Chemical variations of loess from the Chinese Loess Plateau and its implications

ABSTRACT The Chinese Loess Plateau (CLP) is the largest loess deposit on Earth with expansive surface-exposed source rocks of varying origins, ages, and history. Here, we present abundances of elements on representative loess and palaeosol samples from seven classic sections of the CLP. Most elements, including soluble elements (e.g. Rb and Cs), show significant correlations with La or Al2O3. These correlations indicate that these elements are hosted or absorbed in particle minerals during weathering, transport, and deposition (e.g. mica, K-feldspar, and clay minerals). These new observations allow the use of La/X (‘X’ being the element of interest) and the estimated La abundance of 31 ppm in the model upper continental crust (UCC) to estimate the abundances of other elements. The results show higher Cs (Cs = 6.7 ± 1.2 ppm), lower transition metals, Ba, and Ga. Given the high CaO and presence of carbonate in UCC rocks of both vast western China (the primary source for the CLP) and eastern China, we propose that these updates on the element abundances represent a refined model for the carbonate-bearing UCC.


Introduction
The scientific significance of the upper continental crust (UCC) on which we live is self-evident as it represents the surficial end-product of chemical differentiation of the Earth over its long and complex history. As the starting point to understand the formation of continental crust, geochemists have been endeavouring to estimate the average composition of continental crust, especially the UCC, over the past century. Two approaches have been commonly taken (Rudnick and Gao 2003): (1) weighted averages of bedrock compositions through systematic large-scale sampling (Shaw et al. 1967;Gao et al. 1998) and (2) to average compositions of fine-grained sediments or sedimentary rocks (e.g. shales) in the UCC (Condie 1993;Chauvel et al. 2014). These methods provide good constraints on the abundances of major elements and insoluble minor and trace elements of the UCC. However, the abundances of transition metals and soluble elements have not been well constrained because of preferential/ differential chemical weathering and limited analyses.
Loess, formed as the result of wind-blown silt-size particle deposition, occupies ~ 10% of the Earth's land surface (Figure 1;Pye 1995;Taylor et al. 1983). Because it was formed in arid and/or cold climatic conditions (i.e. peridesert loess and periglacial loess; Pye 1995), loess has experienced intensive physical weathering with limited chemical weathering during its transportation and deposition (Gallet et al. 1998;Peucker-Ehrenbrink and Jahn, 2001). Hence, loess best reflects source rock compositions with rather limited element fractionations. Furthermore, global loess shows similar patterns and ratios of rare earth elements (REEs) and Th (e.g. Taylor et al. 1983), which are also comparable to those of the UCC estimated by using other approaches ( Figure 2). Hence, loess can be taken as a natural mixture of materials eroded from the expansive regions of the surficial crust (Taylor et al. 1983;Gallet et al. 1998;Chauvel et al. 2014).
Previous studies, however, show that the contents of soluble trace elements (e.g. K and Ba) in loess are either uncorrelated or poorly correlated with insoluble  Li et al. 2020;Chauvel et al. 2014). (B) Distribution of potential source regions for the Chinese Loess Plateau (CLP) with highlights of sampling locations for this study (b; revised after Nie et al. 2014;Sun et al. 2020). CAO -the Central Asia Orogen, NTP -the Northern Tibetan Plateau, OP -the Ordos Plateau. (C-E) Photos of loess outcrops for representative sampling localities. elements, such as La (e.g. Rudnick and Gao 2003 and references therein), suggesting that soluble elements may have been differentially leached during weathering. Consequently, only the contents of insoluble trace elements (e.g. Nb, Ta, and Th) in loess (Figure 2(a, b)) are used to estimate their abundances in the UCC. Some other studies, however, suggest that the 'weathering signatures' may in fact be inherited from eroded bedrocks, recording previous weathering events (e.g. Peucker-Ehrenbrink and . Moreover, recent studies have also shown that the loess composition can vary with varying provenances in space and time (Chen et al. 2007;Chen and Li 2013;Nie et al. 2014;Sun and Zhu 2010). In addition, eroded materials may have experienced mineral sorting, which can lead to the fractionation of heavy minerals and their hosted elements (Taylor and McLennan, 1995).
Whether a chemical element in loess can be used to infer its abundance in the average UCC thus depends on the source inheritance and mixing processes. Therefore, a clear understanding of elemental behaviours in aspects of sedimentary processes to produce loess can help us to better constrain the average composition of the UCC using the loess composition. Compared to periglacial loess (e.g. western Europe loess, Argentinian), dust in peridesert loess was derived from different source rocks of a larger area with longer distance transport and more effective mixing (e.g. Chauvel et al. 2014). Among these peridesert loess deposits, the Chinese Loess Plateau (CLP) is the thickest and most extensive loess deposit on Earth (e.g. Liu and Ding 1998) with an exposure area in excess of c. 635,280 km 2 and thickness of up to c. 505 m locally .
Here, we present chemical compositions of loess samples from seven representative sections of the CLP (Figure 1(b-e)), and discuss the controls on the behaviours of various elements during sedimentary processes from weathering to erosion and to deposition. Based on the studies of the CLP loess, this study provides a refined model for the average UCC composition that incorporates carbonate, including soluble elements (e.g. K, Rb, Cs, Ba).

The geological setting of the CLP and sampling sections
Dust of the CLP is interpreted to have derived from vast arid lands upwind to the west and northwest (e.g. the Taklamakan Desert in western China, the Qaidam Gobi-Desert on the northern Tibetan Plateau, and the Badain Jaran and Tengger deserts in northern China) and  Gallet et al. 1998). (C) Chondrite normalized REE patterns and (D) primitive mantle normalized trace element patterns of the CLP samples using normalization values of Sun and McDonough (1989). Model UCC compositions by McLennan (1985, p. 1995) and Rudnick and Gao (2003) are plotted for comparison. The plotted data for global loess and palaeosols in the literature are compiled in Supplementary Table 3. originated from erosion of the surrounding mountain ranges (i.e. the Altyn Mountains, Qilian Mountains, and Kunlun Mountains on the north edge of the Northern Tibetan Plateau, and the Tianshan Mountains and Gobi Altay Mountains from the Central Asian Orogenic belt as shown in (Figure 1(b); Chen et al. 2007;Chen and Li 2013;Xiao et al., 2012). Because of the Tibetan Plateau uplift and Northern Hemisphere glaciation (e.g. An et al. 2001;Sun et al. , 2020Sun and Zhu 2010), vast areas of bedrocks have been exposed for erosion to supply aeolian dust to the CLP. Varied provenances and longdistance transport in combination facilitate effective dust mixing compared to periglacial loess (e.g. Western Europe loess; Chauvel et al. 2014;Sauzéat et al. 2015).
In this study, we sampled seven classic sections of the CLP: Luochuan (six loess samples and five palaeosol samples), Yan'an (two loess samples), Hukou (five loess samples), Lingtai (four loess samples), Lantian (three loess samples and two palaeosol samples), Jiuzhoutai (10 loess samples), and Qin'an (five loess samples; Figure 1(b-e)). The Qin'an section is significantly older, deposited at ~22-6.2 Ma as evidenced by palaeomagnetic data and fossils, while all the other CLP sections are Quaternary deposits of < c. 2.6 Ma (e.g. Guo et al. 2002). Bulk-rock loess samples were analysed in this study to avoid artificial fractionation caused by varying grain sizes, which reflect varied sources and complex sedimentary histories.

Analytical methods
The loess samples were carefully hand-crushed in an agate mortar into powders in a clean environment for bulk-rock analysis to avoid potential contaminations. The samples were analysed at the Laboratory of Ocean Lithosphere and Mantle Dynamics (LOLMD) in the Institute of Oceanology, the Chinese Academy of Sciences. Inductively coupled plasma optical emission spectrometer (ICP-OES, Agilent 5100) and inductively coupled plasma mass spectrometer (ICP-MS, Agilent 7900) were used for analysing major and trace elements, respectively. Strontium isotope compositions were measured on a multi-collector inductively coupled plasma mass spectrometer (MC-ICP-MS, Nu II). The analytical data of chemical compositions and Sr isotope compositions are given in Supplementary Tables 1-3, and the detailed information for analytical precision and accuracy is given in Supplementary File A.
For major elements, each dried sample powder of 50 mg mixed with 250 mg lithium borate flux (LiBO 4 ) was melted in a platinum crucible at 1050°C for 0.5 hour in a muffle furnace, followed by heating over a Bunsen burner at 1000°C. Then, the melted droplet was immediately dropped into c. 50 mL 5% HNO 3 solution and diluted to 100 mL with Milli-Q water (Kong et al. 2019). Our repeated analyses of the United States Geological Survey (USGS) rock standards (BHVO-2, AGV-2, and GSP-2) agree with recommended values, better than 5% for accuracy and better than 3% for precision (Supplementary File A). For loss on ignition (LOI), ~500 mg powder for each sample was heated in a muffle furnace at 1000°C for 0.5 hours with the calculated weight loss as the LOI.
For trace element analysis, 50 mg powder for each sample was dissolved in anti-aqua regia (HNO 3 : HCl = 3:1) and HF in a high-pressure bomb (a Teflon beaker in a stainless steel jacket) for 15 hours, followed by 2 hours of re-digestion with 20% HNO 3 , following Chen et al. (2017). Then, each sample solution was diluted to 100 g (with a dilution factor of 2000) in 2% HNO 3 for analysis. During analysis, ICP-MS was set in no gas mode instead of using collision mode, and calibration was performed based on five solutions (1, 10, 25, 50, and 100 ng/mL for all the analysed elements) acquired from multi-element calibration standard solutions (Agilent Technologies, Tokyo, Japan) with a blank. One replicate sample was analysed for every 10 samples, and a given lab-mixed solution was analysed for every four samples to monitor instrumental drift. USGS standards (BCR-2, AGV-2, and GSP-2) were analysed as unknowns to determine accuracy and precision (Supplementary File A). Accuracy, indicated by RE between the analysed values of USGS standards and their recommended values, is generally better than 10% for most trace elements (except for BCR-2 with 12% RE for Ni and Cu, and 15% for Sn), with many elements agreeing with the reference values within 5% (Supplementary File A). Precision, indicated by replicate analyses, is within 5% for most trace elements (Supplementary File A).
For Sr isotope analysis, 50 mg sample powder was decomposed in a high-pressure bomb by using HNO 3 + HCl + HF at 190°C for 15 hours, followed by redigestion with 2 mL 3 N HNO 3 for 2 hours. Then, sample solutions were loaded onto Sr-spec resin columns for Sr separation, and dry plasma mode with Aridus II was used for analysis. As the relative abundances of 83 Kr and 86 Kr are constant and no interference for 83 Kr,86 Kr was calculated based on the measured 83 Kr. The intensity of 86 Sr was acquired by reducing 86 Kr from the total measured isotope at mass 86. The measured 87 Sr/ 86 Sr ratios were normalized to 86 Sr/ 88 Sr = 0.11940 to remove the effect of instrumental mass fractionation. NBS-987 was analysed bracketing every four samples to monitor the instrument drift during the analysis. The repeated measurement of 87 Sr/ 86 Sr ratios of NBS-987 is 0.710244 ± 0.000009 (n = 9, 2σ), consistent with the reference value (0.710243-0.710250; http://georem. mpch-mainz.gwdg.de/sample_query_pref.asp). Analysis of BCR-2 gives 87 Sr/ 86 Sr = 0.705336 ± 0.000007, which agrees well with the recommended value of 0.70492 ± 0.00055.
Mineral phases in thin sections were identified by using focused ion beam (FIB) -scanning electron microscopy (SEM, Zeiss Gemini 2 crossbeam 550) with energydispersive spectrometer (EDS, Bruker QUANTAX) at China University of Petroleum (East China). To determine mineral modal abundances, advanced mineral identification and characterization system (AMICS) software was used.

Mineral counting results
The back-scattered images and mineral modal abundances acquired using SEM-EDS with AMICS are shown in Figure 3. The CLP loess and palaeosols are primarily composed of quartz + feldspar (albite, K-feldspar, and albite-anorthite) + carbonate + micas (both biotite and muscovite) + epidote with allanite + clay minerals (illite, kaolinite, and chlorite). Heavy minerals include amphibole, augite, garnet, rutile, titanite, apatite, and Feoxides. These mineral assemblages we obtained in the CLP loess confirm previous petrographic studies, in which all these minerals are reported either as single particles or as parts of rock fragments in the CLP loess (Jeong et al. 2008(Jeong et al. , 2011. Ca-carbonate minerals are common in the loess and can reach up to c. 40 wt.% (16QA-02) in the Qin'an section. Carbonate mainly occurs as detrital grains, evidenced by its morphology as single angular particles or as parts of rock fragments, together with some secondary carbonate (also see Supplementary File B). In comparison, the palaeosols have little carbonate and much less epidote group minerals, but more clays (Figure 3).

Trace elements
All the CLP loess and palaeosol samples show a uniform trace element pattern (Figures 2(c, d)), i.e. elevated Ba-Rb-Cs-Th-U-Pb-LREEs, low Nb-Ta-Sr, a clear negative Eu anomaly (i.e. Eu N /Eu* = Eu N /(Sm*Gd) 1/2 ≈ 0.6), and a flat HREE pattern. This pattern is similar to that of loess samples elsewhere reported in the literature, except that we do not see large positive Zr-Hf anomalies in the CLP loess and palaeosol samples (Figures 2(c,d)). Lanthanum of the CLP Quaternary loess samples is positively correlated with most analysed elements ( Figure 5(a-k) and Figure 6(a-f)), including both soluble elements (e.g. Rb-Cs-Ba-Pb) and insoluble elements (i.e. REE and Ti-Nb-Ta-Th), but not correlates with Zr or Hf ( Figure 5(l)). Strontium shows a positive correlation with CaO but not with others. In addition, molybdenum and U show scattered positive correlations ( Figure 5(p)), but they show no correlation with any other incompatible elements. Trace element contents of Qin'an loess samples are generally comparable with or lower than those of the CLP Quaternary loess, except for higher Sr, while the palaeosols have comparable or higher contents of most trace elements except for lower Sr (Figures 5-7  and Supplementary Table 1). Ratios of Nb/Ta (c. 13) and Zr/Hf (c. 38) of all our analysed CLP loess and palaeosol samples remain constant ( Figure 5(m, n)).

Strontium isotope ratios
Strontium isotope data for the CLP loess and palaeosols are given in Supplementary Table 2. The 87 Sr/ 86 Sr ratios range from 0.712474 to 0.71905, which increases with increasing 87 Rb/ 86 Sr ratios (Figure 7(c)). The CLP Quaternary loess has generally higher 87 Sr/ 86 Sr ratios than those of the Qin'an loess, and palaeosols show the highest 87 Sr/ 86 Sr ratios (Figure 7(c-e)).

Mineral controls on compositional variations of loess and palaeosols
The correlations between major element contents and mineral modal abundances (Figure 4(g-i)) suggest that the mineralogy controls the bulk-rock variations of for loess and palaeosol samples from the CLP. Major element contents shown in (A-E) have been recalculated following a sum of major elements to be 100% for comparison with the recommended UCC composition of . Rudnick and Gao (2003), as shown by thick crosses (also presented in Figures 5-7). Literature data (Supplementary Table 3) for the CLP loess and palaeosols and other loess on Earth are also plotted for comparison (also in Figures 5, 6). The symbols are the same in the following Figures 5-7 major element contents in the CLP loess and palaeosols. As shown in Figure 3 and Supplementary File B, carbonate minerals are important constituents of the CLP loess, which has also been reported in other CLP sections (e.g. Jeong et al. 2008Jeong et al. , 2011. Together with the variably high CaO and LOI and their positive correlation (Figure 4(f)), the significant correlation of carbonate with CaO and LOI (Figure 4(g, h)) demonstrate that calcite, not dolomite (as there is no correlation with Mg), is a significant constituent of the CLP loess and controls. Furthermore, CaO and LOI do not correlate with any other element, except for Sr. As shown in Figure 4(g) and Figure 7, it is the carbonate modes that control the abundances of CaO, Sr, Rb/Sr, and 87 Sr/ 86 Sr ratios.
Aluminium has the lowest water-rock partition coefficient compared with other major elements during chemical weathering and can be inherited from source rocks through the transformation of feldspar to clay minerals (such as illite and kaolinite). Thus, Al 2 O 3 has been taken as the constant index, and its relationship with other elements has been used to identify the mobility/immobility of these elements during sedimentary processes (e.g. Taylor et al. 1983). The positive correlations of Al 2 O 3 with SiO 2 -TiO 2 -K 2 O-FeOT-MnO (Figure 4) indicate the controls of silicate minerals (e.g. quartz, feldspar, and phyllosilicate minerals) on the varying SiO 2 -TiO 2 -K 2 O-FeOT-MnO abundances. Together with the negative correlations of Al 2 O 3 with CaO-LOI (e.g. Figure 4(a)), it reflects the complementary modal relationship between calcite and silicate minerals in the CLP loess and palaeosols (i.e. the more calcite, the less silicate minerals, and vice versa). In addition, the absence of correlations of Na 2 O and P 2 O 5 with any other elements in palaeosols, together with their lower Na 2 O and P 2 O 5 contents, indicates the loss of these two elements during post depositional weathering process, and P is also possibly removed by uptake in vegetation.

Inherited carbonate minerals in loess and their dissolution in palaeosols
Because carbonate minerals mainly occur as detrital grains indicated by their morphology (Figure 3), they are most likely derived from the provenance. Thus, the common presence of carbonate reflects the inherited signatures of high modes of carbonate and high CaO of source rocks. As carbonate is one of the most susceptible minerals to chemical weathering (e.g. Jeong et al. 2008), high modes of carbonate in loess reflect the limited dissolution of carbonate with little weathering, which is consistent with the arid environment for the formation of loess. Although some secondary carbonate minerals formed after loess deposition (Supplementary File B), they are likely re-precipitated following the dissolution of local primary carbonate (e.g. Jeong et al. 2008).
Compared with the CLP loess, the lower CaO-LOI-Sr and higher Rb/Sr and 87 Sr/ 86 Sr ratios (Figure 7) of the palaeosols reflect lower carbonate modes, which is consistent with the petrological observation ( Figure 3). As evidenced by higher chemical index of alteration (CIA = 100 * molar Al 2 O 3 /[Al 2 O 3 + CaO* + K 2 O + Na 2 O], and CaO* is corrected for apatite and carbonate; McLennan 1993), CLP palaeosols have experienced stronger weathering (64 vs. 58 for CLP Quaternary palaeosols and loess on average, respectively). Thus, the lower carbonate modes of palaeosols can be attributed to the stronger weathering under a relatively humid-warm environment, which can result in the dissolution of primary carbonate and subsequent re-precipitation of calcite at the base of the CLP palaeosols (e.g. Jahn et al. 2001). This further indicates that even if there are any externally derived fluids (e.g. rainfall, groundwater), they will lead to the dissolution of primary carbonate and potential reprecipitation of secondary carbonate, instead of supplying sufficient CaO for the direct precipitation of abundant carbonate.
Therefore, the common presence of calcite and high CaO-LOI in the CLP loess is most likely inherited from the provenance, which is dominated by siliciclastic and carbonate sedimentary rocks, granitoids, and their metamorphic equivalents (Jeong et al. 2008). Thus, our bulkrock analysis without prior chemical leaching is critically important in revealing the 'true' compositional makeup of the CLP loess.

Conservation of trace elements
Most of the analysed elements, including REEs-Th, Nb-Ta, transition metals, and Ga-Sn, show significant correlations with La (Figure 2(b), Figure 5(a-k) and Figure 6 (a-f)). These correlations reflect the inheritance of these elements from the source rocks with limited chemical Figure 6. Correlation diagrams of alkali and alkaline earth elements with La (A-F) and their correlations with micas + clay minerals + K-feldspar, which are thought to the important mineral hosts for these elements (G-H) for loess and palaeosol samples from the CLP. The correlation coefficients are given for analytical data of the CLP loess samples of this study and updated data on the CLP loess by Chauvel et al. (2014), except Be, contents of which are not available in the latter.
alteration during loess formation. The relatively constant Zr-Hf contents are probably caused by differential physical separation of zircons (and possibly other heavy minerals; Figure 5(l)). As previous studies reported for various minerals in river sediments (e.g. , epidote group minerals (epidote, allanite, zoisite), titanite, and clay minerals are significant for hosting REEs and Th; zircon and garnet are important hosts for HREEs, while zircon, rutile, and titanite are hosts for HFSEs (Supplementary File B).
For soluble elements, previous studies only showed good correlations of Li with LREEs in loess (Sauzéat et al. 2015). However, this study shows that soluble elements (i.e. Cs-Rb-Ba-K) of loess are also significantly correlated with La except Li ( Figure 6). These correlations indicate the conservation of these soluble elements during sedimentary processes to produce loess. Furthermore, the correlations of the total modal abundance of micas, K-feldspar, and clay minerals with Cs-Rb-Ba-K-Li (Figures 4(i) and 6(g, h)) reflect the accommodation of these elements by micas, K-feldspar, and clay minerals in the loess, the latter of which may inherit from the weathered source rocks. Because different trace elements are preferentially hosted in different minerals, the conservation of important mineral hosts during sedimentary processes as single particles or rock fragments is important to preserve geochemical signatures of the provenance.
Molybdenum and U are correlated ( Figure 5(p)) but do not correlate with any other elements. Moreover, in contrast to the constant Th/Nb ratio, Th/U ratios of the CLP loess vary significantly from 2.5 to 5.3 with an average value of 4.0 ( Figure 5(o)). Given the insignificant Figure 7. Correlation diagrams of Rb/Sr and 87 Sr/ 86 Sr ratios with CaO, Sr, Rb and mineral modal abundances to illustrate the influence of carbonate on these geochemical variations. More carbonate leads to increasing Sr content, while the decrease of micas + clay minerals + K-feldspar can lead to decreasing Rb content, both of which will result in the decrease of Rb/Sr and 87 Sr/ 86 Sr ratios finally.
chemical weathering under cold and arid conditions during sedimentary processes to produce loess, the variations of Mo and U may inherit geochemical signatures of bedrocks caused by previous weathering history. Because both Mo and U are sensitive to oxidation, this suggests that the source rocks may have suffered significant oxidation. As Carpentier et al. (2013) suggested, higher Th/U ratios of sediments derived from mature continental areas than those from juvenile terranes reflect the loss of U resulting from the long-term weathering history of source materials.

Temporal variations of the CLP loess
Compared to the CLP Quaternary sections, the Qin'an loess originated from the large arid area in the Asian interior during the early Miocene (e.g. Guo et al. 2002). Hence, the Qin'an and Quaternary sections offer the opportunity to determine if there may be any difference between CLP Miocene and Quaternary loess in terms of their dust sources as possible responses to tectonic and climate changes.
The Qin'an loess has higher CIA values than the Quaternary loess (i.e. 61 vs. 58 on average), indicating a stronger degree of chemical weathering. It also has abundances of most elements (i.e. SiO 2 -TiO 2 -Al 2 O 3 -Fe 2 O 3 -alkali elements-Ba-Pb-REEs-HFSEs-transition metals) comparable with or lower than those of the Quaternary loess, but has higher LOI-CaO-Sr (Figures 4-7). This is consistent with petrographically observed higher proportions of carbonate that dilutes the silicate constituents in the Qin'an loess. The welldefined linear correlation of 87 Sr/ 86 Sr with 87 Rb/ 86 Sr also reflects a clear two-component mixing relationship for both Quaternary Loess and the older Qin'an loess (Figure 7(c)). The radiogenic ingrowth is a function of Rb/Sr ratios, which decrease with increasing carbonate component and with the decreasing total modal abundance of micas + clay minerals + K-feldspar. Hence, lower 87 Sr/ 86 Sr ratios of the Qin'an loess compared to Quaternary loess are attributed to more carbonate and less micas + clay minerals + K-feldspar (Figure 7(d, e)).
The silicate components of the Qin'an loess also differ from those of the CLP Quaternary loess as indicated by different slopes in element covariation diagrams (Figures 5 and 6). The significantly lower Rb relative to the difference of Sr reflects that the silicate component of Qin'an loess has lower Rb/Sr ratios and incompatible elements than Quaternary loess (Figure 7(f, g)). Previous studies have also found systematic differences in Sr-Nd-Pb-O isotopic and chemical compositions in silicate fractions after removing carbonate from Miocene loess and Quaternary loess (Sun and Zhu 2010;Chen and Li 2013;Sun et al. 2020).
Moreover, although the pseudochron ages have no chronological significance for the loess deposition, this study gives an older pseudochron age of the Qin'an loess than the Quaternary loess, i.e. 352 vs. 246 Ma (Figure 7(c)). All these variations can be attributed to the change of the source material contribution, which is as the function of tectonic uplift and climate change Sun et al. , 2020Sun and Zhu 2010;Chen and Li 2013).

Using CLP loess to infer UCC
Unlike large-scale surface sampling of shields or orogens, fine-grained terrigenous sediments can supply a naturally well-mixed composition of eroded surfaces, eliminating the deviation caused by biased sampling or weighted averaging that may plague surface sampling (e.g. Condie 1993). Among fine-grained terrigenous sediments (e.g. shales), loess, especially that of peridesert origin, is an excellent candidate to estimate the average UCC composition, because of less chemical weathering and relatively long-distance transport from large source areas. It has been shown that the CLP loess deposits are the thickest and most extensive loess deposit on Earth, and the volume of dust eroded in total can be up to 190,584 km 3 given a thickness of 300 m . Moreover, as source regions of the CLP, bedrocks of surrounding deserts (Taklamakan, Gobi, Badain Jaran, and Tengger) and mountains (Altyn, Qilian, Kunlun, Tianshan, and Gobi Altay; Figure 1(b)), covering over 2 million square kilometres, range from Archaean to Cenozoic (e.g. Nie et al. 2014). Hence, the CLP loess is better representative for estimating the average UCC composition than other loess deposits (e.g. Chauvel et al. 2014;Sauzéat et al. 2015).
To estimate the UCC composition, we used 30 CLP Quaternary loess samples in this study. For the elements that were also reported by Chauvel et al. (2014; everything except Be-Ga-Mo-Sn-Tm-W), we also combined their data for 13 CLP loess samples with the data reported here to estimate the average composition of the UCC (Table 1). We did not use palaeosol samples because of the potential alteration associated with paedogenesis, such as the Sr loss in palaeosols as mentioned above. The Miocene Qin'an loess is also excluded to better constrain the UCC composition because of different correlations of various elements with La (Figures 5-7) and stronger effects of chemical weathering as indicated by their higher CIA values.

Major elements
The average contents of major elements of the CLP Quaternary loess agree well with the recommended values of the average UCC composition (Rudnick and Gao 2003) within ± 20% variation (Figure 8(a); Table 1) except for higher CaO (7.96 wt.%) and lower Na 2 O, the latter of which reflects the loss of Na during the sedimentary processes as discussed above. As this study and many others (e.g. Jeong et al. 2008Jeong et al. , 2011 have shown that abundant carbonate in the CLP loess is inherited from the provenances, high CaO contents of the CLP loess reflect the well-mixed composition of exposed surfaces upwind to the west and northwest of the CLP. Gao et al. (1998) also identified higher CaO contents (8.06 wt.%) and more abundant carbonate in their estimates based on large-scale sampling of eastern China than those estimates using exposed rocks from shields (Gao et al. 1998;Rudnick and Gao 2003). Moreover, the loess from Kaiserstuhl, Rhine Valley has much higher CaO contents, up to c. 23 wt.%, which reflects the presence of abundant limestone from the Alps (Taylor et al. 1983;Hu and Gao 2008). Hence, loess compositions of the CLP, together with large-scale exposed bedrocks in eastern China, may represent a carbonate-bearing model of the UCC composition. Because both the averaged CaO contents of eastern China and the CLP Quaternary loess are c. 8 wt.% (Table 1), we assume that this value may represent the CaO content of the carbonate-bearing UCC.
Previous studies suggested that loess had much higher SiO 2 than the model UCC composition by using the approach of large-scale surface sampling (e.g. Taylor et al. 1983;Gallet et al. 1998) because of preferential transport of quartz into loess and its preservation during sedimentary recycling processes (e.g. Taylor et al. 1983;Rudnick and Gao 2003). However, our study demonstrates that averaged contents of most major elements in carbonate-rich CLP loess are comparable with the proposed UCC composition including SiO 2 (Figure 8(a)). As discussed in previous study (Liang et al. 2009), removing the present carbonate in the CLP loess can lead to a ~3% difference for Al 2 O 3 content between the carbonate-leached CLP loess and those without leaching. Hence, the presence of carbonate dilutes the contents of most major elements in the CLP loess.

Trace elements
Although loess has experienced limited chemical alteration, most soluble elements are known to poorly correlate with La in loess, which may reflect the heterogeneous composition of source rocks after ineffective mixing (Figure 9(a)). Hence, previous studies used surface composites or element ratios to estimate abundances of these elements in the UCC (e.g. Rudnick and Gao 2003 and references therein). Here, we find significant correlations of Li-Be-K-Rb-Cs-Ba-Pb with La and Al 2 O 3 in CLP loess (Figures 4d and 6), demonstrating that these elements are quantitatively transported from source rocks to loess, which allow us to refine the abundances of these elements in the model UCC composition.
The Li value in this study is higher than recommended values (18-24 ppm) of Rudnick and Gao (2003 and references therein) but is identical to the values derived in recent studies, i.e. 35 ± 11 ppm based on correlations between Li and Nb in shales using MC-ICP-MS (Teng et al., 2004), 31 ± 5.2 ppm based on Neoproterozoic and Phanerozoic glacial diamictites (Gaschnig et al. 2016) and comparable with the value of 30.5 ± 3.6 ppm using correlations of Li with REEs in the recent loess study by Sauzéat et al. (2015). Based on the correlation of Li with In in well-characterized UCC samples (shales, pelites, loess, graywackes, granitoids, and their composites), Hu and Gao (2008) also provided a higher recommended Li value of 41 ppm.
The Cs value is also higher than most previous estimates (Figure 9(b)), except for the value of 6 ppm estimated by using sedimentary rocks (McDonough et al., 1992) and 7.3 ppm for marine sediments (Plank and Langmuir 1998). However, this value is comparable with that of loess from the CLP and Tadjikistan  (Supplementary Table 3; Hu and Gao 2008;Chauvel et al. 2014). Of the previous studies, Cs contents of shield composites were only given by Gao et al. (1998;3.6 ppm). However, due to the lack of significant Table 1. Recommended values of the UCC composition in different estimates (major elements are in wt.%, and trace elements are in ppm).
No.  Shaw et al., 1967, 1976Wedepohl, 1995a Fahrig and Eade, 1968Eade & Fahrig, 1973Condie, 1993Gao et al., 1998Gao et al., 1998Taylor and McLennan, 1985McLennan, 2001Rudnick and Gao, 2003Hu & Gao, 2008Gaschnig et al., 2016 Others This study b Ref.    Shaw et al., 1967, 1976Wedepohl, 1995a Fahrig and Eade, 1968Eade & Fahrig, 1973Condie, 1993Gao et al., 1998Gao et al., 1998Taylor and McLennan, 1985 Notes: a Updates of (Shaw et al. 1967(Shaw et al. ), 1976 Values of major elements are the average composition of the CLP Quaternary loess samples recalculated to a major-element sum of 100% in this study. Considering the significant correlations of most analysed elements with La, values of trace elements in the UCC are calculated by using the average La/X ratios (X is contents of the element of interest) and the widely accepted La content of 31 ppm. c The values in bold are the suggested updates of the UCC composition in this study. d As Sr negatively correlates with La, we use the correlation of Sr with 1/La to estimate the Sr value of the UCC by assuming La of 31 ppm.
Values of Mo and U are their average contents in the CLP loess samples, which are presented only for reference considering their differentiation from other elements during the sedimentary processes to produce the CLP loess. e The recommended value is referred to Rudnick and Gao (2003). correlations between Cs and La in sedimentary rocks, previous estimates for Cs primarily relied on the correlation of Cs with Rb and used the average Rb/Cs ratio, which is calculated assuming a content of Rb. The proposed values of Rb can range from 82 to 112 ppm in previous estimates (Hu and Gao 2008;Rudnick and Gao 2003 and references therein), and Rb/Cs ratios used also vary from 15.3 (Plank and Langmuir 1998) to 30 (Taylor and McLennan 1985), leading to high uncertainties of the Cs estimate (Rudnick and Gao 2003). This study is based on the significant correlation of Cs with La, such that it yields a more reliable value for Cs in the UCC. The Ba value is lower than previous estimates (Table 1 and Figure 9(b)). The significant correlations between Ba-Rb-Cs-K and their correlations with La ( Figure 6) suggest that Ba, like the alkali elements, is greatly conserved during sedimentary processes. Hence, given the comparable or considerably higher Rb and Cs contents of the CLP loess relative to their recommended values in the UCC, the lower Ba content should reflect the source rock composition. Based on the loess data of Chauvel et al. (2014), McLennan (2001, and Taylor et al. (1983), the Ba content is highly variable among different loess deposits (c. 190 ppm in Kaiserstuhl to c. 810 ppm in Iowa; Supplementary Table 3). However, because the CLP and Tadjikistan loess are important peridesert loess deposits originated from more expansive surface bedrocks (e.g. Chauvel et al. 2014), the consistently lower Ba contents of the loess from these two deposits may be more representative of Ba in the UCC.

Supplementary Materials Supplementary
The values of Sc-V-Cr-Co-Ni-Cu are lower than the values recommended by Rudnick and Gao (2003) and McLennan (1985, p. 1995) but are generally consistent with those of carbonate-bearing composite by Gao et al. (1998) and those of Gaschnig et al. (2016), which are also comparable with the values of Hu and Gao (2008), except for obviously higher V (see Table 1). The values of Sc, V, and Co are also similar to those of the Figure 8. The recommended values of the UCC composition in this study normalized to those recommended by Rudnick and Gao (2003). Major elements are in wt.% (A), while trace elements are in ppm (B). The lighter grey area represents the agreement of the recommended values within ±20%. The recommended values of Taylor and McLennan (1985) with updates by McLennan (2001), Gao et al. (1998;carbonate-bearing composition), and Gaschnig et al. (2016) are also given for comparison. Figure 9. (A) The modelling diagram to show the difference between peridesert loess and periglacial loess. Although chemical alteration during the processes to produce loess is limited, Ba-Rb-Cs-K as soluble elements lack correlation with La in previously studied loess, which likely reflect the heterogeneous composition of source rocks after an ineffective mixing. Hence, the observed correlation of Ba-Rb-Cs-K with La in Figure 6 of this study indicates that dust for the CLP loess has experienced effective mixing after long-distance transport. (B) The comparison of Ba-Cs contents of this study with previous estimates, i.e. G98 - (Gao et al. 1998) (including both carbonate-bearing and carbonate excluded, the plot with slash); G16 - (Gaschnig et al. 2016); W95 - (Wedepohl 1995); RG03 - (Rudnick and Gao 2003); TM85 - (Taylor and McLennan 1985) (related data are given in Table 1).
Canadian Shield (Eade and Fahrig 1973;Shaw et al. 1967Shaw et al. , 1976; Table 1). Transition metals are thought to be hosted in mafic minerals, which are preferentially incorporated into the fine-grained sediments . However, the significant correlations of these elements with La support that the lower contents of these elements are inherited from their source rocks. Our Ga value is comparable with the reported Ga contents of global loess (except Kaiserstuhl with a lower Ga of c. 7 ppm; Hu and Gao 2008) and with the recommended value of 14 ppm in the UCC by Wedepohl (1995), which adopted the value of Shaw et al. (1967) but is much lower than most other recommended values (Gao et al. 1998;Gaschnig et al. 2016;Hu and Gao 2008;Rudnick and Gao 2003;McLennan 1985, p. 1995).

Conclusions
This study presents chemical compositional data of the CLP loess, which originated from expansive source regions with long geological histories involving multiple sedimentary cycling, long-distance transport, and effective mixing. Given the high CaO and presence of carbonate in UCC rocks from both vast areas of western China (the primary source for the CLP) and eastern China, we propose that the CLP composition represents a possible model for the carbonate-bearing UCC. This study also shows significant correlations of most elements (even soluble elements) with La and Al 2 O 3 , indicating that most elements are conserved during the sedimentary processes that produce the CLP loess. Therefore, the ratios of La/X can provide better constraints on element values of the carbonate-bearing UCC, i.e. higher Cs (Cs = 6.7 ± 1.2 ppm), lower transition metals, Ba, and Ga.