Loess-Like Dust Appearance at 40 Ma in Central China

across the Asian continent with ages varying from the Miocene to the Eocene and linked to various mechanisms including global cooling, Tibetan Plateau uplift and retreat of the inland proto-Paratethys Sea. Here, we study the Eocene terrestrial mudrocks of the Xining Basin in central China and use nonparametric end-member analysis of grain-size distributions to identify a loess-like dust component appearing in the record at 40 Ma. This is coeval with the onset of high-latitude orbital cycles and a shift to predominant steppe-desert vegetation as recognized by previous studies in the same record. Furthermore, we derive wind directions from eolian dune deposits which suggest northwesterly winds, similar to the modern-day winter monsoon which is driven by a high pressure system developing over Siberia. We propose that the observed shifts at 40 Ma reflect the onset of the Siberian High interacting with westerly derived moisture at obliquity timescales and promoting dust storms and aridification in central China. The timing suggests that the onset may have been triggered by increased continentality due to the retreating proto-Paratethys Sea.


Introduction
. The classic sequence of Quaternary loess initiated at ∼2.6 Ma and is linked to the coeval glaciation of the Northern Hemisphere (Z. Ding et al., 1997;Lu et al., 2010). These loess deposits are interbedded with finer-grained reddish paleosols (median grain-sizes of 10-30 µm) formed during interglacials, which are interpreted to have formed under a wetter East Asian summer monsoon (Maher, 2016). Conformably underlying the Quaternary loess is the Red Clay, which, despite the name, consists mostly of silt-sized material similar to the paleosols in the Quaternary loess. These are interpreted to reflect the atmospheric circulation during the warmer and wetter Miocene-Pliocene epochs (Maher, 2016). In the central Chinese Loess Plateau, east of the Liupan Shan, ages of the Red Clay range between 7 and 11 Ma (Ao et al., 2016;Z. L. Ding et al., 1998;B. Wang, Zheng, et al., 2014;Y. Song et al., 2001;D. Sun et al., 1997;Xu et al., 2009;Zheng et al., 1992;Y. Zhu et al., 2008). The onset of these dust deposits is interpreted as the set-up of winter monsoonal circulation and inland aridification both linked to the uplift of the Tibetan Plateau . However, older loess-like deposits extending up to ∼20-29 Ma were identified west of the Liupan Shan (Garzione et al., 2005;Guo et al., 2002;X. Qiang et al., 2011;Y. Zhang et al., 2014) and to the north of the Loess Plateau (Jiang & Ding, 2010). These records suggest that the monsoonal circulation is much older and proposed to be driven by the combined effect of an earlier phase of Tibetan Plateau uplift and the retreat of the inland Paratethys Sea that covered the Eurasian continent during the Paleogene (Guo et al., 2002). In recent years, even older dust deposits have been identified in Eocene strata (Li et al., 2018;Licht et al., 2014), suggesting that Asian atmospheric circulation may have been driven by the India-Asia collision, greenhouse to icehouse cooling or proto-Paratethys Sea incursions, all occurring during the Eocene.
Some of the oldest loess-like deposits in central China have been identified in the Eocene mudrocks of the Xining Basin ( Figure 1) and are shown to extend from 34 Ma to at least 40 Ma (Licht et al., 2014). These deposits have been used as an indicator for aridity and winter monsoonal circulation (Licht et al., 2014). However, it remains unclear when dust deposition started in the basin which has important implications for the origin of Asian dust and associated atmospheric circulation. Moreover, recent sedimentological and provenance studies show that the pre-Quaternary dust records in central China contain significant contributions from local fluvial sources which obscures the dust signal (Alonso-Zarza et al., 2009;Liu et al., 2019;Nie et al., 2014;Y. Zhang et al., 2014;H. Zhang et al., 2018).
Here, we reassess and extend the sedimentary record of Licht et al. (2014) to ∼50.5 Ma using multiple previously dated sections in the Xining Basin as well as a new sedimentary section (Baimasi) containing widespread eolian sedimentary structures of deflated gypsum grains to derive paleo-wind directions ( Figure 1b). Using these sections, we construct a composite record of grain-size distributions (GSDs) to identify windblown dust in the mudrocks. The measured GSDs of these mudrocks represent a mix of the different fluvial and eolian components operating in the basin. These are subsequently unmixed using a nonparametric end-member analysis to quantify the relative contributions of eolian dust (Paterson & Heslop, 2015;van Hateren et al., 2018;. End-member modeling of GSDs has been successfully applied to distinguish eolian and fluvial contributions in marine cores (e.g., Weltje and Prins, 2003) as well as lacustrine records (e.g., Dietze et al., 2014) and in various other sedimentary environments (e.g., see appendix I in van Hateren et al., 2018). In addition, the end-member models have been extensively tested and shown to perform well using both artificial and natural data sets (Dietze & Dietze, 2019;Paterson & Heslop, 2015;van Hateren et al., 2018;Varga, Gresina, et al., 2019). Using this approach, we identify a loess-like dust component appearing in the record at ∼40 Ma, coeval with the onset of obliquity-forced atmospheric moisture cycles and an increase in steppe-desert vegetation previously recognized in the Xining Basin by Bosboom et al. (2014). We subsequently extend both the cyclostratigraphic and palynological records to the early Eocene with additional samples which corroborate the previous findings of Bosboom et al. (2014), but propose an alternative interpretation involving the onset of the Siberian High at 40 Ma.

Geologic Setting and Previous Work
The modern-day Xining Basin is bounded by the eastern Qilian Shan in the north, the Laji Shan in the south and the Riyue Shan in the west. The basin constitutes the westernmost part of the Cenozoic Longzhong Basin, which stretches toward the Liupan Shan in the east (Horton et al., 2004). Various tectonic settings have MEIJER ET AL.

Lithostratigraphy
Cenozoic deposition commenced during the early Eocene and unconformably overlies Cretaceous strata or older basement rocks (L. G. Fan et al., 2019;Fang et al., 2019). Paleogene deposits include the Qijiachuan, Honggou, Mahalagou and Xiejia Formations (BGMRQP, 1965;Dai et al., 2006;Horton et al., 2004;QBG-MR, 1985 and Figure 1c). The Qijiachuan Formation consists of organic-rich evaporites and fluvio-lacustrine mudrocks (Fang et al., 2019;Meijer et al., 2019). These are followed by brick-red, massive mudrocks of the Honggou Formation alternated with lacustrine mudrocks containing carbonate beds and evaporites . Fluvial sandstones are found locally in the lower part, but decrease upwards . The Honggou Formation successions are interpreted mainly as a dry mudflat in a distal alluvial environment based on the occurrence of evaporites and the massive structure of the mudrocks resulting from subaerial reworking such as bioturbation and pedogenesis . The overlying Mahalagou Formation is characterized by orbitally forced alternations between massive red mudrocks and gypsum beds (Abels et al., 2011;Dupont-Nivet et al., 2007;G. Q. Xiao et al., 2010 and Figure 2a). The gypsum beds in the Xining Basin disappear after the Eocene-Oligocene Transition (EOT) due to regional aridification (Dupont-Nivet et al., 2007). The uppermost gypsum bed at the EOT is used as a stratigraphic marker for the boundary to the Oligocene Xiejia Formation composed predominantly of brownish-red mudrocks (Fang et al., 2019).
Neogene formations include the brownish-yellow alluvial mudrocks and fluvial sandstones of the Chetougou and Guanjiashan Formations and the conglomerates of the Mojiazhuang Formation (Fang et al., 2019;R. Yang et al., 2017;. The Neogene deposits are upwardly coarsening due to increased uplift of the eastern Qilian Shan, located north of the Xining Basin (Fang et al., 2019;R. Yang et al., 2017), and of the Laji Shan located to the south (G. Xiao et al., 2012). Subsequent incision of the Huangshui River resulted in fluvial terraces which are capped by up to 200 m of Pliocene Red Clay and Quaternary loess (Fang et al., 2019;.

Previous Work
Previous studies in the Xining Basin have interpreted the massive mudrocks of the Eocene Mahalagou Formation as loess-like dust (Licht et al., 2014). This interpretation is based on the surface textures of the silt-sized quartz grains indicating eolian transport, the GSDs of the mudrocks being similar to Quaternary loess and the lack of sedimentary structures indicative of fluvial deposition (Licht et al., 2014). In addition, the U-Pb age spectra of detrital zircons indicate a provenance similar to the Quaternary Chinese loess (Licht et al., 2016). It has also been noted that the meter-scale cycles in the Mahalagou Formation between gypsum beds and mudrocks may be the Eocene equivalent of the orbitally driven loess-paleosol alternations observed in the Quaternary records of the Chinese Loess Plateau (Licht et al., 2014).
These Eocene cycles are observed in the center of the Xining Basin (at the Dasigou, Shuiwan, Tiefo, East Xining, Tashan and Xiejia sections; Figure 1b) and are identified as predominantly driven by the 41-kyr obliquity cycle and to a minor extent by the 100-kyr eccentricity cycle (Abels et al., 2011;Dupont-Nivet et al., 2007;Huang & Hinnov, 2019;G. Q. Xiao et al., 2010). The gypsum beds are interpreted as wet periods with an elevated groundwater level, whereas the red mudrocks reflect a dry mudflat (Abels et al., 2011;Dupont-Nivet et al., 2007). Bosboom et al. (2014) identified the onset of these cycles at 40 Ma after a brief transitional phase between ∼40.5 and 40 Ma which is characterized by laterally discontinuous gypsum beds, which become continuous afterward. In addition, Bosboom et al. (2014) proposed that the onset of these cycles correlates with an increased prevalence of steppe-desert vegetation observed via the pollen record, but this was poorly constrained between 40.7 and 39.5 Ma due to a lack of productive palynological samples in this interval.
Both the onset of obliquity cycles and the increased dominance of steppe-desert vegetation have been linked to aridification due to global cooling shortly after the Middle Eocene Climatic Optimum (MECO; Bosboom et al., 2014). However, a recent study using clumped isotopes identified no atmospheric cooling in the Xining Basin at 40 Ma (Page et al., 2019) thereby challenging this interpretation. The stratigraphic record in the Xining Basin has subsequently been extended down to ∼50.5 Ma and dated using magnetostratigraphic correlations constrained by the U-Pb age of a tuff at 50.0 ± 0.4 Ma ). An alternative age model has been proposed afterward by F. Yang et al. (2019), but this study did not include the MEIJER ET AL.  radiometric age of the tuff layer and missed several reversed polarity zones in the gypsiferous upper part of the stratigraphy, likely due to a strong normal overprint observed in these beds. It remains unclear to what extent the deposits of the early to middle Eocene are orbitally forced and whether they are formed by the accumulation of eolian dust as observed in the Mahalagou Formation.

Sections
We sampled previously described and dated sections (Table 1; Figure 1b) for grain-size analysis and new fossil pollen assemblages to create a composite record covering the early to late Eocene. These sections include the 125 m-thick Tiefo section (Bosboom et al., 2014), and a freshly excavated quarry in the upper 170 m of the Xiejia section (Dupont-Nivet et al., 2007;Meijer et al., 2019), both covering the upper Honggou to lower Mahalagou Formations. We also sampled the previously studied, 275 m-thick Caijia section , covering the Qijiachuan to Honggou Formations. This section was sampled for X-Ray Fluorescence (XRF) scanning and color measurements to construct time series for cyclostratigraphic analysis in addition to grain-size analysis and fossil pollen. Detailed methods and sampling resolution of the various analyses follow below. The Baimasi section (J. Zhang, Wang, et al., 2016), was logged using a Jacob's staff and sampled for magnetostratigraphy and grain-size analysis. Additionally, we collected 11 grain-size samples from a 12-m profile of the Oligocene deposits north of the Shuiwan section (Abels et al., 2011;Dupont-Nivet et al., 2007), and 10 grain-size samples from a 1-m loess profile near the Xiejia section for comparison with the Eocene GSDs. This profile consists of loosely packed silt-sized sediments with a massive sedimentary structure and is found at high elevation (2,370 m) indicating an eolian origin ( Figure S1). These deposits are therefore interpreted as Quaternary loess following the criteria of Pye (1995).

Magnetostratigraphy
Paleomagnetic samples were taken at the Baimasi section using a portable electric drill and a compass mounted on an orientation stage at ∼1 m resolution where possible resulting in 68 samples. These samples were analyzed in the laboratory of paleo-and rock magnetism at the German Research Centre for Geosciences (GFZ) in Potsdam. After cutting, the samples were thermally demagnetized with increments of 120°C for the first step, followed by increments of 50°C up to 620°C and a final step at 650°C resulting in 12 temperature steps in total. A viscous normal overprint was removed during demagnetization up to 200°C ( Figure S2a). Subsequent heating up to 650°C resulted in a linear demagnetization toward the origin (Figure S2a). The thermal demagnetization pattern is similar to previous paleomagnetic studies in the Xining Basin and indicates the dominant magnetic carriers being magnetite, due to a drop in remanent magnetization at ∼580°C, and hematite, due to retaining remanent magnetization beyond 650°C (Dai et al., 2006).

Grain-Size Analysis
Mudrocks of the studied sections were sampled at ∼1 m resolution where possible resulting in 442 samples in addition to the 10 sediment samples from the Quaternary loess profile mentioned above. Grain-size MEIJER ET AL. Xiejia loess profile 36°31'12"N, 101°52'17"E This study Table 1 Locations of Studied Sections analysis was performed on these samples in the sediment laboratory (Sedlab) of Section 4.6 Geomorphology at the GFZ in Potsdam. Beds of evaporites and carbonates were not sampled. The samples were pretreated in several steps to remove organic matter, carbonate and gypsum and to disperse the grains. To enhance the efficiency of the chemical treatments, the rock samples were first disaggregated using a mortar and pestle, but carefully to make sure no individual grains were broken. Second, ∼1 g of sample was used for further treatment and 2.5 ml of 30% hydrogen peroxide (H 2 O 2 ) was added to remove the organic matter. The samples were left to react on a shaking table for at least six days until the reaction stopped. Third, 2.5 ml of 10% hydrochloric acid (HCl) was added and the samples were left on a shaking table for two days to dissolve the carbonate. Fourth, 50 ml of deionized water was added and the samples were left in an overhead shaker for three days to dissolve the gypsum. The deionized water was refreshed every day. Fifth, 0.1 g of sodium pyrophosphate (Na 4 P 2 O 7 *10H 2 O) in a 2.5 ml solution was added and the samples were left in an overhead shaker for at least 12 h to disaggregate the grains.
The samples were measured in suspension using a Horiba Partica LA-950 v2 laser diffraction particle sizer with a reported measurement range of 0.01-2,500 µm and therefore suitable for analyzing mudrocks. The diffraction data were converted to GSDs using the Mie optical model, which is recommended for particle sizes below 30 µm (Varga, Gresina, et al., 2019). However, the Mie model requires the refractive index and the absorption coefficient of the measured particles as input. These are difficult to constrain because the samples are composed of an unknown mix of minerals. Therefore, we tested a wide range of refractive indices (1.4-2.45) and absorption coefficients (0-1) and compared the GSDs ( Figure S3). Artificial peaks appear in the submicron-range when the absorption coefficient approaches zero, as is reported in previous studies (Sperazza et al., 2004;Varga, Gresina, et al., 2019). Otherwise, the GSDs are consistent and show only minor variations. Hereafter, we used a refractive index of 1.55 and an absorption coefficient of 1. GSDs are given in 92 size bins logarithmically distributed between 0.011 and 2,500 µm. Some of the measured residues were observed and photographed using an optical microscope to check the reliability of the measurements ( Figure S4).
To unmix the GSDs, we used the nonparametric end-member analysis implemented in the AnalySize software package (Paterson & Heslop, 2015). Although this analysis is nonparametric and therefore requires no predefined function of the end-members, the number of end-members needs to be prescribed. Tests with synthetic mixtures have shown that the use of an incorrect number of end-members leads to unreliable results (van Hateren et al., 2018). We ran the analysis using 1 to 10 end-members and found that four end-members are necessary to account for most (98.6%) of the variance in the data set (R 2 = 0.986). Increasing the number of end-members does not significantly improve the quality of fit ( Figure S5a) and using fewer end-members does not sufficiently cover the entire range of possible grain-sizes, especially below 2 µm ( Figure S5b). Furthermore, we found that the end-member interpreted as loess-like and its increase at 40 Ma are robust features of the grain-size record regardless of the predefined number of end-members. The median grain-size, end-member abundances and sample-wise R 2 of the individual sections are shown in Figure S6. A composite grain-size record was created by linear interpolation of the samples between correlated polarity zone boundaries in the various sections.

Cyclostratigraphy
We constructed time series of the Caijia section by measuring XRF scanning and color at 0.5 m resolution resulting in 515 samples in total. The samples were carefully ground, assembled in 32 mm cups, covered by a 4 µm polypropylene film and measured using an ITRAX XRF core scanner. XRF intensities (in counts per second) were obtained for the 13 elements Al, Si, S, Cl, K, Ca, Ti, V, Mn, Fe, Rb, Sr, and Zr. All statistical analyses were performed on element intensity records after log-ratio or centered-log-ratio transformation. We used principal component (PC) analysis as implemented in the ItraXelerate software (Weltje et al., 2015) to examine the covariance between the individual elements in our data set. The results are presented in a biplot of PC1 and 2, which reveals three groups of elements ( Figure S7). Lithogenic elements (Fe, Ti, K, Si, Al, and Rb) are positively correlated and clearly separated from the authigenic elements (Ca, S, and Sr) by PC1, which explains 72.7% of the variance. The authigenic elements can be subsequently separated into carbonates (Ca) and gypsum (S and Sr) along PC2 explaining another 12.7% of the variance. We construct time series by using the dominant elements of these three groups (Ti, Ca, and S). Hence, relative varia-MEIJER ET AL.
10.1029/2020PA003993 7 of 24 tions of lithogenics, carbonates and gypsum are represented by log-ratios of Ti/Ca, S/Ti and S/Ca, which provide the most easily interpretable signals of relative changes in the chemical composition (Weltje & Tjallingii, 2008).
RGB-values were extracted from high-resolution photographs produced by the corescanner to construct a color record. The average values of 10 representative pixels were used and converted to CIELAB color space for interpretation purposes. These values include: L* (lightness), a* (red over green), and b* (yellow over blue). However, note that these values are not calibrated and should only be used to study relative changes throughout our record. Spectral analysis was performed using Redfit (Schulz & Mudelsee, 2002) as implemented in the Past software (Hammer et al., 2001) with no segmentation, no oversampling and a Welch window. The time series were bandpass filtered using the AnalySeries software (Paillard et al., 1996).

Palynology
Productive palynological (pollen and spore) samples were collected from gypsum beds and laminated mudrocks in the Tiefo, Xiejia and Caijia sections. Samples collected from red beds in these sections were barren of palynomorphs. The samples were prepared by Palynological Laboratory Services (Anglesey, UK) using standard palynological techniques (HCL, HF, sieving at 10 µm) and examined at the University of Amsterdam, The Netherlands. Ten samples were productive; nine from the Mahalagou Formation, but only one from the Honggou Formation. These samples were added to the previously studied palynological record of the Xining Basin (Bosboom et al., 2014;Dupont-Nivet, Hoorn, & Konert, 2008;Hoorn et al., 2012;Page et al., 2019 and Figure S8). Samples were published as supplementary data for a recent review of Asian steppe ecology (Barbolini et al., 2020) but are first described here for climatic reconstructions of our studied sections. Palynomorphs were identified using the Genera File of Fossil Pollen and Spores (Jansonius & Hills, 1976 and recent updates), Z. C. Song et al. (1999) and the Chinese palynological literature. Plant Functional Types (PFTs) were assigned using the Nearest Living Relatives (NLR) approach applied by Hoorn et al. (2012).

Baimasi Section
The late Eocene deposits in the Xining Basin are characterized by mudrocks alternated with gypsum beds (Figure 2a) formed in saline lakes as recognized by previous studies (Abels et al., 2011;Bosboom et al., 2014;Dupont-Nivet et al., 2007). The newly studied Baimasi section is located on the eastern margin of these saline lake deposits and is described in the following. The section is subsequently dated and correlated to the other sections by using magnetostratigraphy and marker beds.

Observations
The lithostratigraphy of the Baimasi section consists of red gypsiferous mudrocks and gypsarenites (Figures 2b and 2c). The mudrocks have a horizontally laminated or massive structure. The gypsarenites are horizontally bedded and contain dm-scale sets of planar cross-bedding and laterally extensive horizontal bounding surfaces (Figures 2b and 2c). The foresets have a tangential base and a maximum dip of ∼20°.

Interpretations
The depositional environment of the gypsiferous mudrocks is interpreted as a dry mudflat and as a saline lake for the gypsum beds (Abels et al., 2011;Dupont-Nivet et al., 2007;Smoot & Lowenstein, 1991;Talbot et al., 1994). The siltstone beds with climbing ripples suggest rapid deposition by unconfined fluvial flows (North & Davidson, 2012). Gypsarenites are rare and almost always associated with eolian transport of gypsum grains deflated from nearby salt pans (Smoot & Lowenstein, 1991). This is corroborated by the rounding and sorting of the gypsum grains and the cross-bedded gypsarenites are thus interpreted as eolian dunes following J. Zhang, Wang, et al. (2016). The horizontally bedded gypsarenites are interpreted as eolian sand sheets. However, we note that the cross-beds observed here are thin with gentler foresets than is common for eolian sands (e.g., McKee, 1966). We argue that is may be due to the low accumulation rates observed in the Xining Basin (∼3 cm/kyrs; Abels et al., 2011;Meijer et al., 2019) resulting in only the more gently steeping base of the dunes being preserved.
The measured paleocurrent directions of the gypsarenites are corrected (red arrows in Figure 3) for the clockwise 20° vertical axis rotation of the Xining Basin during the late Paleogene as reported in previous paleomagnetic studies (Dupont-Nivet, Dai, et al., 2008). Although only four beds of cross-bedding are preserved, the paleocurrent directions of these beds are consistently to the south and southeast. This suggests that north to northwesterly winds occurred during the Eocene, which would be similar to the winter monsoons dominating the region northeast of the present-day Xining Basin (Figure 1). However, we note that the Eocene wind directions may be affected by local topography which is poorly constrained.

ChRM Directions
The Characteristic Remanent Magnetization (ChRM) directions were calculated using eigenvector principal component analysis (Kirschvink, 1980) on at least four temperature steps. The resulting directions reveal two clusters of either normal or reversed polarity ( Figure S2b). The means of MEIJER ET AL.
10.1029/2020PA003993 9 of 24 these clusters were calculated using Fisher statistics (Fisher, 1953) and samples with an angle of more than 45° from the mean were rejected from further analysis (open symbols in Figure 3). The reversals test was performed using the means of the two clusters (McFadden & McElhinny, 1990), but failed due to an unresolved normal overprint as observed in previous studies in the Xining Basin (Dupont-Nivet, Dai, et al., 2008;Meijer et al., 2019). This would make our data set unsuitable for rotational analysis, but does not affect the reliability of the magnetostratigraphy. The Virtual Geomagnetic Poles were calculated and polarity zones are defined on at least two successive samples with the same polarity.

Correlation to the GTS
The Baimasi section contains a reversed polarity zone at the base (R1), followed by a ∼40 m-thick normal polarity zone (N1) and a ∼20 m-thick reversed polarity zone at the top (R2). The Baimasi section is mapped as the Mahalagou Formation (Dai et al., 2006: QBGMR, 1985 and Figure 1a) and is observed to grade upwards into the brownish-red mudrocks of the Xiejia Formation (Figure 2e). Furthermore, the characteristic meter-scale gypsum beds of the Mahalagou Formation are observed both at the top and base of the section (Figures 2e and 3). These observations confidently constrain the age of the Baimasi section to the Mahalagou Formation, which serves as a starting point for our magnetostratigraphic correlations. The chrons observed in the Mahalagou Formation include: C13 to C18 (Figure 4; Abels et al., 2011). The only long reversed chron in this interval is C13r and we therefore correlate it to R2. C13r is preceded by a short normal and reversed chron (C15n and C15r), which are not observed in the Baimasi section, possibly because this interval is poorly exposed (Figures 3 and  4). Below C15r is a long normal chron, C16n, which we correlate to N1 and R1 is correlated to C16r. This suggests the Baimasi section covers the 36.5-35 Ma interval but our interpretation is limited by the missing reversals of C15n and C15r. Nevertheless, we argue that based on the lithological observations listed above, we can confidently correlate the Baimasi section to the Mahalagou Formation spanning from 40 to 34 Ma.

Grain-Size Distributions
The GSDs of the studied sections are dominated by fine silt with a median grain-size of 4-6 µm ( Figure 5a). The GSDs measured in this study differ from the late Eocene grain-size record from the Xining Basin reported in Licht et al. (2014), which are characterized by bimodal distributions with modes at 16-31 µm and at 1.6-2.8 µm. The samples in both studies were prepared using similar pretreatments, but measured using a different laser diffraction device. Comparison studies have shown that although laser diffraction measurements are precise and highly reproducible (Goossens, 2008), their accuracy is limited and using different devices may result in different GSDs (Varga, Gresina, et al., 2019). Therefore, care should be taken when comparing GSDs between studies and we thus exclude the results of Licht et al. (2014) in this study. The comparison with the GSD of Quaternary loess (Figure 5b) is made with samples measured using the same pretreatment, laser diffraction device and optical settings.

End-Member Analysis
End-member analysis of the data set reveals four end-members (Figure 5a) based on the best correlation (R 2 ) per grain-size class. Three of these end-members have modes at 2.6 µm (EM1), 5.1 µm (EM2), and 10.1 µm (EM3). A fourth end-member is fine skewed with a mode at 51.5 µm (EM4). The composite record ( Figure 6) shows that the lower part of the stratigraphy (50.5-40 Ma) is dominated by EM1, EM2, and EM3, while the abundance of EM4 is low (5% on average). The upper part of the record (40-36 Ma) is dominated by EM3 and EM4, while the abundance of EM1 and especially of EM2 are reduced. At 40 Ma, the abundance of EM4 increases abruptly to an average of 19% in different sections (Tiefo and Xiejia) at the same stratigraphic level suggesting a laterally continuous shift ( Figure S6), coeval with the previously recognized onset of obliquity cycles and increase in steppe-desert pollen (Bosboom et al., 2014). Samples from the Oligocene mudrocks show similar end-member abundances as in the upper part of the Eocene record ( Figure S6).
MEIJER ET AL.

End-Member Interpretations
The GSD of EM4 is remarkably similar to the Quaternary loess in the Xining Basin (Figure 5b), which corresponds to the coarse local dust component observed in global loess populations (1.b.1 in Vandenberghe, 2013, Figure 5c) and in Tibetan lake deposits (Dietze et al., 2014, Figure 5d). However, it should be noted that the GSDs of loess may vary considerably depending on the source and distance to the source (e.g., Pye, 1995;Újvári et al., 2016;Vandenberghe, 2013). Furthermore, similar GSDs are observed on mudflats formed by fluvial transport (van Toorenenburg et al., 2018) and therefore the GSD alone is not a diagnostic feature for eolian transport. Additional evidence is provided by the U-Pb age spectra of detrital zircon grains indicating a provenance for the silt-sized fraction (10-45 µm) different from Eocene fluvial deposits but similar to Quaternary loess (Licht et al., 2016). Furthermore, previous surface morphology analyses of the >20 µm quartz grains identified features such as dish-shaped depressions and adhering clay particles, which suggest a prolonged eolian transport for these grains (Licht et al., 2014). However, it should be noted that it remains unclear whether these features are formed by the eolian transport of silt-sized grains or are inherited from older phases of eolian transport (e.g., Krinsley & McCoy, 1977). Lastly, the abrupt, laterally continuous increase of EM4 suggests an eolian rather than a fluvial origin linked to local dynamics. Cross-bedded silt-and sandstone beds indicating fluvial transport are observed as well, but decrease upwards . Such channels only reappear during the Miocene related to the uplift of the nearby mountain ranges (G. Xiao et al., 2012;R. Yang et al., 2017), while there is no evidence of a tectonic event at 40 Ma. This suggests a reduction in fluvial activity rather than an increase at 40 Ma as observed in EM4. We note that the observations listed above, when considered on their own, are not strictly diagnostic of windblown transport , but argue that when taken together, these observations suggest an eolian origin as the most likely. We therefore interpret the increase of EM4 as the onset of a loess-like dust component, which remains a consistent part of the sedimentary record afterward.
The finer-grained end-members may occur in several sedimentological settings and their origin is more difficult to assess because these grains are too small for single-grain provenance and quartz surface morphology analyses. In terms of eolian dust, these end-members may occur as longterm suspension transport, possibly by upper-level westerlies (Pye, 1995;J. Sun, 2002), or as aggregates in short-term suspension as observed in modern-day dust storms (M. Qiang et al., 2010). We also note that EM3 might reflect the fine-grained dust component commonly observed in the Mio-Pliocene records of the Red Clay Formation (Shang et al., 2016). Alternatively, a fluvio-lacustrine origin may be inferred either in suspended load or as mud aggregates transported in bedload (Wright & Marriott, 2007). Because of these difficulties, we refrain from deriving any sedimentological and paleoclimatic information using these end-members and focus on EM4 instead. Note, that because we only interpret EM4 with an average abundance of 19% as being eolian in origin, the Xining deposits are not predominantly composed of windblown silt in contrast to previous interpretations (Licht et al., 2014(Licht et al., , 2016. Therefore, these deposits are not classified as lithified loess (Pye, 1995), but rather interpreted as a diverse depositional environment containing a loess-like component as well as other fluvio-lacustrine and/or dust components (Figure 7). This interpretation is in line with recent depositional models for the Miocene Red Clay   (Dietze et al., 2014). GSD, grain-size distribution.

Cyclostratigraphy
The increase of loess-like dust at 40 Ma is coeval with the onset of obliquity cycles in the Xining Basin previously recognized by Bosboom et al. (2014) in the middle-late Eocene records. In the following, we perform time series analysis on the Caijia section, which covers the Qijiachuan and Honggou Formations, to examine whether the record in the Xining Basin is orbitally forced in the early to middle Eocene. MEIJER ET AL.

10.1029/2020PA003993
13 of 24 Figure 6. Composite of the early late Eocene record of the Xining Basin based on magnetostratigraphic correlations between the studied sections. The composite shows the lithostratigraphy, cyclostratigraphy and end-member abundances of the grain-size record and Plant Functional Type abundances of the palynological record with new samples from this study and Barbolini et al. (2020) indicated in red. The magnetostratigraphy is correlated to the GTS16 (Ogg et al., 2016). Note that the increase in the loess-like end-member (EM4) at 40 Ma is coeval with the onset of 41 kyr obliquity cycles and the increase in steppedesert pollen previously recognized by Bosboom et al. (2014).

Cyclicity
The Honggou Formation is dominated by massive brick-red mudrocks, similar to the overlying Mahalagou Formation and interpreted as a dry mudflat (Figure 8; Meijer et al., 2019). The red mudrocks are alternated with meter-scale intervals of brown to green mudrocks with fine horizontal laminations interbedded with cm-scale beds of carbonate and gypsum. These laminated mudrocks are interpreted as lacustrine and reflect wetter atmospheric conditions (Figure 8; Meijer et al., 2019). The alternations between the red and lacustrine mudrocks are regular, but with a long period of 10-12 m compared to the meter-scale cycles in the overlying Mahalagou Formation. Thirteen of these low frequency cycles can be identified in the Caijia section ( Figure 8). No lacustrine deposits are identified in cycle 4 (Figure 8), but a meter-scale channel body of very fine sandstone is observed here instead. The channel body contains trough cross-bedding and grades laterally into cm-scale beds of very fine sandstone with ripple laminations interpreted as fluvial deposits . Two of the cycles contain gypsum beds instead of lacustrine mudrocks (cycle 5 and 8; Figure 8) indicating more saline conditions . Gypsum beds are more common in the upper part of the Honggou Formation and the cyclicity is less obvious. Two groups of meter-scale gypsum beds with a lacustrine interval in between may represent three more cycles (cycle 11-13; Figure 8). Afterward, no more cycles are observed in the upper part of the section. No cyclicity is observed as well in the Qijiachuan Formation at the base.

Spectral Analysis
Spectral analysis (Figure 8) of the color records reveals a >95% confidence peak at a period of 12 m in a* (red over green) and b* (yellow over blue). L* (lightness) shows a >99% confidence peak at 19.6 m. The bandpass filter of L* shows that this period follows the occurrence of gypsum and carbonate beds, but fails to identify some of the lacustrine mudrocks which have a similar lightness as the red mudrocks. Therefore, the 19.6 m period may represent a double cycle of the 12-m cyclicity. The XRF proxies Ca/Ti and S/Ti are dominated by higher frequency cycles with periods of 1-3 m, which reflect the individual gypsum and carbonate beds that occur in the lacustrine intervals, but lack the lower frequency cycles observed in the color records. The lower frequency cycles are represented in the Ca/S record which shows a >95% confidence peak at 10.6 m ( Figure 8). No cyclicity is observed in the grain-size record, which is discussed below.
MEIJER ET AL.

Correlations
The cycles in the 50-45 Ma interval of the Caijia section can be correlated bed-to-bed to the East Xining section located ∼9 km to the west ( Figure S9). Similar cycles are observed as well in the Xiejia section, located ∼14 km to the south. However, these cycles consist of sandstone beds in the lower part instead of lacustrine mudrocks ( Figure S9). The Xiejia section is located in a more proximal setting toward the southern margin of the basin which may explain the occurrence of fluvial sandstones during the wet phase of the cycles. Similar orbital cycles are observed in the mudflat deposits of the Teruel Basin in Spain with distal lacustrine deposits correlating to proximal fluvial beds during the wetter phases (Ventra et al., 2018). The correlations in the Xining Basin suggest that the cycles are basin-wide and possibly orbitally forced. The accumulation rates are relatively constant at 3 cm/kyrs for the Honggou Formation , which suggests that the cycles of 10-12 m have a duration of 333-400 kyrs and may be forced by the long eccentricity cycle of 405 kyrs. To identify possible phase relations and driving mechanisms of the Caijia sedimentary record, we correlated the record with various astronomical solutions (Laskar et al., 2011, Figure 9). The sedimentary cycles are dated using the magnetostratigraphy of  and various age models for the chron boundaries (Gradstein et al., 2012;Westerhold et al., 2015;2017 and Figure 9). The most recently tuned age model (Westerhold et al., 2017)   In the following, we explore the occurrence of the ∼400-kyr cycles in terms of orbital forcing. Climate model simulations testing the role of orbital forcing under modern-day boundary conditions suggest that both the summer monsoons and wintertime westerlies are driven directly by orbital forcing through Northern Hemisphere insolation, although the phase is opposite (Bosmans et al., 2018;Shi et al., 2011;X. Zhang, Jin, et al., 2016). At the midlatitudes, this insolation is dominated by the precession cycle and, to a minor MEIJER ET AL.

10.1029/2020PA003993
16 of 24 Figure 9. Color record a* (red over green) of the Caijia section correlated to the astronomical solutions (a-d) of eccentricity, e (Laskar et al., 2011). The ages of the Caijia section are calculated by linear interpolation between chron boundaries identified by . The base of the Caijia section is correlated to the top of C23n. Chron ages of the following timescales are used: GTS12 (Gradstein et al., 2012), W15 (Westerhold et al., 2015) and W17 (Westerhold et al., 2017). W17 shows wet phases of cycle 1 to 13 during 405-kyr eccentricity minima. extent, by the obliquity cycle. The 405-kyr eccentricity cycle is only a modulation of the shorter precession and 100-kyr eccentricity cycles and therefore, the lack of these higher frequency cycles in our record is puzzling. Other studies have identified similar low frequency cyclicities in the absence of higher frequency cycles and ascribed this to limitations of the paleoenvironmental record in preserving the higher frequency cycles  or to a clipped insolation response resulting in only the most extreme amplitudes being recorded (Nie et al., 2008). Alternatively, the paleoenvironmental response to orbital forcing may be modulated by other mechanisms. Marine records show that the lower frequency cycles such as the 405-kyr eccentricity cycle are preferentially amplified in the carbon cycle during the Cenozoic (Kocken et al., 2019) and therefore may have affected the Xining Basin via CO 2 forcing. Additionally, teleconnections with the Antarctic ice sheets are proposed for the occurrence of 100-kyr eccentricity cycles in various Miocene records across the northeastern Tibetan Plateau (Z. Wang et al., 2019) and the incipient Eocene ice sheets may have similarly modulated the paleoenvironment in the Xining Basin. Regardless of the explanation for the 400-kyr cycles, our cyclostratigraphic analysis reveals no higher frequency cycles in the record before 40 Ma, even though the sampling resolution of ∼0.5 m corresponds to ∼17 kyrs (based on the average accumulation rate of 3 cm/kyr from Meijer et al., 2019) and should be enough to detect obliquity and 100-kyr eccentricity cyclicities. In addition, no cycles at all are identified in the record between 45 and 40 Ma. This is in stark contrast to the obliquity cycles observed after 40 Ma by previous studies (Abels et al., 2011;Bosboom et al., 2014;Dupont-Nivet et al., 2007;Huang & Hinnov, 2019;G. Q. Xiao et al., 2010) suggesting a major paleoenvironmental shift at this time.

Palynology
To study the vegetation response to the 40 Ma event, we complement the existing palynological record of the Xining Basin withsamples from the early and middle Eocene ( Figure 6). Only one sample from the early Eocene was productive and reveals a semi-arid steppe-forest ecosystem occupied by xerophytic grasses and halophytic shrubs (Graminidites and Nitraria respectively), that also hosted considerable areas of temperate woodland and ferns surrounding watercourses (Figure 7). The middle Eocene is represented by multiple samples indicating a more arid steppe-desert environment than the early Eocene, with a now-subordinate component of temperate broad leaved forest. This is followed by a major increase in both the abundance and diversity of the xerophytic and halophytic desert and steppe shrubs, Ephedra and Nitraria, and a concurrent decrease in temperate broad-leaved forest diversity. After this shift, steppe-desert vegetation (alternating between Ephedra and Nitraria being dominant) comprises around three-quarters or more of the palynological assemblage, with many samples including 80-95% steppe-desert pollen. This reflects a significant drying event previously recognized by Bosboom et al. (2014) and now better constrained by our pollen samples to occur between 40.7 and 39.9 Ma. This shift is again coeval with the onset of obliquity cycles (Bosboom et al., 2014) as well as the appearance of the loess-like end-member observed in our grain-size analyses ( Figure 6). The late Eocene record is characterized by the appearance of conifers at ∼37 Ma, as recognized in previous studies, and linked to cooling and aridification (Abels et al., 2011;Barbolini et al., 2020;Dupont-Nivet, Hoorn, & Konert, 2008;Page et al., 2019).

Discussion
Our study shows an abrupt shift occurring in the Xining Basin with the appearance of a loess contribution (EM4) at 40 Ma (Figure 6), coeval with the onset of obliquity-forced moisture cycles and a large increase in the proportional abundance of steppe-desert pollen previously recognized by Bosboom et al. (2014). Although the latter likely indicates a marked expansion of low-growing, bushy shrubs that would have acted as efficient local dust traps (Pye, 1995), the vegetation shift alone cannot explain the onset of dust in the basin because these taxa were also relatively abundant at intervals prior to 40 Ma ( Figure 6). Furthermore, these sudden changes are unlikely to be related to the slow local tectonics occurring in the Xining Basin at this time (L. G. Fan et al., 2019), but rather imply a regional reconfiguration of the atmospheric circulation.
MEIJER ET AL.

Onset of the Siberian High
Modern-day observations show that the northwesterly winter monsoons are driven by the Siberian High resulting in dust storms due to the break-up of this high-pressure system during spring (Roe, 2009). Therefore, the appearance of both the dust in the grain-size record and the prevailing northwesterly winds evidenced in the Baimasi section may indicate the establishment of a similar atmospheric configuration at 40 Ma. This is corroborated by provenance data showing that the source of the Xining dust was identical to the Quaternary loess suggesting that the Asian atmospheric circulation has been a resilient feature since the late Eocene (Licht et al., 2016). The establishment of the Siberian High would also explain the cyclicity appearing at 40 Ma, because this high-latitude pressure system is primarily driven by changes in insolation related to obliquity at orbital timescales (Shi et al., 2011) and interacts with atmospheric moisture from the monsoons and westerlies. A stronger Siberian High during high obliquity deflects more atmospheric moisture, whereas low obliquity results in a weaker high-pressure system and more moisture transported inland. During the Eocene, the Xining Basin is proposed to have been dominated by moisture derived from the wintertime westerlies  and therefore would have been especially susceptible to the Siberian High. Overall, the presence of the Siberian High would have enhanced the aridification of the region and driven the shift to widespread steppe-desert vegetation. We thus propose that the changes observed at 40 Ma could mark the onset of atmospheric circulation associated with the establishment of a high-pressure system developing over Siberia. While some climate model simulations using Paleogene boundary conditions suggest the onset of a proto-Siberian High as early as the late Paleocene with wind directions in agreement with our observations (Farnsworth et al., 2019), others suggest an absence of the Siberian High until at least the middle Eocene due to the warmer Eocene winter temperatures . Instead, these simulations show a belt of high atmospheric pressures between 20 and 45°N during winter with dominant surface wind directions in the Xining Basin toward the east  or northeast (Huber & Goldner, 2012). These ambivalent results may reflect variations in internal model properties representing winter conditions, and other boundaries conditions that may have driven the onset of the Siberian High: Paleogene atmospheric cooling, Tibetan Plateau uplift and proto-Paratethys Sea retreat all occurring during this time window. In the following, we consider the respective contributions of these proposed driving mechanisms.

Driving Mechanisms
In a first-order view atmospheric cooling is proposed to drive aridification (Dupont-Nivet et al., 2007;Lu et al., 2010) and expand Arctic sea ice strengthening the Siberian High (Roe, 2009). Global temperatures dropped during the Eocene starting at 50 Ma (Cramwinckel et al., 2018) and ephemeral sea ice in the Barents Sea is recorded since at least 45 Ma (Tripati & Darby, 2018), both predating the 40 Ma event in the Xining Basin. Furthermore, a recent study using clumped isotopes to reconstruct atmospheric temperatures in the Xining Basin found no cooling event at 40 Ma, but rather a two-tiered cooling at 37 Ma and at the Eocene-Oligocene Transition (EOT), 33.5 Ma (Page et al., 2019). Both cooling steps are interpreted to correlate with aridification events identified as a decrease in gypsum beds (Abels et al., 2011;Dupont-Nivet et al., 2007) and a decrease in reconstructed soil water δ 18 O . However, we do not observe an increase in the loess-like end-member at 37 Ma (Figure 6), nor in the Oligocene deposits after 34 Ma ( Figure S6), even though atmospheric cooling is expected to intensify the Siberian High. It seems that a threshold has been crossed previously at 40 Ma and subsequent changes did not significantly enhance dust deposition in the Xining Basin.
Climate model sensitivity simulations suggest that the winter monsoons and Siberian High are sensitive to uplift of the Tibetan and the Mongolian Plateau (Sha et al., 2015;Yu et al., 2018;C. Zhu et al., 2019). Uplift of the latter is poorly constrained to be late Cenozoic (Vassallo et al., 2007) and therefore postdates the onset of dust in the Xining Basin. In contrast, numerous studies report that the Tibetan Plateau has expanded northwards since 40 Ma (Rohrmann et al., 2012;C. Wang, Dai, et al., 2014) and may have intensified the Siberian High and dust transport (C. Zhu et al., 2019). However, it remains unclear whether this long-term process could have resulted in the abrupt shifts observed at 40 Ma in the Xining Basin.
Finally, we consider the sea incursions of the inland proto-Paratethys Sea (Figure 10). Recent studies constrained the extent and timing of these sea incursions and dated a major sea retreat to ∼41-40 Ma (Kaya MEIJER ET AL.

10.1029/2020PA003993
18 of 24 et al., 2019), coeval with the shifts in the Xining Basin. Sensitivity climate model simulations using various paleogeographies of the proto-Paratethys Sea show that the Asian climate is especially sensitive to this particular sea retreat from the Tarim and Siberian Sea (Z. Zhang et al., 2007). According to these simulations, this regression acts as a tipping point resulting in both the strengthening of a high-pressure system over Siberia due to increased continentality as well as the aridification of northwestern China by removing a major moisture source for this region (Figure 10). Although the aridification itself can explain the observed shifts in vegetation and the increase in dust at 40 Ma, it does not explain the onset of obliquity cycles and the northwesterly wind directions. Therefore, we propose that the sea retreat driving the appearance of the Siberian High is the most plausible mechanism to explain the shifts observed in the Xining Basin. However, it should be noted that other mechanisms such as atmospheric cooling and Tibetan Plateau uplift may have played a role as well and should be tested by future climate model simulations.

Regional Aridification and Dust Records
The onset of dust in the Xining Basin precedes the eolian deposits observed in most other regions of Central Asia, except for a poorly dated record at ∼51 Ma in the Xorkol Basin (Li et al., 2018), which is located outside of the Loess Plateau region and likely driven by a different atmospheric configuration. Dust is observed to significantly increase in the grain-size record of the neighboring Lanzhou Basin at ∼26 Ma (Y. Zhang et al., 2014). However, a recent reanalysis of this magnetostratigraphic record suggests a dust increase at ∼40 Ma , coeval with that observed in the Xining Basin and therefore suggesting a regional signal. Sedimentary records from Tajikistan show eolian dunes and loess shortly after the third Paleogene regression of the proto-Paratethys Sea (Carrapa et al., 2015),   Bosboom et al., 2011) showing the retreat of the proto-Paratethys Sea. This may have resulted in aridification and the onset of a high-pressure system over Siberia driving the East Asian winter monsoon and dust storms in the Xining Basin. of loess in Mongolia as well as eolian dunes in the Tarim Basin are observed at 34 Ma and are linked to the final retreat of the proto-Paratethys Sea at the EOT (Blayney et al., 2019;J. Sun & Windley, 2015). These records suggest that the continued retreat of the proto-Paratethys Sea resulted in the desertification of western Central Asia. Subsequent aridification due to atmospheric cooling and uplift during the Miocene may have progressively enhanced dust accumulation on the Chinese Loess Plateau as observed by the onset of dust at 26-20 Ma in the western part (Garzione et al., 2005;Guo et al., 2002;Jiang & Ding, 2010;X. Qiang et al., 2011;W. T. Wang et al., 2011) and at 11-8 Ma in the eastern part (Ao et al., 2016;B. Wang, Zheng, et al., 2014). Alternatively, these basal ages may reflect a tectonic control on the accommodation and preservation of loess rather than an atmospheric shift (B. Wang et al., 2018).
A poorly dated record from the eastern Pacific Ocean reveals the appearance of a minor detrital dust component between 50 and 40 Ma, followed by a large increase of this component at 20 Ma (Ziegler et al., 2007). Provenance data show that the northern Pacific Ocean is dominated by Asian dust since 40 Ma (Pettke et al., 2002). Dust accumulation rates of this record are low during the Cenozoic followed by a doubling at 25 Ma and an even larger increase in accumulation rate at 3-4 Ma (Janecek & Rea, 1983). However, a recent study with improved age constraints suggests that the accumulation rates increased continuously since at least 25 Ma rather than stepwise (W. . These records add to the notion for the onset of an initial dust component at 40 Ma followed by a progressive aridification of the Asian interior resulting in a large increase in dust accumulation during the late Oligocene to early Miocene.

Conclusions
We identified the appearance of a loess-like dust component in the terrestrial record of the Xining Basin at 40 Ma, coeval with the onset of obliquity cycles and a shift to dominance of steppe-desert vegetation previously identified by Bosboom et al. (2014). We propose that these shifts are linked to retreat of the proto-Paratethys Sea which, in addition to the long-term Tibetan Plateau uplift and global cooling, may have set up the Siberian High resulting in winter monsoonal circulation and associated dust storms. This obliquity-driven high-pressure system would have deflected westerly moisture and driven the cyclicity observed in the Xining Basin as well as regional aridification and corresponding biotic shifts. However, more Central Asian records are needed to confirm that the 40 Ma events are indeed a regional signal. Our study suggests that the Siberian High may have driven cold surges over eastern Asia since 40 Ma and should be taken into account when reconstructing temperatures in this region (Tang et al., 2015). Middle to late Eocene Asian dust may have played a critical role in the global cooling and set-up of the greenhouse to icehouse transition by scattering incoming radiation, promoting the formation of clouds and affecting the carbon cycle by fertilizing phytoplankton production in the Pacific Ocean (Jickells et al., 2005;Martin, 1990;Shaffer & Lambert, 2018). Especially because middle to late Eocene loessites have been identified in South and North America as well (Bellosi, 2010;M. Fan et al., 2020). The role of this Eocene dust as well as the mechanisms driving the Siberian High should be further explored and tested in future proxy records and climate modeling studies.

Data Availability Statement
Raw paleomagnetic data is available at earthref.org/MagIC/16967 and ChRM directions, grain-size, cyclostratigraphic (XRF and color) and palynological data are available at https://doi.pangaea.de/10.1594/ PANGAEA.923996 as well as in the Supporting Information.