a contribution of terrestrial sulphate during glacials is likely much larger than previously assumed, consistent with that estimated from sulphur isotope mass balance11. This leads to a conclusion that DMS-derived sulphate decreases in glacials, although the degree of decrease remains uncertain depending on glacial/interglacial changes in R2 values.

Reduced DMS emissions in the AZ of the SO during glacials

Sulphate aerosol observations at EDC and coastal Antarctic sites display a clear seasonal pattern with a maximum in austral summer37. High surface DMS concentrations and emission fluxes over the modern SO in austral summer have also been reported38,39. The dominant source of biogenic sulphate in the Antarctic interior is thus most likely DMS emitted from the SO. The flux of marine biogenic sulphate deposited in the Antarctic interior would then be controlled by DMS emissions in the source regions, the location of these source regions (i.e., distance to Antarctic interior sites), DMS oxidation chemistry, and depositional processes. As glacial/interglacial changes in oxidation chemistry and deposition are likely to be small, the decreased biogenic sulphate flux during glacials would be caused by reduced DMS emissions and/or longer transport distances40 (Supplementary Discussion). Transport distances depend strongly on the summer sea ice extent around Antarctica40 (Supplementary Discussion), but only limited information is available for glacials. In the Indian Ocean sector, the summer sea ice extent at the LGM was only slightly greater than the present day41, whereas in the Atlantic Ocean sector, the sporadic occurrence of summer sea ice considerably farther north is indicated41 (Supplementary Discussion). Although data from other oceanic sectors are very sparse, Gersonde et al.41 speculate that the summer sea ice field around Antarctica changed from 4 × 106 km2 (present day) to 5–6 × 106 km2 (LGM).

DF and EDC display similar residual nssSO42− fluxes with glacial/interglacial ratios of 1/3 to 1/4 (Fig. 3). The LGM/Holocene ratios (~1/3) at DF, EDC, and EDML are consistent with the MSA flux at Siple Dome (West Antarctica, Supplementary Fig. 1) where MSA can be used as a proxy for marine biogenic sulphur deposition because its post-depositional loss is minimal42. In other words, the four sites facing different sectors of the SO, which includes the Indian sector where summer sea ice extent increased only slightly at the LGM, display similar glacial/interglacial ratios of biogenic sulphur species. The similar ratios would be mainly associated with glacial-interglacial changes in DMS emissions in the SO, and specifically the AZ because it is a major DMS source region for biogenic sulphate in the present-day Antarctic interior38,43 (Supplementary Discussion). The reduced biogenic sulphate fluxes at the LGM could be partly due to the increased transport distances. However, the LGM increase in the summer sea ice extent around Antarctica by 1.25 to 1.50 times41 would only slightly increase the transport distances to the Antarctic interior sites, which are affected by a mixture of air masses from different oceanic sectors44,45 (Supplementary Discussion and Supplementary Fig. 7). Thus, lower biogenic nssSO42− fluxes during glacials indicate reduced DMS emissions in the AZ, suggesting that primary production, as well as export production, decreases during glacials, which is consistent with marine sediment records4.

Discussion

Sea surface temperature (SST), solar radiation, sea ice extent, and nutrient and iron supply can affect DMS emissions1,43,46,47 in the AZ. Power spectra of residual nssSO42− show strong powers in the 41-kyr and 93-kyr bands (Fig. 4c). Powers in similar bands (41-kyr and 98-kyr) are also observed in the δ18O record, which is closely linked to SST and sea ice extent in the AZ. To our knowledge, the relationship between DMS emissions and SST has not been directly investigated. The growth rate of phytoplankton (unicellular algae), however, shows little dependence on SST near the melting point of sea ice48, which is the major source of DMS46,49. Hence, covariance of the residual nssSO42− flux and δ18O record at DF (Figs. 3, 4, Supplementary Fig. 5) does not imply that decreased summer SST is a major cause of reduced DMS emissions. Although the integrated summer insolation at 55°S, the latitude of a major source region of DMS, shows strong spectral power in the 41-kyr band, variability in solar radiation could not be a major cause of the reduced DMS emissions during glacials because it is less than 3% (Supplementary Discussion). The large seasonal difference in sea ice extent during glacials implies large areas of melting sea ice in summer, which would lead to enhanced DMS emissions because melting sea ice is an important DMS source46,49. However, this is not the case because DMS-derived sulphate decreases in glacials (Figs. 3, 4, Supplementary Fig. 5). Thus, the change in winter sea ice extent does not directly affect overall DMS emissions in the AZ on orbital timescales.

Vertical mixing and upwelling appear to dominate the nutrient and iron supply in Antarctic surface waters4. Expanded winter sea ice during glacials would enhance AZ stratification, weaken mixing and upwelling, and decrease the supply of nutrients and iron in winter4. This would decrease the nutrient/iron abundance and thus DMS emissions in summer. Reduced vertical mixing and upwelling during glacials should also reduce the CO2 exchange between the ocean interior and atmosphere, thereby sequestering CO2 into the ocean and leading to decreased atmospheric CO2 concentrations, as is proposed by4. This study also implies that reduced DMS emissions during glacials may reduce cloud albedo, resulting in a negative feedback by biogenic sulphate aerosol-cloud interaction1,2. Although an improved understanding of the precise mechanisms controlling nssSO42− flux variations and their links to climate change is needed, the data provided here can be used to constrain the sulphur cycle and climate models. Ongoing analyses of sulphur isotopes of SO42− (δ34S) in DF core will reduce the estimation uncertainty of DMS-derived sulphate, and enable more quantitative discussion on the interaction between DMS-derived sulphate and climate.

Methods

Ion data

We use ion data from DF1 and DF2 cores after and before 300,000 BP, respectively14. Na+, Ca2+, Mg2+, NO3−, and SO42− were measured from both cores using ion chromatography. In addition, K+ was measured from DF2 core. For DF1 core, we use previously published data50 after re-examination and removal of some data points because of large measurement errors. Fifty-nine samples were newly cut from DF1 core, re-measured, and the new data were added to the earlier dataset. Measurement errors were generally less than 10% but may be higher for low concentrations. For DF2 core, 10-cm-long samples were cut every 0.5 m and measured on two Dionex DX-500 ion chromatographs: one for anions and the other for cations. Measurement errors were estimated to be less than 3%. Sea salt (ss) Na+ and non-sea-salt (nss) Ca2+ concentrations were calculated from Na+ and Ca2+ concentration data using the weight ratios of Ca2+/Na+ for seawater (0.038) and average crust (1.78), as described in previous studies7,8,9,16,51. The nssSO42− concentrations were calculated assuming a sea ice source7,8 of ssNa+. Similar values are obtained if we assume an open ocean source7,8 of ssNa+. Fluxes of nssCa2+ and nssSO42− were calculated by multiplying concentrations by estimated accumulation rates14.

Chronology and accumulation rate estimation

We use the DFO-200652 timescale for the past ~342,000 years and the AICC201253 timescale for the period older than ~344,000 years14. The AICC201253,54 chronology is used for EDC and EDML. The accumulation rates at DF were deduced from the δ18O record by Dome Fuji Community members14, and those at EDC and EDML were taken from [53,54].

Spectral analysis

Spectral analyses were carried out with the Analyseries software package55 (Fig. 4c). Blackman-Tukey spectra (30% lag) using a Bartlett window with a bandwidth of 0.00702905 are shown in Fig. 4c. The amplitudes of the spectra were normalized. The δ18O and residual nssSO42− data used for the spectral analysis were resampled at a 200-yr interval using linear interpolation. For resampling, δ18O data from14 and residual nssSO42− data provided in the Source Data file were used.