Japan sea sediments consist of various detrital materials of eolian and riverine origin. Understanding the provenance of eolian dust is vital for reconstructing the variability of wind patterns and monsoons in the past. New and complete eolian accumulation rates from Taklimakan, Gobi, and Ordos are reconstructed at Integrated Ocean Drilling Program (IODP) Site U1425 in the Japan Sea using parallel factor (PARAFAC) endmember modeling. Our results show that Taklimakan dust is dominated by the silt fraction, while Gobi dust is dominated by the clay fraction, and they are controlled by the relative contributions of different pathways of dust transport, such as the westerly winds and East Asian winter monsoon (EAWM). Clay-size dust from Gobi increased during three periods, late Miocene global cooling (LMGC), intensification of Northern Hemisphere Glaciation (iNHG), and mid-Pleistocene Transition (MPT), which reflected increased EAWM winds associated with global cooling and glaciation. Taklimakan became the major dust contributor to the Japan Sea sediments during the warmer climate periods in the latest Miocene to early Pliocene and the Late Pleistocene, where westerly wind activity dominated eolian transport. Dust from Ordos increased greatly 0.95-0.85 Ma due to enhanced Asian aridification since 2 Ma in Northwest China. Detrital flux from Japanese islands suggests that the East Asian summer monsoon (EASM) was stronger 9.6-8 Ma and weakened from 8 Ma to the Pleistocene.

Eolian sedimentation in the North Pacific Ocean has been studied to evaluate the change in Earth’s climate and wind circulation in the Cenozoic (e.g., LL44-GPC-3, Ocean Drilling Program (ODP) Sites 885/886 and 576/578; [1, 2]). Westerlies and East Asian monsoons have prevailed in Asia since the early Miocene [36], in which East Asian summer monsoon (EASM) precipitation controls riverine detritus input to marine sediments and dust availability in desert areas, while the East Asian winter monsoon (EAWM) and westerly winds control dust transportation from Asian desert sources to marine sinks. Major climate changes in the late Cenozoic were accompanied by pronounced changes in the pattern and intensity of atmospheric circulation [7]. An increase in ice sheets during the Northern Hemisphere Glaciation intensified the EAWM [8] in northern China. The latitudinal position of westerlies migrated equatorward by ~3-5° during the Last Glacial Maximum [9], while it was weakened and more poleward during the warm Pliocene [10]. Currently, westerlies are migrating poleward in response to global warming. Accordingly, the interplay and response of atmospheric circulation to global climate changes are important factors for predicting future climate, and the reconstruction of past flux and grain size of dust accumulated in the downwind areas from inland deserts to the East Asian region could give us an opportunity to understand the mechanisms modulating activities of westerlies and EAWM winds.

Composition and flux of dust components could record the dust transportation mode and contribution from desert source materials, which could be well preserved in the sediments of the Japan Sea located at the eastern edge of East Asia on the main pathway of westerlies and under the influence of monsoon climate. Previous studies conducted at ODP Site 797 and IODP Site U1430 [1113] in the Japan Sea presented history of dust provenance change from the late Miocene to Quaternary. Although the relative contributions of dust to the Japan Sea sediments from different regions of Asian deserts have been intensely investigated, their associations with the wind circulations of westerly, EAWM, and the EASM are still poorly understood. It also remains unclear how the EASM and aridity have evolved over time in the inland deserts. The long-term aridification of the inland Asia might have associated with an uplift of the Himalayan-Tibetan complex in the Cenozoic [4, 14], while the aridity change and evolution of EASM and EAWM as well as the latitudinal position of westerly jet during the late Cenozoic are still controversial [1217].

Examination of temporal and spatial variations of eolian deposits from different dust sources regions could provide a critical information of the response of westerlies and monsoonal fluctuations to the global climate change in late Cenozoic and the potential driving factors for these consequences, which are potentially relevant to our future climate change. Numerous Asian dust provenance studies are primarily based on the comparison of mineralogy [1, 18, 19] which can effectively reveal the mineral composition of parent rock in the dust source region that controls the relative proportions of different minerals in the material carried by the wind. The mineralogical change in a specific eolian source region occurs with over tens of millions of years time scale [1, 20] that provide a stable and significant feature to reconstruct the past atmospheric activity. Although mineralogical composition is an effective proxy for eolian provenance study, precise quantification of contribution from different sources is still challenging. In this paper, parallel factor (PARAFAC) analysis was applied to mineral datasets in the silt and clay fractions of Japan Sea sediments to identify dust subcomponents from the complex mixture of materials, which enabled us to distinguish the variability of effects from atmospheric circulation of westerlies, EAWM, and aridity changes in the dust source region for the last 10 million years.

2.1. Integrated Ocean Drilling Program (IODP) Site U1425

A total of 180 samples were collected from IODP Site U1425 (39°29.44N, 134°26.55E) (Figure 1(a)), which is situated on the terrace of the Yamato Bank in the center of the Japan Sea. The higher topographic setting of the site minimizes the influence of turbidites, and slow sedimentation is ideal for detecting eolian dust [21].

The sediment at Site U1425 is defined by three lithologic units [21]. Unit I has an age from the Late Pliocene (2.7 Ma) to Holocene and consists of a higher proportion of terrigenous materials of clay and silty clay with small amounts of diatom-bearing clay. The most distinguishing sedimentary feature in this unit is the alternating bedding of dark and light intervals. Unit II has an age from the late Miocene (7.4 Ma) to late Pliocene (2.7 Ma) and is mostly composed of a mixture of fine-grained clay minerals and biosiliceous components of diatom-rich clay, which indicates high biological productivity in this unit. Unit III, covering 7.4-9.7 Ma in this study, consists of clay, diatom-rich clay, diatom ooze, and siliceous claystone and has relatively higher terrigenous content in the sediment than Unit II. Volcaniclastic materials in this core represent a minor component of the sediments in general [21].

The age model for this site was established based on an integration of datums suggested by three age models by Tada et al. [22], Kurokawa et al. [23], and Kamikuri et al. [24]. From 0-1.45 Ma, we follow the age model of U1424, where the gamma-ray attenuation (GRA) density profile was tuned with LR04 [25], and the age was projected to U1425 using the correlation of dark-light sedimentary cycles by Tada et al. [22]. From 1.5-9.21 Ma, an age model orbitally tuned with long (405 ky) and short (100 ky) eccentricity cycles by Kurokawa et al. [23] was used. From 9.22-9.69 Ma, biostratigraphic age controls by Kamikuri et al. [24] were adopted (Figure 2).

Samples at Hole D of Site U1425 from the top to ~370 m CCSF-D_Patched_rev20170309 (revised core splice depth below sea floor; [26]) cover 9.6 Ma, where the average sample resolution was approximately 53 ky. Stratigraphic positions of samples collected from the interval out of the splice were projected to the corresponding stratigraphic position in Hole B by examining the core photograph. Details about the samples used such as projected stratigraphic positions on the splice, revised core splice depth below the sea floor (m CCSF-D_Patched_rev20170309), estimated ages, and the related sedimentation rates are shown in Supplementary Table S1.

2.2. Potential Source Materials

The main sediment sources of the Japan Sea are the eolian dust from Central Asian deserts and riverine input from the nearby Japan islands. Asian dust are mainly emitted from the 10 deserts and sandy land in Central Asia such as the Mongolian Gobi Desert, Gurbantuggut Desert, Taklimakan Desert, Qaidam Basin, Badain Juran Desert, Tengger Desert, Hobq Desert, and Mu Us sandy land according to the meteorological observations [27, 28]. The detritus sources of these 10 deserts are nearby mountains and underlying bedrock in the three major regions of Northern Tibetan Plateau (NTP), the Central Asian Orogen (CAO), and the Ordos Plateau in the North China which have been formed by tectonic activities in central to northern Eurasia since Mesozoic [29]. NTP contains Tarim Craton and Qaidam Craton that provide clastic particles to the Taklimakan desert from the mountains of Kunlun Shan, Pamirs, and Altun [30]. CAO contains the Junggar Craton and Siberian Plate which supplied detritus to the Gobi and Mongolian deserts from the Gobi-Altai, Hangai mountains, and Tianshan Mountains (north) [29, 31]. The Ordos Plateau is situated in the North China Craton which delivered detrital materials to the eastern Mu Us Sandy land, Hobq Desert, and Ulan Buh Desert, although their contributions to the total Asian dust emissions are limited (Figure 1(a); [31]).

Despite NTP and CAO are regional metamorphic mountain ranges, combinations of the rock types and nature are different. Taklimakan desert is situated on the Tarim Craton in the NTP region which consists of marine shale and sandstone in the basin [32]. The Tarim sandstone mainly derived from the interior of the Tarim Craton which is dominated by monocrystalline quartz (Figure 1(b)) [33]. In contrast to the Tarim Basin, volcanic rocks and volcanogenic sedimentary rocks dominate in the Junggar Basin and these sandstones bear the compositional imprint of arc volcanics with felsic composition and plagioclase feldspar (Figure 1(b)) [33]. The East Gobi Basin was formed by extension and intracontinental rifting during the late Mesozoic, while the basement rock in southern Mongolia consists of volcanic islands with composition from lithic- and plagioclase feldspar-rich volcanic detritus to quartz and feldspar-rich granitic material [3436]. The distinctive features of the mineralogical composition in the two regions are inherited from the parent rocks which provide a stable and significant feature for provenance study (Figure 1(b)).

Quartz, K-feldspar, and muscovite are resistant to chemical weathering [37], while plagioclase will be partially weathered in the annual precipitation over 700 mm [38]. However, plagioclase (albite) is stable and unchanged in the semiarid and arid environment which rainfall is less than 500 mm. Because dust source regions in inland Asia such as Junggar Basin, Gobi, Mongolia, and Tarim Basin underwent prolonged semiarid to arid environment since early Miocene to present [4, 14], primary detrital minerals such as quartz, K-feldspar, and plagioclase (albite) could be altered insignificantly under such dry environment. Long-term resistivity of minerals under arid condition was well exemplified in L1 and L2 loess layers in the Chinese Loess Plateau (CLP) [38].

To determine provenance for the detrital endmembers in U1425 sediments, we collected loess and desert materials from the 3 primary dust source areas of the Tarim Basin near the NTP, Tian Shan Mountains, Mongolian and Gobi Deserts in the CAO, Hobq, Inner Mongolia, and Mu Us Deserts on the Ordos Plateau [3941] and neritic sediments close to the Japanese coast to serve as a riverine supply from the Japanese islands (Figure 1(a)). The list of source samples from the Taklimakan Desert, Mongolian Gobi Desert, Ordos Plateau, and Japanese islands is shown in Supplementary Table S2.

3.1. Grain-Size Separation

To distinguish possible transport modes of detrital subcomponents such as westerlies, low-level wind during the winter monsoon, and river suspension and to eliminate mineralogical fractionation during sorting, it was necessary to measure compositional variability for various grain-size fractions within sediments. Nagashima et al. [42] pointed out that eolian dust dominates the silt fraction of hemipelagic sediments in the Japan Sea, while riverine input is in the clay fraction. Therefore, we decided to separate sediment samples into different size fractions of silt (>4 μm) and clay (<4 μm). Grain-size separation was conducted by the repeated pipette method [43].

3.2. Powder X-Ray Diffraction (XRD) Analysis

Mineral composition was analyzed for both silt (>4 μm) and clay (<4 μm) fractions, as well as potential source materials by using a powder XRD method. A powder sample mounted on a glass holder was scanned from 2° to 40° 2θ by 1° 2θ/min at 40 kV and 20 mA using CuKα radiation, where the diffraction intensity (counts per second: cps) was recorded with a 0.02° 2θ step. Identification of major and clay minerals was performed according to the position of diagnostic peaks of these minerals on the XRD diffractograms (Figure 3; quartz 20.8°, illite 8.8°, chlorite 25.2°, kaolinite 24.2°, smectite maximum between 5 and 8°, K-feldspar 27.5°, anorthite 27.8°, and albite 28.0°). Since the 26.6° peak of quartz may overlap strong reflection of illite, the 20.8° peak was used as the relative measure of quartz abundance [19]. The reproducibility of these measurements is better than 5%. The full width at half maximum (FWHM) of illite was also examined as a half height width (∆ °2θ) of the 10 Å (8.8°) peak on the XRD diffractograms (Figure 3). This is a measure of the crystallite size of illite and is often used to trace the possible source regions in marine sediments [44]. Higher values indicate poor crystallinities, and low values indicate well-developed crystallinities [45]. According to the Kübler index [46], the illite crystallinity value is useful to determine what type of metamorphic conditions existed when a rock was formed.

3.3. Mass Accumulation Rate (MAR) Calculations

The MAR (g/cm2/ky) for each detrital subcomponent was calculated using the dry bulk density (DBD) and linear sedimentation rate (LSR). DBD is defined as the mass (weight) of the dry solids divided by the total volume of the wet sample [47]. DBD (g/cm3) was estimated from wet bulk density (WBD) assuming constant grain density (GD) and pore water density by Equation (1). As a measure of WBD for each sample, shipboard GRA density was used, which represents the WBD of the entire core and was measured for every 2.5 cm interval for all sediment core sequences [21]. To maintain consistency between wet bulk density directly determined by shipboard “moisture and density (MAD)” measurements [21], we first calibrated GRA-based WBD (WBDGRA) to MAD-based WBD (WBDMAD) using a correlation between them, such as WBDMAD=0.12901+0.89581WBDGRAS.E.=±0.10. A GD of 2.59±0.15 (g/cm3) averaged for all MAD measurements for the U1425 samples was assumed for all samples. The pore water density was assumed to be 1.024 g/cm3, as used in the MAD calculation [21].

LSR (cm/ky) were determined from the revised core splice depth below the sea floor (m CCSF-D_Patched_rev20170309) and the integrated age model. MAR (g/cm2/ky) of each endmember (EM) was calculated by the method of Rea and Janecek [48] as MAR=fractionofEMLSRDBD.

4.1. Theoretical Framework

The XRD diffractogram of each sample could be regarded as a linear combination of the diffractograms of subcomponents, such as dust from various sources, riverine input from surrounding lands, biogenic materials, and their diagenetic products. Each diffractogram of the subcomponent and its fraction in the sample should be nonnegative, and the total fraction of subcomponents should be (close to) unity. PARAFAC analysis was applied to decompose XRD diffractograms into those of subcomponents (endmembers) and to quantify their contributions using drEEM toolboxes running on MATLAB® software [49]. The PARAFAC model was applied to data that were arranged in three-way arrays such as the excitation-emission matrix (EEM) of fluorescent organic matter (fluorescenceintensityofsample×excitationwavelength×emissionwavelength), where xijk is the intensity of fluorescence corresponding to the j-th emission wavelength and the k-th excitation wavelength for the i-th sample, and eijk is the residual representing the variability not accounted for by the model. Assuming the number of endmembers is F, the decomposed matrices aif, bjf, and ckf are the factor score (relative contribution of subcomponent in the i-th sample), factor loading spectrum along emission wavelength at a k-th excitation wavelength, and magnification factor at k-th excitation wavelength for f-th endmember, respectively, where aif, bjf, and ckf are nonnegative [49].

The calculation is conducted to obtain an endmember-mixing model that minimizes eijk. In the case of XRD in this study, the dataset can be regarded as a special case with k=1, where xijk is the intensity of diffraction at the j-th diffraction angle (2-40° 2θ with 0.02° 2θ interval; 1j1900) for the i-th sample (1i180).

4.2. Determination of the Number of Endmembers

An appropriate mixing model was found to minimize the sum of squares of the residuals [50]. In this study, model development was initiated with a series of (F =) 2 to 7 endmembers (EMs), which were applied to the 180 XRD profiles of silt and clay fractions with a nonnegativity constraint. Increasing the number of endmembers from 2 to 7 reduced the score error and total residual sum of squares (RSS), and a significant decrease in error occurred between the 5- and 6-endmember models. As shown in Figure 4(a), for the silt datasets, the 4- and 5-endmember models showed higher RSS and larger score errors from 15-25%, while the 6- and 7-endmember models showed smaller score errors within 10%. On the other hand, some samples had a larger RSS in the 7-endmember model than in the 6-endmember model (Figure 4(a)). In the clay datasets, the 4-, 5-, and 7-endmember models showed higher RSS and larger score errors than the 6-endmember model, which had the smallest RSS and score errors within 10% (Figure 4(b)). Therefore, 6-endmember models were adopted for both silt and clay datasets in this study.

4.3. Scaling Factor Calculations and Reproducibility of Endmember Modeling

After validation of the 6-endmember PARAFAC model, scores (aif) and spectral loadings (bjf) of each endmember were obtained from the model, where ckf was unity in this case (k=1). Because the total sum of fractions of each endmember should be unity, scaling factors, αf, were calculated to satisfy f=16αfaif1 for all i, where the square sum of residual errors (1f=16αfaif) was minimized. Using αf, the fraction (absolute contribution) of the f-th endmember for the i-th sample was calculated as αfaif. To obtain the XRD intensity profile of each endmember, outputs of drEEM toolboxes [49] for spectrum loadings (bjf) were first divided by the maximum of bjf in terms of j for each f-th endmember, which were set as bjf. Then, bjf was divided by the scaling factor αf to obtain a “realistic” diffractogram of each endmember.

The 6-endmember PARAFAC was modeled 20 times (Figures 5(a) and 5(b)) to ensure that the result was stabilized and not a local minimum [51]. Similar solutions with minor differences were reached, which implied that the obtained mixing model was reproducible and reliable and could be used for a direct comparison between the mineral compositions of endmembers and potential source materials. Therefore, the result was reported as the average of 20 models. The nonnegativity-constrained 6-endmember model explained 99.57% of the variation in silt datasets and 99.43% of the variation in clay datasets.

5.1. Mineral Composition in Silt and Clay Fractions in Japan Sea Sediments

Quartz, plagioclase, and illite were the most abundant minerals in the silt fraction during the Pleistocene, suggesting many dust-derived materials in this period (Supplementary Table S3). FWHM of illite in the silt fraction was low (high crystallinity) with a narrow range of approximately 0.2 ∆°2θ in the Pleistocene. According to the Kubler Index [46], highly crystallized illite (FWHM<0.25) was sourced from regional metamorphic rocks in mountain ranges, such as those on the NTP or in the CAO. Poor illite crystallinity (FWHM>0.4) corresponded to diagenetic rock sources, such as those in the Japan Island arc.

Based on the mineralogical differences in the Tarim Basin (higher quartz content) and Junggar Basin in the Gobi Desert (higher plagioclase content) [33] mentioned in above Section 2.2 (Figure 1(b)), the distribution of albite (Ab) and quartz (Q) ratios and illite crystallinity (FWHM) could effectively differentiate dust and riverine sources in silt and clay sediments in the Japan Sea. Taklimakan eolian sediment had the lowest FWHM values and the lowest Ab/Q ratios, and Gobi dust had a low FWHM but a higher Ab/Q ratio. Japanese riverine sediment had the highest FWHM values and higher Ab/Q ratios (Figure 6(a)). These interpretations are also applicable to the clay fraction (Figure 6(b)). This preliminary identification of the 3 sources is not sufficiently quantitative because of a lack of knowledge about exact mineral ratios for each detrital subcomponent and difficulty in eliminating contributions from biological components. Therefore, the results of more quantitative statistical endmember unmixing by PARAFAC are examined in the next section.

5.2. Provenance Assessment of PARAFAC Endmembers

To determine the provenance for the 6 EMs in U1425 sediments calculated by PARAFAC, firstly, we compared the diffractogram patterns of EMs with those of potential source materials. Figure 5 shows that desert source materials from the Taklimakan Desert, Gobi deserts, and Ordos Plateau had high intensities of quartz, feldspar, and illite which match well with EM1, EM2, and EM5. Then, characteristic mineral ratios such as Ab/Q and K-felspar (K-fel)/Ab were used to distinguish the source materials (Figures 7(a) and 7(b)). Namely, Gobi and Ordos dust sources are characterized by higher Ab/Q, while Taklimakan shows lower Ab/Q (Figure 7(a)). On the other hand, Ordos material is characterized by higher K-fel/Ab (Figure 7(b)).

Taklimakan materials and EM1 had lower Ab/Q ratios than Gobi materials and EM2 (Figure 7(a)). Sandstone in the Tarim Basin is dominated by monocrystalline quartz, which was a typical continental block-recycled orogen provenance of Paleozoic metamorphic rocks in the NTP [33, 52, 53], while Mongolian Gobi is characterized by back-arc basin deposits with plagioclase, feldspar-rich, volcanic detritus [33, 36]. Ordos materials and EM5 had high K-fel/Ab ratios (Figure 7(b)). The Ordos Plateau is situated in the northern part of the North China Craton, which contains largely undisturbed sedimentary rocks that are Carboniferous to Jurassic in age and Cretaceous sandstones [54]. Therefore, we could attribute EM1, EM2, and EM5 to Taklimakan, Gobi, and Ordos, respectively. Again, it was also confirmed that EM1, EM2, and EM5 exhibited similar XRD diffractogram patterns to those of the desert materials from Taklimakan, Gobi, and Ordos, respectively (Figure 5).

Japanese island materials and EM3 had XRD diffractogram patterns different from those of the Asian dust sources and revealed significant peaks of smectite, low illite intensities (Figures 7(c) and 7(d)) and high anorthite/albite ratios. The Japanese islands are characterized by basaltic volcanic and sedimentary rocks [55]. Volcanic rocks often produce smectite by chemical weathering under wet and humid environments on Japanese islands [56]. An attribution of EM3 to Japanese Island materials could be also justified by the similar temporal variations of illite FWHM and EM3 contribution (Figures 6(c) and 6(d)). Larger illite FWHM (low crystallinity) is generally associated with higher contribution of EM3 and both profiles show similar trend and critical changes at 8, 6, and 2.7 Ma both for silt and clay fractions. EM4 and EM6 were attributed to biogenic opal and opal-CT subcomponents, respectively (Figure 5).

5.3. Contributions and Fluxes of Endmembers

A six-endmember PARAFAC model was established, and 3 Asian dust sources, Taklimakan (EM1), Gobi (EM2), and Ordos (EM5); a riverine source from the Japanese islands (EM3); and 2 biogenic sources, diatomaceous (EM4) and Opal-CT (EM6), were identified, and their contributions were quantified (Figure 8 and Supplementary Table S4). Sediments at IODP Site U1425 in the Japan Sea indicate that biogenic blooms (EM4+EM6) occurred from the late Miocene to Pliocene. The contribution of Asian dust (EM1+EM2+EM5) to the Japan Sea was low in the late Miocene to Pliocene, at 20-40% (Figure 9(b)), and the MAR ranged from 0.2 to 1.5 g/cm2/ky but increased dramatically to 70% in the Pleistocene, with a maximum MAR of 3.1 g/cm2/ky 0.03 Ma (Figure 10(a)). The total Asian dust MAR in this study is similar to previous estimations of 1-3 g/cm2/ky at ODP Site 797 [11] and 2-3 g/cm2/ky at MD01-2407 [57] in the Japan Sea and is also consistent with the exponential decreasing trend along the eastward transportation of Asian dust [58] during the late Quaternary.

5.3.1. Taklimakan Desert (TAK: EM1)

The contribution of Taklimakan dust to the Japan Sea ranged from 10 to 50% (Figures 8(a) and 8(b)), and the flux increased gradually in the Pleistocene to a maximum of 1.67 g/cm2/ky at 0.29 Ma, in which both silt and clay fractions had the highest MAR at the same time (Figures 10(b) and 10(d)). TAK became the main dust contributor to the Japan Sea from 0.5 to 0.16 Ma and 5.6-4 Ma (Figure 9(a)). Lower flux occurred in the late Miocene to Pliocene at <0.5 g/cm2/ky. The temporal distribution of the Taklimakan dust flux in the silt and clay fractions had a similar trend and maintained a regular ratio between them from the late Miocene to the present, although the silt fraction dominated the total Taklimakan dust fractions (Figure 10(d)), which implies that both silt and clay dust were transported at the same time by the same transportation mode of wind.

5.3.2. Gobi Desert (Gobi: EM2)

The contribution of Gobi dust to the Japan Sea ranged from 20 to 60% since 10 Ma (Figures 8(a) and 8(b)). The highest MAR occurred 0.96 Ma at 1.7 g/cm2/ky and it was lowest in the Pliocene (Figure 10(b)). Clay-sized Gobi dust had a higher flux than the silt-sized fraction from 9.24 to 0.73 Ma. The highest MAR in the clay fraction was 1.56 g/cm2/ky in 0.96 Ma. The MAR of the silt fraction was generally low at <0.8 g/cm2/ky. The temporal variation in Gobi fluxes in silt and clay fractions showed a great difference in pattern and trend, as well as dust quantity (Figure 10(c)), which implied that both were transported by different wind circulation from a dust source area to the Japan Sea.

5.3.3. Ordos Plateau (Ordos: EM5)

Ordos dust to the Japan Sea was significant in the Pleistocene since 2 Ma with a 10-40% contribution (Figures 8(a) and 8(b)). The highest MAR occurred 0.92 Ma with 1.26 g/cm2/ky and decreased to 0.83 g/cm2/ky in the Holocene (Figure 10(b)). The silt fraction was dominant with Ordos dust since 1 Ma (Figure 10(e)).

5.3.4. Japan Island Arc (Japan: EM3)

The contribution of riverine input from Japanese islands to the Japan Sea was high during the late Miocene at 30-50% and decreased significantly in the Pleistocene (Figure 9(b)). The highest MAR was 2.98 g/cm2/ky, occurring 9.66 Ma but decreased to 0.2-1.2 g/cm2/ky from 8 Ma to the present, showing minima 8-7, 6-4.5, 2.7-2, and 1.5-0.6 Ma (Figure 10(a)).

6.1. Grain-Size Variation of Dust from Taklimakan and Gobi

Figure 10(d) shows that Taklimakan dust in general was dominated by silt fraction (Silt_TAK) from the late Miocene to Pleistocene, which may reveal a typical characteristic of a long-range transport of dust from Taklimakan by westerly wind. Atmospheric circulation in the Tarim Basin is controlled by high-level westerly winds, which are responsible for long-range dust transport (>5000 km) to the Japan Sea and Pacific Ocean [27]. Desert materials in Taklimakan were entrained to an elevation of >5 km in the troposphere and were transported by rapid and strong westerly winds to downwind areas (Figure 11). During transportation, no dust fall was observed in the proximal areas of Chinese loess until it traveled downwind to the Japan Sea [59]. Hence, all entrained materials in the deserts were transported directly downwind which is supported by the result of a climate model study in the Asian dust transportation by westerly [60]. Typical 4 μm far-traveled Asian dust and larger particles (>10 μm) were transported to Canada in April 2001, indicating a strong and rapid westerly jet in the troposphere [61].

In the case of Gobi materials, clay-size fraction (Clay_Gobi) had a higher flux than the silt-size fraction (Silt_Gobi) from 9.24 to 0.73 Ma (Figure 10(c)). Because temporal flux variations of silt (Silt_Gobi) and clay (Clay_Gobi) fractions showed very different pattern and trend as well as their magnitude, silt and clay fractions were considered to be transported by different wind circulation modes from Gobi to the Japan Sea. Most desert materials in the Mongolian Gobi were entrained to an elevation of <3 km and carried by EAWM winds to a medium distance (500 to 3000 km) to the Chinese Loess, southeastern China, and the Japan Sea [62]. When Gobi dust materials ascend to the highest elevation, the materials begin to descend and cause heavy dust to fall in the proximal region of CLP [59, 62]. According to Liu et al. [63], CLP has been subdivided into three zones of sandy loess, loess, and clayey loess zones from northwest to the south depending upon particle size which were sorted by wind blowing from the desert sources. Jeong et al. [38] also found coarse silt was progressively replaced by the fine silt and clay from west to eastward on CLP. Coarse particle (5-20 μm) dust could be deposited on CLP [60], 2-5 μm dust has settled on the Korean Peninsula [64], and only fine-grained dust (clay) could be suspended further to the Japan Sea and Southeast Asia (Figure 11). Although the effects of weathering should be considered, the gravitational settling during dust transportation by low-level winds of the EAWM is obvious. Therefore, it is reasonable to consider that clay-sized Gobi dust dominated from 9.24 to 0.73 Ma in the Japan Sea was mainly transported by low-level atmospheric circulation by EAWM winds (Figure 10(c)). Figures 12 and 13 show that cooling events and trends found in the benthic δ18O record [65] correspond well with the periods of higher flux of Gobi clay, which suggests that EAWM might be enhanced during cooling phase and glacial periods since 10 million years ago.

On the other hand, silt-sized Gobi dust in the Japan Sea revealed a similar MAR pattern and trend as those of Taklimakan dust, implying that coarser Gobi dust (silt) was mainly transported by westerlies to the Japan Sea. Sun [62] and Tsai et al. [59] also observed dust storm events in the Mongolian Gobi Deserts in which dust was entrained to the troposphere and transported by westerly winds to the Japan Sea and Pacific Ocean or America.

6.2. Detrital Provenance Variability in Japan Sea Sediments

6.2.1. Late Miocene

High riverine flux from Japanese islands 9.6-8 Ma indicates a strong summer monsoon climate in East Asia, which is consistent with the clay mineral study at Linxia Basin in the NE Tibet [17]. A provenance shift occurred from 8 to 7 Ma; the Japanese riverine flux decreased to the lowest value, 0.2 g/cm2/ky, and the eolian flux increased from 7.8 to 6.6 Ma in the Japan Sea (Figure 10(a)). Increase of eolian sedimentation at 8-7 Ma was also confirmed in the Loess Plateau, NE Tibetan Plateau, the Japan Sea, and the North Pacific [12, 6668].

These data imply that Asian aridification approximately 8 Ma coincided with late Miocene global cooling (LMGC) between ~7 and~5.5 Ma and with a short period of Northern Hemisphere glaciation between 6 and 5.5 Ma (Figure 12; [69, 70]). Eolian flux in general was low but occasionally increased to 1.39 and 1.02 g/cm2/ky in 6.72 and 5.65 Ma, respectively, suggesting the aridity of the Asian interior and the intensification of the winter monsoon in East Asia (Figure 10(a); [71]). This intensification of the EAWM was also confirmed by the increase of grain size in the eolian deposits on the CLP in 7.4 Ma and 5.3 Ma [15].

6.2.2. Pliocene

Warming Event appearing in the early Pliocene [70, 72] after the late Miocene cooling was characterized by very low fluxes of eolian and riverine material, 0.2-1 g/cm2/ky (Figure 10(a)), but had the highest biogenic contributions to the Japan Sea (Figure 9(b)). Shifts in dust provenance sources occurred from Gobi to Taklimakan during the Warming Event in 5.6-4 Ma (Figure 9(a)). According to a study by Ao et al. [72], for eolian records in CLP, this Warming Event only increased summer monsoon moisture in East Asia, but enhanced aridification by increasing evaporation than precipitation in most parts of Central Asia. The riverine input from Japan islands did not show a prominent increase during the warming period (Figure 10(a)), which may suggest continuously invariable summer precipitation at eastern margin of Asia at that time. A gradual increase of eolian material 3.2 Ma with decreasing of biogenic material followed sudden cooling 3.3 Ma (Figures 9(b) and 13).

6.2.3. Pleistocene

A substantial increase in eolian MARs to 2.5 g/cm2/ky occurred after 2.7 Ma (Figure 10(a)) in the Japan Sea during the intensified Northern Hemisphere Glaciation (iNHG), which established a cold and dry environment, accelerating sediment erosion in the Central Asian Mountain ranges and strengthening the atmospheric circulation systems in the Northern Hemisphere. Low temperatures during iNHG caused a greater equator-to-pole temperature gradient and stronger thermal gradients along polar frontal systems [73], resulting in stronger westerlies [74] and intensification of winter monsoon winds.

Gobi was the greatest dust contributor to the Japan Sea in the Early Pleistocene but decreased sharply after 0.95 Ma (Figure 13). Ordos dust increased only 0.95-0.85 Ma during the mid-Pleistocene transition (MPT) due to enhanced aridification in Northwest China which is evidenced by Li et al. [75] for the Ulan Buh Desert and southern Inner Mongolia. The formation of deserts in the Mu Us and southern Inner Mongolia could be as old as the MPT (1.1-0.9 Ma) [75] that resulted in increased dust materials from the Ordos Plateau (Figure 13).

Taklimakan became the main dust contributor to the Japan Sea after 0.5 to 0.16 Ma during the Mid-Brunhes Event (MBE) (Figures 10(b) and 13), which corresponded to a period when Earth eccentricity was close to 0 [76, 77], and the benthic δ18O record shows the largest variation between 3.2 and 5.4‰ in the Late Pleistocene [65] (Figure 13).

6.3. Implications for Atmospheric Circulation Variability during the Past 10 Million Years

Based on our dust records, we reconstructed the variability of paleoatmospheric circulation of westerly winds by using fluxes from Taklimakan, as well as silt-sized dust from Gobi (Gobi silt). The EAWM was inferred from clay-sized dust from Gobi (Gobi clay). The EASM and aridity changes in East Asia were inferred from Japanese riverine sediment during the last 10 million years (Figure 14).

6.3.1. Dominance of the EASM 9.6-8 Ma

The dominance of the EASM in East Asia 9.6-8 Ma suggested warm air and moisture from the southeast, which was inferred from the high Japanese riverine flux record (Figure 14). The eolian deposits accumulated on the CLP had a low deposition rate and strong pedogenesis, which revealed intensified precipitation during this period [78]. In the same period, TAK and Gobi silt dust dominated 9.6-9.4 and 8.9-8 Ma, indicating that westerly winds were strong. In addition, high flux of Gobi clay 9.3-8.9 Ma suggested that the EAWM winds were enhanced during a less pronounced, late Miocene cooling step ~9.0 Ma (Figure 12; [79]). The high-frequency fluctuations in dust grain size and dust source materials during this period suggested that unstable atmospheric circulation prevailed in East Asia during the dominance of the EASM in the late Miocene (Figure 14).

The EASM was strong 9.6-8 Ma but relatively weak in 8-7, 6-4.5, and 3-2 Ma in the Japan Sea region that might contradict with some CLP records in China which proposed EASM increase in the Miocene–Pliocene climate transition [72]. The evolution of EASM in East Asia during this period and its controlling mechanisms remains controversial [80] and complicated because EASM precipitation also depended on the variation of regional and local environments. The controls on aridification in inland Asia at ∼8 Ma might involve the retreat of EASM, global cooling, and the uplift of the Tibetan Plateau. It seems widely accepted that the uplift of the Tibetan Plateau played a significant role on the long-term aridification of inland Central Asia during Cenozoic by modulating the atmospheric circulation and blocking moisture from southeast Asia [6, 81]. Based on climate modeling study, Zhang et al. [14] demonstrated the growth of the Tibetan Plateau and uplift of the northern Tibetan Plateau expand drylands in central Asia to north of ∼40°N which resulted in same dryland zone in higher latitude as present. In the latest Miocene or Pliocene, the northern and eastern part of Tibetan Plateau and adjacent mountain ranges were still uplifted [82] that may have intensified aridity in inland and Central Asia at 8 Ma and afterwards.

6.3.2. Dominance of the EAWM during Global Cooling and the Early Pleistocene

The dominance of the EAWM was linked to global cooling, such as LMGC (Figure 12), as well as the increase in ice sheets in the Northern Hemisphere, such as iNHG and MPT (Figure 13). This observation suggested that global cooling could have strengthened the low-level EAWM winds [83] and increased dust emissions from the Mongolian Gobi and northern China deserts to the Japan Sea.

During the late Miocene cooling period, Gobi clay significantly contributed to the total eolian flux in the Japan Sea (Figure 12) from 7.8 to 6 Ma, suggesting Asian aridification and intensification of the EAWM. Although this fast late Miocene cooling was considered to be triggered by abrupt decrease of CO2 or uplift of Tibetan Plateau, the true mechanism is still controversial and under hot debate [12, 15, 17, 84]. Slight increases of Gobi clay 4, 3.6, and 3.2 Ma (Figure 13) suggest progressive global cooling, which was associated with the occurrence of ice-rafted debris (IRD) at the north pole in the late Pliocene until the beginning of the Northern Hemisphere Glaciation [85, 86].

During the iNHG, increasing Siberian High activity strengthened the EAWM winds [8] in northern China, which increased dust storm events as well as dust transportation from the Gobi and northern China deserts. Gobi clay increased substantially from the late Pliocene to Early Pleistocene until 0.6 Ma (Figure 13). On the other hand, some studies suggested that the strong Siberian High and EAWM forced the westerly winds to migrate southward [9, 87, 88] during the strong glacial and cooling periods. Another study on the loess in the Tarim basin suggests that expansion of Northern Hemisphere ice sheets could intensify the strength of the westerly jet and push it southward, which enhanced aridity and intensified dust storm activity in the Tibetan Plateau [89]. The increase of accumulation of Taklimakan dust and Gobi silt in the Japan Sea indicated a prolonged interval of westerly wind blowing on inland desert area since early Pleistocene (Figure 14).

6.3.3. Dominance of Westerly Winds during Warmer Periods

During the warmer periods in the Warming Event in the late Miocene to early Pliocene and the Late Pleistocene, weakened cold air masses from the Siberian High could permit a northward migration of westerly winds in an earlier season [10, 16, 90]. Dust provenance changes from Gobi to TAK 5.8-4.2 Ma (Figures 9(a) and 14) during the Warming Event revealed the dominance of westerly winds but was characterized by low dust deposition in the Japan Sea.

The influence of westerlies increased gradually reflecting the increasing fluxes of TAK and Gobi silt from 2.5 to 1.1 Ma and became dominant in the Late Pleistocene from 0.5 Ma to the present. The early northward migration of westerlies increased the duration of westerlies over the Taklimakan Desert and resulted in an increase in the length of season for dust storms and dust transportation from the Taklimakan Desert, as well as deserts in northern China, to the Japan Sea. A similar situation can currently be observed with a longer dust season from April to August [91]. During the dominance of westerlies, Gobi silt also increased accordingly and had a higher flux than Gobi clay from 0.5 Ma to the present (Figure 13).

Under recent global warming conditions, the dominance of westerly winds is obvious, which calls our attention to the influence on the regional and local climate. Extreme weather events such as heat waves and large storms are likely to become more frequent or more intense in the coming future. Although the behaviors of regional climate and weather are complicated to understand and forecast, long-term global climate changes in the past, such as the warming in the Pliocene or the dominance of the summer monsoon in the late Miocene, may provide the basis to solve the riddle and the problems that we are facing now or will face in the future.

This work provides empirical evidence for the major changes in paleoatmospheric circulation of westerlies, the EAWM and aridity change in the dust source region since 10 Ma using a sediment sequence from IODP Site U1425 in the Japan Sea. Mineralogical analyses of the sediment at IODP Site U1425 effectively established a proxy for eolian provenance study. Mineral compositions could distinguish the various detrital sources, such as dust from Taklimakan and Gobi and riverine input from the Japanese islands. In particular, PARAFAC endmember modeling is a powerful analytical tool to quantitatively differentiate eolian dust sources, such as those in Taklimakan, Gobi, and the Ordos Plateau, as well as the other subcomponents in the sediment.

The EASM was strong 9.6-8 Ma, while it showed fluctuation at moderate intensity after 8 Ma to the Pleistocene in the Japan Sea region. Stepwise Asian aridification occurred 8, 6, and 2 Ma and finally induced desertification in the Ordos Plateau area. A great increase in Ordos dust contribution to the Japan Sea in the Late Pleistocene was related to the increase in aridity in Inner Mongolia and the Ordos Plateau since 2 Ma and the enhanced aridification 1.5-0.6 Ma induced desertification in southern Inner Mongolia during the MPT.

The dominance of the EAWM was linked to the LMGC, iNHG, and MPT associated with global cooling and the increase in ice sheets in the Northern Hemisphere, which strengthened the low-level EAWM winds and increased the deposition of fine clay dust from Mongolian Gobi. In contrast, during the dominance of westerlies in the warmer periods during the Warming Event in the late Miocene to early Pliocene, as well as in the late Pleistocene, early seasonal northward migration of westerlies prolonged the duration of westerlies over the inland desert area, which resulted in deposition of coarser silt-sized eolian dust from the Taklimakan and Gobi Deserts.

The authors confirm that the data supporting the findings of this study are available within the article.

The authors declare that there is no conflict of interest regarding the publication of this article.

We would like to express our sincere thanks to all the participants of IODP Exp. 346 for their efforts to provide high-quality continuous sediment records from the Japan Sea. We also thank Profs. Atsuko Sugimoto and Masanobu Yamamoto for their valuable advice during the course of this study. Profs. Ryuji Tada, Hongbo Zheng, Youbin Sun, and Hitoshi Hasegawa kindly permitted to use their loess and desert samples. This work is a part of the doctoral thesis of A. Lee under the guidance of T. Irino. This work was supported by a grant from IODP Exp. 346 After Cruise Research Program, Japan Agency for Marine-Earth Science and Technology (2014-2017; awarded to Ryuji Tada, T. Irino, and others), JSPS KAKENHI grant number 16H01765 (2016-2019; awarded to Ryuji Tada), and JSPS grant number JPMXS05R2900001 (2017-2020; awarded to Masanobu Yamamoto).

Exclusive Licensee GeoScienceWorld. Distributed under a Creative Commons Attribution License (CC BY 4.0).