Introduction

Westerlies are the prevailing winds that blow from west to east between latitudes of 30° and 60° in both hemispheres and are a key component of global atmospheric circulation1. They play an essential role in transporting heat and moisture, and in balancing the energy distribution between low- and high-latitude regions2,3,4,5. The westerlies have shifted poleward since the 1960s in response to global warming, which caused more severe midlatitude weather events, such as extreme storms and precipitation6,7,8. The increasing effects of such extreme events make it necessary to understand how the westerlies responded to different climate states. The last glacial cycle (~115–11.6 ka) was characterized by extreme millennial climatic oscillations between cold and warm states9, and the prevailing paradigm attributes these millennial climatic oscillations to AMOC reorganizations10,11, associated with massive iceberg discharges during Heinrich (H) stadials9,12. This study investigates how these oscillations affected the downstream behaviour of the westerlies as far away as the North Pacific Ocean.

So far, geological studies of northern hemisphere westerly variations in the last glacial cycle have relied mainly on terrestrial and marginal sea sediments, using records such as those from the Chinese Loess13, Lake Qinghai14, Japan Sea15, and Philippine Sea16. Studies indicate that stronger westerlies usually occur in cold stadials17, and vice versa. However, how the westerly path changed during the last glacial cycle is still debated. Some studies suggest that the main westerly axis shifted southward during stadials14,15 while others suggest the opposite view13. This controversy can be attributed largely to limitations of proxy evidence in the studied regions; unambiguous proxy indicators of the westerlies over Asia are difficult to obtain because they are also affected by the monsoon circulation16,18,19. In contrast, atmospheric circulation over the North Pacific Ocean (NPO) is dominated by the westerlies and past westerly behaviour can be inferred from NPO aeolian dust records directly or from palaeoceanographic records indirectly20,21,22,23,24. Considering the generally low sedimentation rates and the challenge of dating sediments deposited below the calcite compensation depth25, few late Quaternary sedimentary records from the NPO have been explored26,27. This hinders understanding of millennial-scale westerly behaviour and the underlying driving mechanisms during the last glacial cycle.

Core NP02 (40.48°N, 150.10°E, 5,177 m water depth) is located downwind of Asian dust source areas and adjacent to the main westerly axis, which enables sensitive detection of westerly shifts (Fig. 1). A recent study suggested that core NP02 sediments were deposited at an average of ~11.6 cm kyr−1 (ref. 28), so this core is expected to capture westerly responses to abrupt climate transitions between warm interstadials and cold stadials within the last glacial cycle. Here, we present a dust record for core NP02 over ~21–57 ka, and we assess westerly path changes that are compared with Transient Climate of the Last 21,000 Years (TraCE-21ka) simulation results to further assess possible underlying mechanisms.

Fig. 1: Schematic map of continental Asia and the North Pacific Ocean.
figure 1

Locations of core NP02 (40.48°N, 150.10°E, red circle) and other cores mentioned in this study are indicated. The primary aeolian dust source (yellow) in central Asia is from the 1:100,000 Tarim River Basin desert distribution dataset provided by the National Tibetan Plateau Data Center (http://data.tpdc.ac.cn). Blue arrows indicate the wind vector (unit: m s−1) at 200 hPa in the spring between 1979 and 2020, provided by the NCEP/DOE Reanalysis II data (https://psl.noaa.gov).

Full size image

Results and discussion

Core NP02 age model reconstruction

An initial age model for core NP02 is presented in ref. 28. Here, we tuned the core NP02 geomagnetic relative paleointensity (RPI) record to RPI variations recorded in subarctic Pacific Ocean cores GC1 and PC729,30 (Supplementary Fig. 1a). Depth and age uncertainties for each tie-point are shown in Supplementary Table 1. Tie-points T2 and T6 are identified as the Rockall and Laschamp excursions in ref. 28 with ages of 26,525 ± 175 years and 41,100 ± 350 years ago, respectively31,32. Age uncertainties of tie-points T1, T3, T4, and T5 are estimated as the mean squared estimates of the reference chronology uncertainty. Tie-point depth intervals are represented by Gaussian functions, with uncertainties expressed as ±1σ.

The reconstructed age-depth model is shown in Supplementary Fig. 1b. For the intermediate part (18–395 cm) of core NP02, ages are estimated from the probability density functions using the age constraints listed in Supplementary Tables 1 and 2 with the MATLAB Undatable software33. For the upper (<18 cm) and lower (>395 cm) parts of core NP02, ages are obtained by extrapolation of linear regressions over the 23.5-26.5 ka and 45.8-54.6 ka intervals, respectively. Given that there are no independent age constraints for the upper (<18 cm) and lower parts, we have tested several fitting methods (Supplementary Fig. 2). Second-order polynomial fitting also works well for the upper parts, and age offsets between the linear regression and second-order polynomial fitting are less than 0.34 kyr. However, the second-order polynomial fitting gives unrealistically young ages at the 396–400 cm depth interval. Therefore, the second-order polynomial fitting is not used here. Moreover, we focus on the well-constrained parts of core NP02 (18–395 cm), so that the upper and lower parts with limited age constraints do not affect this study.

Dust proxy

High-coercivity magnetic minerals – hematite and goethite – usually occur as nanoparticles in the Asian interior24,34. They have long atmospheric residence times and can be transported over long distances; therefore, they have been studied frequently as dust proxies35,36,37,38,39. The ‘hard’ isothermal remanent magnetization (HIRM) is used widely to evaluate the high-coercivity magnetic mineral concentration40,41,42,43,44. However, its interpretation can be complicated by changes in the properties of high-coercivity magnetic minerals, such as the coercivity distribution and isomorphous cation substitution45,46,47. The L-ratio enables semi-quantitative identification of coercivity variations46. Samples from the Spfa−1 and Kt-3 tephra layers in core NP02 (ref. 28) have relatively high and low L-ratio values, respectively; some neighbouring samples also have anomalous L-ratio values. This is likely to result from admixed volcanic ash. After excluding tephra samples and those with anomalous values, L-ratio variations fall within a relatively narrow range (shaded area in Supplementary Fig. 3a), which indicates that the Al substitution degree is relatively stable (Supplementary Fig. 3b). Moreover, the magnetization acquired above 300 mT has a linear relationship with IRM carried by component 3 of IRM decomposition results28 (Supplementary Fig. 4), which further supports that HIRM can be used to indicate hematite and goethite concentration variations in core NP02 sediments. Goethite is often volumetrically more abundant than hematite in sediments. For example, the goethite concentration is 2 to 3 times higher than that of hematite in loess-palaeosol sequences on the Chinese Loess Plateau48. However, its concentration is difficult to measure because <10% of its IRM is thought to be acquired at applied fields up to 3 T (ref. 49,50). Therefore, HIRM provides more information about hematite content than goethite content. In core NP02, hematite variations are identified by IRM acquisition curve decomposition28. Hematite does not have an authigenic origin in the NPO38 and the studied core is far from sources of riverine and ice-rafted sediment, so hematite is treated as aeolian in origin. Thus, HIRM is used here as a hematite dust concentration proxy for core NP02.

North Pacific Ocean dust content changes and mechanisms for variability

NPO dust is transported long distances by the westerlies in the middle and upper troposphere20,51,52, and its content variations are influenced by the location and spatial extent of dust source regions, the westerly dust transport path and efficacy, and the dust deposition process (i.e., dry deposition vs wet removal). The NPO is sedimentologically simpler than other oceans, with the Asian interior serving as the major dust source where dust storms occur mainly in spring throughout the Cenozoic22,53,54,55, as verified by a La-Th-Sc ternary diagram for the siliciclastic fraction in core NP02 (Supplementary Fig. 5). Extremely low kaolinite contents (<4%) throughout core NP02 indicate a lack of intense Asian interior chemical weathering56 (Supplementary Fig. 6d). The smectite/(illite + chlorite) ratio does not have an obvious overall trend, with variations around 0.77 (Supplementary Fig. 6e). This lack of variability indicates that the Asian interior climate was relatively stable and arid throughout the 21–57 ka interval.

Interpretations of relationships between source regions and the climatic significance of dust remain controversial. Increased aeolian dust contents have been traditionally attributed to increased aridity. For example, increasing hematite contents at ~2.5 Ma were attributed to progressive Asian continental aridification38,57. Alternatively, it has been suggested that enhanced dust contents were caused by increased sediment availability associated with increased snow and glacial meltwater that resulted from climatic cooling39. In this interpretation, increased aeolian dust inputs resulted from the provision of renewable sources of erodible silt that was later deflated as dust during arid climatic phases. In our results from NP02, the source area weathering regime seems to have remained constant, and neither HIRM nor the smectite/(illite + chlorite) ratio suggests any evident relationship as required in the interpretations in ref. 42 or ref. 39 (Supplementary Fig. 6e and f). Hence, we rule out dust source region variations as a primary driver of dust content changes in core NP02.

Instead, the westerlies can play an important role in controlling the observed dust content changes. Considering that westerly intensity has no obvious effect on late Quaternary dust content variations over the NPO20,58,59, we propose instead that westerly transportation pathway changes explain the dust variations in core NP02. Tropical tropospheric warming and/or polar stratospheric cooling increase the meridional temperature gradient60, which will affect the location of baroclinic instability that in turn affects westerly strength and movement60,61,62. For example, high-latitude North Atlantic cooling through AMOC slowdown has modulated Holocene westerlies over the western Mediterranean region63. On millennial timescales, abrupt dust content decreases in core NP02 are identified at approximately 24, 39, 48, and 54 ka, at times of H2, H4, H5, and H5a (Fig. 2d and Supplementary Data 1), as recorded in Greenland ice cores (Fig. 2a and b). The significance of these features has been tested (Supplementary Note 1) and, importantly, they all have ages that fall near multiple age tie-points or, for the oldest feature, close to a well constrained tephra age (Fig. 2d and Supplementary Fig. 1b). Studies indicate that these abrupt glacial climate changes were associated with major AMOC slowdowns64,65,66,67. The AMOC transports heat from low to high latitudes, and its slowdown reduces meridional ocean heat transport by as much as ~60% to the north of 40°N (ref. 68). Moreover, modelling studies have identified that AMOC slowdown affects the midlatitude NPO through oceanic and atmospheric teleconnections69,70,71,72. Therefore, AMOC slowdowns are expected to affect the westerly path and cause dust variations in core NP02.

Fig. 2: Westerly and North Atlantic climate records over the 60-20 ka interval.
figure 2

a, b GISP2 δ18O and Ca2+ records on the GICC05modelext chronology105, respectively. c Sediment 231Pa/230Th; light and dark green curves represent records from ODP Site 1063 over the 20–30 ka interval106 and core CDH19 over the 25–60 ka interval66, respectively. d Hard isothermal remanent magnetization (HIRM) for core NP02, plotted with error bars representing age uncertainties (1σ). Brown dots with error bars below Fig. 2d denote age constraints with 1σ uncertainties used in this study. e The median grain size of the Yumin loess section13. f Westerly climate index (WI, flux of >25 μm fraction) from Lake Qinghai14. The dashed line in (c) denotes the 231Pa/230Th production ratio of 0.093 in sea water88. Locations of Dansgaard-Oeschger interstadials (black number) and Heinrich (H) events are indicated, after ref. 91.

Full size image

Similar to H2, 4, 5, and 5a, the Younger Dryas (YD) and H1 events are also considered to have been associated with major AMOC reduction73,74,75,76,77. Here we use outputs for the YD and H1 events from TraCE-21ka model experiments, with both full and single forcings (Supplementary Table 3), respectively, to evaluate the impacts of westerly changes on Asian dust transportation. We note that the role of orbital forcing may be underestimated as the precessional forcing is stronger in the observations (the last glacial cycle) than in the model simulations. The model-output westerly climatology forced by meltwater mimics that of the full forcings, while those of other single forcing experiments do not (Fig. 3 and Supplementary Fig. 7), which indicates a primary role of freshwater forcing in inducing circulation anomalies over Asia. Fresh glacial meltwater “hosing” into the North Atlantic Ocean at high latitudes causes a reduction of northward AMOC heat transport in the model simulations68. Based on simulated temperatures over the North Atlantic Ocean, the experiments capture rapid cooling well and are consistent with the cooling presented in other water hosing experiments78,79,80,81,82 for the YD and H1 events. Regarding freshwater hosing in H1, the differences in westerly climatology between two 300-year periods before (19-18.7 ka) and during the peak cooling (16-15.7 ka) in freshwater-forced experiments were calculated to reflect how such a North Atlantic cooling signal affects downstream westerly circulation over Asia. The mid-latitude westerly wave train response that propagates eastward across Eurasia in the middle to upper troposphere acts to intensify atmospheric high pressures and anticyclonic circulation over the Tibetan Plateau (TP; Fig. 3a). Strengthened anticyclonic circulation drives intense southerly winds to the west of the TP and intense northerly winds to its east, throughout the troposphere (Fig. 3b). The modern Asian dust transport pathway is consistent with the westerly path83; assuming that this also the case for the last glacial cycle, then these circulation changes should substantially modify dust transportation pathways in the Asian dust source region. Notably, downstream of the Asian interior, dust is more likely to be transported to the south by northerly winds, rather than eastward to the NPO. Similar changes to the westerlies are seen in the TraCE-21ka YD simulations (Supplementary Fig. 8).

Fig. 3: Westerly response to water-hosing in Heinrich stadial (H) 1 in the TraCE-21ka experiment.
figure 3

a 200 hPa wind vector (black arrows, in ms−1) and zonal deviations of geopotential height (filled colours, in gpm); and (b) vertical profile of meridional wind velocity (filled colours, in m s−1) for 35–45°N. The differences are based on model results in the TraCE-21ka freshwater-forcing-only experiment and are denoted as those between two 300-year periods before (19-18.7 ka) and during peak cooling (16-15.7 ka) over the North Atlantic in H1, respectively. Blue dots from left to right are the locations of Yumin Loess, Lake Qinghai, and core NP02, respectively.

Full size image

The TraCE-21ka results indicate no significant meridional westerly shift over the NPO during H stadials (Fig. 3a), which is consistent with the findings of ref. 84 that the main westerly axis remained at 38–40 °N over the NPO throughout the last glacial cycle. Instead, we infer that altered westerly pathways over inland Asia (Fig. 3a) favoured dust deposition patterns on the continent during H2, H4, H5, and H5a, which—given the low HIRM values in core NP02—left less dust available for transport and deposition over the NPO (Fig. 2d). Dust contents did not decrease in core NP02 during H3 (Fig. 2d). In contrast to H2, H4, H5, and H5a, when ice-rafted debris identified in the North Atlantic Ocean was derived mostly from the Laurentide ice sheet, H3 contains ice-rafted debris derived mostly from the European ice sheet; this different source/location of iceberg input is thought to have had less impact on AMOC intensity12,85,86,87. Accordingly, 231Pa/230Th ratios did not reach a production ratio of 0.093 during H3, which indicates less AMOC weakening88,89 (Fig. 2c). AMOC during H3 is, thus, interpreted to have been relatively deep and vigorous64,67,90, compared to the other Heinrich events. We infer that a weaker northward heat transport reduction in the North Atlantic during H3 had less impact on the westerly wave train than during the other H stadials and, thus, we observe no abrupt dust content decrease in core NP02 (Fig. 2d). We note that although westerly strengthening over the NPO is presented in the TraCE-21ka results, as it is in other models69,71, we have not taken the higher wind speed into consideration because past studies indicate a lack of correlation between wind speed and dust content in the open ocean20,58,59. In terms of the dust deposition process over the NPO, wet deposition is the dominant removal process23. Simulated precipitation and moisture anomalies from the TraCE-21ka experiments change little during the H1 event over the studied area (Fig. 4), which excludes it as an additional contributor to core NP02 dust variations.

Fig. 4: Differences in precipitation and moisture flux (black vectors) to water-hosing forcing during H1 in the TraCE-21ka experiment.
figure 4

Differences are based on model results for freshwater-forcing TraCE-21ka experiments and are denoted as those between two 300-year periods before (19-18.7 ka) and during peak cooling (16-15.7 ka) over the North Atlantic Ocean in H1, respectively. Locations of the Greenland GISP2 ice core and core NP02 are indicated by green dots.

Full size image

Our findings also reconcile apparently contrasting interpretations of a previously reported westerly change in China. For proximal dust deposition in China, larger aeolian dust grain size indicates stronger wind intensity due to westerlies shifting closer to dust deposition regions14. Based on grain-size analyses, loess records indicate northward-shifted westerlies during H stadials, whereas evidence from Lake Qinghai sediments suggests southward-shifted westerlies13,14 (Fig. 2e and f). We note that eastern Central Asian loess records (e.g., Yumin Loess) fall under the influence of intensified winds with a strong southerly component during H stadials (Fig. 3a), which would have transported coarser grains. Conversely, Lake Qinghai on the northeastern TP (Fig. 1) was influenced by intensified winds with a strong northerly component during H stadials (Fig. 3a) and, therefore, received more coarse-grained material. When viewed in isolation, these regional wind anomalies might suggest extensive northward or southward displacement of the westerlies belt, respectively, but our results identify them as merely regional AMOC-collapse-related meridional anomalies in the westerly wave train.

Finally, we consider distal dust deposition in Greenland, which reveals contrasting temporal variability to that in core NP02 (Fig. 2b and d). In Greenland ice cores, maximum (Asian) dust contents, characterized by higher Ca2+ concentrations, are found during H stadials, along with lesser peaks during Dansgaard-Oeschger (D-O) stadials91. The apparent inconsistency between signals in Greenland ice and core NP02 can be explained in terms of increased dust transport efficacy to Greenland during D-O stadials. Lower temperatures during cold stadials lead to reduced high northern latitude precipitation78,92,93, which reduced wet dust removal from the atmosphere and allowed dust to be transported more efficiently to Greenland. Reduced precipitation en-route has been reproduced for H1 in experiments with both full forcings and meltwater forcings only (Fig. 4), which facilitates dust transport to Greenland. Moreover, both the size distribution of dust in ice cores and circulation models suggest that dust transport times to Greenland were shorter during stadials94. Reductions in both transport time and en-route wet deposition increase dust transport efficacy, which in turn is conducive to enhanced dust accumulation in Greenland during D-O stadials. Apart from dust transport efficacy, recent studies suggest that European glacial dust and/or North African dust are potential sources for the last glacial Greenland ice core dust in addition to the Asian interior95,96, which may have further increased dust fluxes in Greenland. Our study is more Asian-dust specific and indicates that during H stadials, dust transportation pathways in the Asian dust source region were modified substantially, so that more dust was deposited in the Asian interior, which in turn decreased dust transportation and deposition over the NPO, and impacted the Earth system83,97, such as by weakening the NPO iron fertilization effect of phytoplankton productivity.

Conclusions

Dust plays a crucial role in the Earth system by modulating the global radiation budget and the global carbon cycle83,97. NPO dust contents are commonly linked to Asian interior source region climate conditions, with interpretations revolving around extensive poleward westerly shifts during warm states over timescales from a few decades to millions of years6,98. However, our reconstructions offer a more nuanced view of changes during abrupt millennial-scale climate events in response to AMOC slowdowns. We find that the westerly dust-transportation trajectory changed markedly over the Asian interior, setting up a pattern of substantial regional meridional anomalies, while no significant westerly change occurred over the NPO. Our deduced pattern of substantial regional meridional anomalies in westerly circulation over Asia reconciles apparently conflicting results from previous studies that focused on specific regions only. We also reconcile contrasting dust-content changes in Greenland ice cores relative to the NPO, in terms of dust transport efficacy changes between Asia and Greenland with shortened transport times and reduced en-route wet dust deposition during cold stadials.

Methods

Laboratory measurements

Magnetic remanences were measured for cubic samples (8 cm3) with a 2-G Enterprises Model 760 R cryogenic magnetometer (sensitivity 1 × 10−12 Am2). Isothermal remanent magnetizations (IRMs) were imparted in DC fields of 1 T, −0.1 T, and −0.3 T, which are defined as the saturation IRM (SIRM), IRM−0.1 T, and IRM−0.3 T, respectively. The HIRM, HIRM−0.1 T, and L-ratio are defined according to refs. 40,46

$${{{{{\rm{HIRM}}}}}}=({{{{{\rm{SIRM}}}}}}+IR{M}_{-0.3\, T})/2,$$
(1)
$$HIR{M}_{-0.1\, T}=({{{{{\rm{SIRM}}}}}}+IR{M}_{-0.1\, T})/2$$
(2)
$$L-{{{{{\rm{ratio}}}}}}={{{{{\rm{HIRM}}}}}}/[0.5\,\ast \,({{{{{\rm{SIRM}}}}}}+IR{M}_{-0.1\, T})],$$
(3)

Trace element and rare earth element analyses were made using a Thermo Scientific XSERIES 2 inductively coupled plasma-mass spectrometer (ICP-MS). Samples were freeze-dried and digested with HF-HNO3 (1:1) in closed Teflon bottles before ICP-MS analysis. Analytical accuracy was assessed with the GSD-9 standard, blank, and replicate measurements.

Clay minerals were identified by X-ray diffraction. Before measurement, samples were treated with 15% H2O2 and then 5% HCl to remove organic matter and carbonate, respectively. Clay-sized particles (<2 μm) were extracted and concentrated with a centrifuge, and were then smeared on glass slides. Analyses were performed with a Rigaku D/max 2500 diffractometer with CuKα radiation at 40 kV voltage and 100 mA current intensity. Peak parameters were estimated semi-quantitatively following ref. 99. Relative contents of smectite, illite, and kaolinite + chlorite were estimated from peak positions at 1.7 nm, 1 nm, and 0.7 nm, respectively100. Relative contents of kaolinite and chlorite were separated according to the relative 0.357 nm and 0.354 nm peak areas.

Model simulations

Model data from TraCE-21ka experiments include one full forcing simulation (TraCE-FULL, with transient forcing changes in orbitally driven insolation, atmospheric greenhouse gases, ice sheets, and meltwater fluxes) from 22 ka BP to 1990 CE and four single-forcing simulations (TraCE-ORB, TraCE-GHG, TraCE-ICE, TraCE-MWF), and were employed to evaluate the westerly response to different forcings during past climate changes. For the TraCE-21ka experiments, all simulations were performed using the Community Climate System Model version 3 (CCSM3) from the National Center for Atmospheric Research (NCAR) at T31_gx3v5 resolution101. The atmospheric model is the Community Atmosphere Model 3 with 26 hybrid coordinate levels. The land model includes soil columns, plant types, and land units that specify glaciers, lakes, wetlands, urban areas, and vegetated regions. The ocean model is an implementation of the Parallel Ocean Program in vertical z-coordinate with 25 levels. The sea ice model is the NCAR Community Sea Ice Model that includes a subgrid-scale ice thickness distribution. The TraCE-FULL experimental design can be found in Supplementary Fig. 9 and ref. 102. Each single-forcing experiment was performed with only one evolving forcing with remaining forcings fixed at the same initial conditions as the TraCE-FULL experiment.

TraCE-21ka simulations have been shown to reproduce the H1 and YD events well, which were characterized by a prominent northern hemisphere cooling, especially over the North Atlantic Ocean103,104. In this study, two periods of 300 years before and during peak cooling (19-18.7 ka and 16-15.7 ka for H1, 13.2-12.9 ka and 12.3-12 ka for the YD, respectively) were chosen to assess the role of freshwater hosing on westerly climatology differences (Supplementary Note 2).