Glacial and interglacial eolian dust dispersal patterns across the Chinese Loess Plateau inferred from decomposed loess grain-size records

Authors


Abstract

Previous studies have indicated that a genetically meaningful decomposition (unmixing) of loess grain-size distributions can be accomplished with the end-member modeling algorithm EMMA. The independent decomposition of two series of loess grain-size records from the NE Tibetan Plateau and Loess Plateau spanning the last glacial-interglacial cycle indicates that the two data sets are described by very similar mixing models. The average mixing model presented here is regarded as representative for the vast loess region in northern China and allows quantification of the contribution of three loess components to the loess grain-size distributions. A genetic interpretation and the paleoclimatic significance of the average mixing model have been provided by comparison of the modeled loess components with modern dust samples in terms of their grain-size distribution and flux rates, and by the distribution patterns of the loess components across the Loess Plateau reconstructed for the last two glacial-interglacial cycles. The sandy and silty loess components represent the coarse dust fraction supplied by saltation and short-term suspension processes over the proximal part of the Loess Plateau during major dust outbreaks in spring and early summer. The low-level winter monsoon (northwesterly wind system) is the likely transporting agent for these dust events. A clayey loess component represents the fine dust component supplied over the entire Loess Plateau by long-term suspension processes during major dust outbreaks and as part of a background supply system. The clayey loess component in the glacial loess deposits is dominantly supplied during major dust outbreaks by the northwesterly winter monsoon, whereas the clayey loess component in the interglacial paleosols is mainly supplied by non-dust-storm processes, possibly with a significant contribution by the high-level subtropical jet stream (westerly winds).

1. Introduction

Many geological investigations in the loess-covered region of north central China have used grain-size measurements as a basis for differentiating widespread loess and paleosol units, correlating them regionally, and relating them to the deep-sea isotope stratigraphy, and as a proxy for winter-monsoon strength [e.g., Liu, 1985; Kukla and An, 1989; An et al., 1991; Vandenberghe et al., 1997; Lu and Sun, 2000; Ding et al., 2002; Rokosh et al., 2003; Nugteren and Vandenberghe, 2004; Yang and Ding, 2004]. A fundamental problem with the interpretation of loess grain-size variability is to determine whether the observed grain-size variation should be attributed to mixing of eolian dust from multiple sources, to size-selective dispersal, or to some combination of both.

An inverse model of (un)mixing is ideally suited to obtain genetically meaningful interpretations of observed loess grain-size distributions (GSDs). Recent studies indicate that such a decomposition of loess GSDs can be accomplished with the end-member modeling algorithm EMMA [Vriend and Prins, 2005; Prins et al., 2007]. The method regards the measured grain-size distributions as a series of mixtures and allows estimation of the end-member (loess) components and their proportional contributions in the analyzed (loess) samples [Weltje, 1997; Prins and Weltje, 1999; Weltje and Prins, 2003, and references therein]. The independent decomposition of two series of loess grain-size records from the NE Tibetan Plateau and the Loess Plateau spanning the last glacial-interglacial cycle [Prins et al., 2007; M. Vriend et al., Contrasting dust supply patterns across the north–western Chinese Loess Plateau during the last glacial-interglacial cycle, submitted to Palaeogeography, Palaeoclimatology, Palaeoecology, 2007 (hereinafter referred to as Vriend et al., submitted manuscript, 2007)] indicate that the two data sets are described by very similar mixing models. The unmixing results revealed the existence of two contrasting dust supply patterns which were active over the Loess Plateau during the last glacial-interglacial cycle. A background sedimentation pattern dominant during interglacial periods, especially over the central and southern parts of the Loess Plateau, is reflected by the constant flux of a fine-grained loess (clayey silt) component. An episodic, thus highly variable sediment input pattern, dominant during glacial periods throughout the Loess Plateau and noticeable during interglacial periods only over the northern Loess Plateau, is reflected in the admixture of two coarse-grained loess (sandy silt) components.

The present paper provides a synthesis of the unmixing results obtained for ten loess grain-size records extending across the NE Tibetan Plateau and Loess Plateau in China. The data set includes the records presented by Prins et al. [2007] and Vriend et al. (submitted manuscript, 2007), as well as a few additional records presented by Nugteren and Vandenberghe [2004] and an unpublished record [Boos and Vermeer, 2006]. Most of the records cover the last glacial-interglacial cycle (n = 8); some records also cover the penultimate glacial-interglacial cycle (n = 5), thus extending the end-member modeling results further back in time compared to the previous published results [Prins et al., 2007; Vriend et al., submitted manuscript, 2007]. The aim is (1) to quantify the proportional contributions of the modeled loess components in the loess grain-size distributions with a forward mixing model (the end-members of the forward mixing model have been obtained by averaging the end-members of the “western mixing model” (Vriend et al., submitted manuscript, 2007) and the “eastern mixing model” [Prins et al., 2007]) and (2) to relate the spatial-temporal distribution and flux of the modeled loess components to possible sediment transport processes (suspension versus saltation) and dust supply patterns (winter monsoon versus westerlies).

2. Material and Methods

2.1. Loess Sections

The locations of the ten loess-paleosol sequences, which are distributed across the loess region in central northern China, are indicated in Figure 1. Two sites are located at the northeastern edge of the Tibetan Plateau in the NW–SE orientated valley of the Huang Shui River near Xining city, and are referred to as the Tuxiangdao (TXD) and Ledu (LD) sections. The Zhongjiacai (ZJC) section is located on the western Loess Plateau, ∼100 km northeast of Lanzhou. Description of these sections and grain-size distribution data are taken from Vriend and Prins [2005] and Vriend et al. (submitted manuscript, 2007). The remaining seven sections are obtained from the central Loess Plateau along two approximately north–south oriented transects. The western transect is formed by the Huanxian (HX), Xifeng (XF), Xunyi (XY) and Weihe Bridge (WB) sections. The eastern transect runs along the Yanan (YN), Luochuan (LC) and Duanjiapo (DJ) sections. Data of these sections (except the DJ section) are taken from Nugteren and Vandenberghe [2004]. The Duanjiapo section (34°10′N, 109°11′E) is located on the southeastern edge of the Loess Plateau near Lantian city [Boos and Vermeer, 2006].

Figure 1.

Location of studied loess sections which are distributed across the northeastern Tibetan Plateau and Loess Plateau in China. TXD, Tuxiangdao; LD, Ledu; ZJC, Zhongjiacai; HX, Huanxian; XF, Xifeng; XY, Xunyi; WB, Weihe Bridge; YN, Yanan; LC, Luochuan; DJ, Duanjiapo.

The sections cover variable time spans, including the last glacial-interglacial cycle (TXD, ZJC, WB) or last two glacial-interglacial cycles (HX, XF, XY, YN, LC). The LD and DJ sections only reach the last glacial period. In other words, the loess-paleosol sequences contain the S0-L1-S1-L2-S2 loess-paleosol complexes representing the last two glacial-interglacial cycles. Following An and Lu [1984], Kukla and An [1989], and Nugteren and Vandenberghe [2004], we assume that the boundaries between the loess and paleosol units are time-equivalent with the glacial-interglacial stage boundaries of the marine isotope stratigraphy [Martinson et al., 1987] (see Table 1). We realize that this might not be true in detail; however, for the purpose of highlighting the glacial-interglacial variability in sediment composition and dust flux, this assumption is valid.

Table 1. Age Estimates of the Lithostratigraphic Boundaries in the Loess-Paleosol Sequences According to the Marine Isotopic Stage Boundary Ages as Determined by the SPECMAP Recorda
Stratigraphic BoundarySPECMAP Isotope EventAge, ka B.P.
S0/L1-12.012.05
L1-1/L1-23.024.11
L1-2/L1-34.058.96
L1-3/S15.073.91
S1/L2-16.0129.84
L2-1/L2-2-159.89
L2-2/L2-36.5175.05
L2-3/S2-17.0189.61
S2-1/S2-2-220.14
S2-2/S2-3-233.20
S2-3/L38.0244.18

2.2. Grain-Size Analysis

The loess sections were sampled in vertical trenches at a 2.5 to 5-cm resolution, resulting in total in 5068 samples for the ten sections. Prior to the grain-size measurements the samples were prepared according to the methods described by Konert and Vandenberghe [1997]. About 1–2 grams of bulk sediment were pre-treated with H2O2 and HCl to remove organic matter and carbonates, respectively. In consequence, the results of the grain-size analysis reflect the grain-size distribution of the siliciclastic loess fraction. All measurements were performed on a Fritsch Analysette 22 laser particle sizer, which results in grain-size distributions with 56 size classes in the size range 0.15–2000 μm.

2.3. Decomposition of Grain-Size Distributions

Grain-size analysis of a loess sample results in a grain-size distribution referred to as the analyzed distribution ga. The analyzed distribution ga can be approximated by a theoretical grain-size distribution, referred to as the least squares fit distribution gf. The least squares fit distribution gf is a mixture of three end-members (EM-1, EM-2, EM-3) with proportions px:

equation image

where i represents a size class (1, 2, … 56), gf (i) defines the relative sediment “mass” (expressed in volume%) associated with the ith size class, equation imagepx = 1, 0 ≤ px ≤ 1, x refers to end-members EM-1, EM-2 and EM-3, and Sx refers to the grain-size distribution of the end-members. The compositions of the three end-members have been obtained by averaging the end-members of the “western mixing model,” based on the TXD, LD, ZJC and HX sections (Vriend et al., submitted manuscript, 2007), and the “eastern mixing model,” based on the HX, YN, LC and XY sections [Prins et al., 2007]. The least squares approximation of ga by gf involves minimization of the square root of the sum of the squared errors (RMS error) between ga and gf for all i size classes simultaneously. Hence the RMS error is given by

equation image

Minimization of the RMS error was achieved using Microsoft Excel Solver. Solver uses the Generalized Reduced Gradient (GRG2) nonlinear optimization code developed by Lasdon et al. [1978]. The method retains a unique minimum value for the RMS error that is represented by similarly unique end-member proportions px. This way a (forward) mixing model has been constructed that express all the loess samples (n = 5068) as mixtures of the three end-members.

2.4. Flux Model

Mass accumulation rates (MAR, in g/cm2/ka) of ten well-constrained lithological units (L1-1, L1-2, L1-3, S1, L2-1, L2-2, L2-3, S2-1, S2-2, and S2-3) have been calculated according to

equation image

where SR is the sediment accumulation rate (in cm/ka), and BD is the sediment dry-bulk density (in g/cm3). SR values have been calculated on the basis of sediment thickness values and age estimates for the loess and paleosol intervals on the basis of pedostratigraphy. Site specific BD values have been determined for the TXD, LD and ZJC sections (Vriend et al., submitted manuscript, 2007). For the other sections, BD values of 1.48 g/cm3 have been assumed [cf. Kohfeld and Harrison, 2003]. Data are listed in Tables 1 and 2.

Table 2. Data of the Studied Loess Sections Used to Calculate the End-Member Specific Fluxes, Including Sedimentation Rate, Mass Accumulation Rate, and End-Member Proportionsa
SectionStratigraphyThickness, cmSR, cm/kaBD, g/cm3MAR, g/cm2/kapEM-1pEM-2pEM-3
  • a

    SR, sedimentation rate; MAR, mass accumulation rate; pEM-1, pEM-2, pEM-3, end-member proportions.

TuxiangdaoS0185---0.480.010.51
L1-118526.41.3736.20.700.150.16
L1-2180556.41.3777.30.550.390.06
L1-357538.31.3752.50.590.300.12
S185515.31.3721.00.210.450.33
LeduS040---0.170.140.69
L1-163052.51.2766.70.470.070.46
L1-2139055.61.2770.60.240.450.31
ZhongjiacaiL1-151042.51.2754.00.140.700.16
L1-262017.71.2722.50.010.610.38
L1-362041.31.2752.50.040.740.23
S163011.31.2714.40.000.340.65
HuanxianS040---0.260.480.26
L1-180066.31.4898.20.210.730.06
L1-254515.61.4823.10.000.620.37
L1-355036.81.4854.50.070.810.12
S13856.91.4810.20.020.210.77
L2-1133044.31.4865.50.420.490.09
L2-221514.21.4821.00.030.580.39
L2-332522.31.4833.00.250.600.16
S2-11605.21.487.80.140.210.65
XifengS0175---0.030.570.41
L1-126021.61.4831.90.000.750.25
L1-246013.21.4819.50.000.410.59
L1-336524.41.4836.10.000.600.40
S13155.61.488.30.040.150.81
L2-138512.81.4819.00.020.690.30
L2-21258.251.4812.20.010.420.56
L2-315010.31.4815.30.010.540.45
S2-11454.81.487.00.020.260.73
S2-218013.81.4820.40.020.700.28
S2-311010.01.4814.80.020.280.69
XunyiS0105---0.040.170.78
L1-127522.81.4833.80.000.370.62
L1-23108.91.4813.20.000.160.83
L1-323015.41.4822.80.000.310.69
S13005.41.487.90.030.050.92
L2-149016.31.4824.10.000.570.43
L2-2905.91.488.80.000.210.78
L2-321514.81.4821.90.000.390.61
S2-11705.61.488.20.050.050.90
S2-2856.51.489.60.010.350.64
S2-3908.21.4812.10.040.090.87
WeiheS0160---0.020.300.68
L1-136029.91.4844.20.000.530.47
L1-238010.91.4816.10.000.300.70
L1-320013.41.4819.80.000.360.64
S14107.31.4810.90.010.070.92
YananS050---0.220.600.17
L1-152543.51.4864.40.240.700.06
L1-257016.41.4824.20.010.710.28
L1-339026.11.4838.60.050.880.08
S13305.91.488.70.050.500.45
L2-1122540.81.4860.30.470.470.06
L2-219512.91.4819.00.020.630.35
L2-333022.71.4833.50.170.790.05
S2-12157.01.4810.40.090.430.49
S2-219514.91.4822.10.250.690.06
S2-317015.51.4822.90.110.470.41
LuochuanS045---0.000.650.35
L1-121517.81.4826.40.000.710.29
L1-22808.01.4811.90.000.410.59
L1-324516.41.4824.30.000.560.44
S12003.61.485.30.040.170.79
L2-133511.21.4816.50.030.780.19
L2-21006.61.489.80.010.400.59
L2-31107.61.4811.20.000.590.41
S2-11354.41.486.50.050.200.74
S2-2604.61.486.80.000.640.36
S2-3454.11.486.10.040.250.71
DuanjiapoS0123---0.000.060.94
L1-119316.01.4823.60.000.060.94
L1-22136.11.489.00.000.001.00

Fractionated mass accumulation rates (fluxes) for the modeled end-members (FEM-x, in g/cm2/ka) have been calculated according to

equation image

where pEM-x is the proportional contribution (dimensionless) of end-member EM-x, and equation image px = 1. In these calculations we assume that the loess-paleosol samples are entirely composed of siliciclastics, which is just a first-order approximation as it neglects the contribution of other (non-siliciclastic) sediment phases like carbonates (detrital, pedogenic) and organic carbon.

3. Results

3.1. Mixing Model

3.1.1. End-Members

The composition of the three end-members of the forward mixing model have been obtained by averaging the end-members of the “western mixing model” (Vriend et al., submitted manuscript, 2007) and “eastern mixing model” [Prins et al., 2007]. The “average” end-members and the two sets of “original” end-members are shown in Figure 2. The “average” end-members are characterized by unimodal, fine-skewed grain-size distributions with a mode at ∼63 μm (EM-1), ∼37 μm (EM-2) and ∼22 μm (EM-3), respectively. The sand (>63 μm):silt (8–63 μm):clay (<8 μm [cf. Konert and Vandenberghe, 1997]) ratio is 46:45:9 for EM-1, 16:65:19 for EM-2, and 1:58:41 for EM-3. Hence EM-1 represents a sandy loess, EM-2 a silty loess, and EM-3 a clayey loess component [cf. Pye, 1995]. The median RMS error for the analyzed loess distributions is ∼4.5% (Figure 3), indicating that the analyzed grain-size distributions are approximated closely as mixtures of the three end-members.

Figure 2.

End-members (bold distributions with symbols) of the forward mixing model applied to the northeastern Tibetan Plateau and Loess Plateau data set. The end-members of Prins et al. [2007] (dashed curves) and Vriend et al. (submitted manuscript, 2007) (thin lines) are shown for comparison. The end-members represent sandy loess (EM-1, modal size ∼63 μm), silty loess (EM-2, modal size ∼37 μm), and clayey loess (EM-3, modal size ∼22 μm).

Figure 3.

Frequency and cumulative distribution of the root mean square (RMS) error for all the analyzed loess samples (n = 5068) in the ten loess sections.

3.1.2. Temporal Distribution of End-Members in the Yanan Record

The mixing model expresses the loess GSDs as mixtures of the three end-members. As an example, the median grain-size and the proportional distribution of the three end-members with depth at the Yanan site are shown in Figure 4. The major loess and paleosol units are characterized by significantly different end-member contributions. The coarse-grained loess units L1-1 and L2-1, which are deposited during the full glacial periods corresponding with marine-isotopic stage (MIS) 2 and the last half of MIS 6, are dominated by the sandy and silty loess end-members (EM-1 and EM-2). In contrast, the fine-grained paleosol units S1, S2-1 and S2-3, which are deposited during the full interglacial periods corresponding with MIS 5 and (part of) 7, are characterized by high contributions of the clayey loess end-member (EM-3) and (near) absence of the sandy loess end-member (EM-1). Loess units L1-2 and L2-2, which are in fact weakly developed paleosols within loess units L1 and L2, are very similar in composition to paleosols S1 and S2. Paleosol unit S2-2, actually a loess unit intercalated between paleosols S2-3 and S2-1, is characterized by high proportions of the coarse-grained end-members EM-1 and EM-2. The overall trend in end-member contributions with depth thus clearly mirrors the grain-size record; both records are characterized by a clear glacial to interglacial variability, which appears to be typical for the Loess Plateau [see also Prins et al., 2007; Vriend et al., submitted manuscript, 2007].

Figure 4.

Median grain size (D50) record and proportional contribution of the end-members in the Yanan loess-paleosol sequence compared to loess (L1, L2, L3) and paleosol (S0, S1, S2) lithostratigraphy [cf. Nugteren and Vandenberghe, 2004].

3.1.3. Spatial Distribution of End-Members Across the Loess Plateau

The average proportional contributions of the end-members in the major loess and paleosol units as recorded in all studied loess-paleosol sequences are given in Table 2. Contour plots of the average pEM-3 and the average [pEM-1/(pEM-1 + pEM-2)] ratio across the loess region in central China have been constructed for eight lithological units spanning the last two interglacial-glacial cycles (last ∼220 ka). The contour lines have been drawn by hand to highlight apparent geographical trends in the loess grain-size distributions. The “trend analysis” is based solely on the data of the eight sections distributed across the Loess Plateau; sections TXD and Ledu have been excluded from this analysis as those two sections are known to be characterized by very different end-member distribution patterns (Vriend et al., submitted manuscript, 2007).

The spatial distribution of the clayey loess component in terms of its average proportional contribution to the bulk loess and paleosol deposits (pEM-3) is shown in Figure 5. All eight lithological units show clear north-to-south trends in pEM-3, with low pEM-3 values recorded in the proximal sites at the northern margin of the Loess Plateau (ZJC, HX, YA) and high pEM-3 values recorded at the distal sites located near the southern edge of the plateau (XY, DJ). The two sections from the central part of the Loess Plateau (XF, LC) are always characterized by intermediate pEM-3 values. The WB section is somewhat exceptional as it recorded relatively low pEM-3 values in units L1-1 and L1-2, which do not fit within the general observed compositional trends (Figures 5a and 5b; numbers between brackets).

Figure 5.

Contour plot of average proportional contribution of end-member EM-3 in major loess and paleosol units across the northeastern Tibetan Plateau and Loess Plateau. (a) L1-1, (b) L1-2, (c) L1-3, (d) S1, (e) L2-1, (f) L2-2, (g) L2-3, (h) S2-1. Triangles indicate the location of the studied loess sections (compare with Figure 1); small numbers indicate end-member proportions for the individual sections; large numbers and bold dashed lines indicate contoured end-member proportions; and the Huanghe (Yellow) and Weihe rivers are indicated by the thin lines.

Although the north-to-south trend of increasing pEM-3 values is observed in both the loess and the paleosol units, the compositional gradient appears to be variable through time. The compositional gradients turn out to be steepest in the full-glacial loess units, and gentler in the interglacial paleosol units. This can readily be seen when the contoured pEM-3 distributions in the loess units L1-1 and L2-1 (Figures 5a and 5e) are compared with the distributions in paleosol units S1 and S2-1 (Figures 5d and 5h). The change in gradient between glacial and interglacial conditions has been illustrated also in Figure 6, where the pEM-3 distributions in the L1-1 and S1 units in the eight sections from the Loess Plateau are projected on a ∼400 km long NW–SE transect running between the HX section in the north to the WB section in the south (X–Y profile; see Figure 1). Disregarding section WB, a quasi-exponential north-to-south increase of the clayey loess component EM-3, ranging between 6% and 94%, is recorded in the L1-1 unit. The gradient is considerably gentler in the S1 unit, varying between 45% and 92%.

Figure 6.

Spatial distribution of the proportional contribution of the fine-grained loess component (pEM-3) in loess unit L1-1 and paleosol S1 along a north-to-south profile across the Loess Plateau. The profile runs approximately between the Huanxian site in the north and the Weihe Bridge site in the south (profile X–Y, indicated in Figure 1). Note the exceptional low pEM-3 value in loess unit L1-1 at the WH site.

The total contribution of the two coarsest-grained loess components (end-members EM-1 and EM-2) to the loess and paleosol deposits can be “constructed” from the pEM-3 distribution patterns shown in Figures 5 and 6, as the three end-members sum up to 1 (or 100%) by definition (pEM-1 + pEM-2 = 1 − pEM-3). Hence the proportional contribution of the coarse loess fraction (pEM-1 + pEM-2) shows a north-to-south decreasing trend in all the loess and paleosol units. End-member EM-1 reaches significant contributions only in the loess units, especially in the L1-1 and L2-1 units, and is (nearly) absent in the paleosol units (Table 2). To illustrate the geographical distribution of the relative contribution of end-member EM-1 to the coarse fraction, contour maps of the average [pEM-1/(pEM-1 + pEM-2)] ratio in the loess units L1-1 and L2-1 have been constructed (Figure 7). From Figure 7 (and from Table 2) it can be seen that significant contributions of end-member EM-1 are mainly recorded in the three sections (ZJC, HX, YA) located near the northern margin of the Loess Plateau. The recorded [pEM-1/(pEM-1 + pEM-2)] gradient appears to be steeper in the penultimate glacial loess unit L2-1 compared to the gradient in the last glacial loess unit L1-1. Moreover, it must be noted that exceptional high [pEM-1/(pEM-1 + pEM-2)] ratios are recorded in sections TXD and LD in loess unit L1-1 (Figure 7a).

Figure 7.

Contour plot of average end-member ratio [EM1:(EM1 + EM2)] in loess units (a) L1-1 and (b) L2-1. Triangles indicate the location of the studied loess sections (compare with Figure 1); small numbers indicate end-member ratios for the individual sections; large numbers and bold dashed lines indicate contoured end-member ratios; and the Huanghe and Weihe Rivers are indicated by the thin lines.

3.2. Dust Flux Model

3.2.1. Fractionated Dust Flux Estimates

Bulk mass accumulation rates (MAR) and fractionated fluxes of the three end-members (FEM-1, FEM-2, FEM-3) were calculated for the eight pedostratigraphic units for each of the ten sections separately on the basis of the data listed in Tables 1 and 2. Bulk MAR values range between ∼5 and ∼98 g/cm2/ka, with the minimum value established for the S1 paleosol unit of the LC section and the maximum value established for the L1-1 loess unit of the HX section. On the Loess Plateau, the paleosol units S1 and S2-1 are characterized by relatively low MARs (5–11 g/cm2/ka), paleosol unit L1-2 and L2-2 by intermediate MARs (9–24 g/cm2/ka), and the loess units L1-3, L2-3 (11–55 g/cm2/ka) and especially L1-1 and L2-1 by relatively high MARs (17–98 g/cm2/ka). The northern sections (ZJC, HX, YN) are characterized by overall higher MARs when compared with the central and southern sections (see also Nugteren and Vandenberghe, 2004). The Tibetan Plateau sections (TXD, LD) are characterized by overall high MARs, with the highest MARs in the L1-2 unit, intermediate MARs in loess units L1-1 and L1-4, and low MARs in paleosol unit S1.

A scatterplot of the summed fractionated fluxes of end-members EM-1 and EM-2 (FEM-1 + FEM-2) and the bulk mass accumulation rates (MAR) recorded in the eight lithological units in the ten loess-paleosol sequences is shown in Figure 8. Overall, a linear relation appears to exist between the two parameters which is adequately described by the following regression equation:

equation image

The observed offset between the regression line and the line y = x is obviously due to the contribution of end-member EM-3 to the bulk MAR. Overall, the average EM-3 fluxes (FEM-3) vary between 1 and 30 g/cm2/ka. However, equation (5) indicates that most of the estimated EM-3 flux (FEM-3) values fall within a narrow range centered at ∼9 g/cm2/ka. The FEM-3 is thus relatively low and especially constant compared to the highly variable bulk MAR. These results suggest that the pEM-3 record, contour plots and profiles shown in Figures 4 to 6 can be interpreted as “dilution records.” The variations in EM-3 content are thus dominantly caused by a variable input of end-members EM-1 and EM-2, with low EM-3 contents reflecting high MARs (high FEM-1 + FEM-2) and high EM-3 contents reflecting low MARs (low FEM-1 + FEM-2). Very similar results have been described by Prins et al. [2007] and Vriend et al. (submitted manuscript, 2007); their regression equations are shown in Figure 8 for comparison.

Figure 8.

Scatterplot of summed EM-1 and EM-2 fluxes (FEM-1 + FEM-2) versus total mass accumulation rate (MAR). Average flux estimates of all major loess (L1-1, L1-2, L1-3, L2-1, L2-2, and L2-3) and paleosol (S1, S2-1, S2-2, and S2-3) units in the ten loess-paleosol sequences are shown. Linear regression for the complete data set is shown (bold line; y = 0.99x + 8.6, r2 = 0.93). Linear regressions (dashed lines) of the mixing models of Prins et al. [2007] and Vriend et al. (submitted manuscript, 2007) are shown for comparison.

3.2.2. Fractionated Dust Flux Distribution Patterns Across the Loess Plateau

The spatial distribution of the fractionated flux estimates of end-members EM-1 and EM-2 (FEM-1 + FEM-2) within the eight major loess and paleosol units spanning the last two glacial-interglacial cycles is illustrated in Figure 9. The (FEM-1 + FEM-2) flux patterns show clear north-to-south decreasing trends in all eight loess and paleosol units. Overall, the highest fractionated dust fluxes are recorded in loess units L1-1 and L2-1, somewhat less high fluxes in loess units L1-3 and L2-3, intermediate fluxes in the weakly developed paleosols L1-2 and L2-2, and lowest fluxes in paleosol units S1 and S2-1. Within every “time slice,” relatively high flux rates are recorded at the proximal sites near the northern margin of the Loess Plateau (ZJC, HX, YN), intermediate flux rates at the sites on the central plateau (XF, LC) and low flux rates at the distal sites located near the southern edge of the plateau (XY, DJ). The glacial to interglacial variability in fractionated dust flux is illustrated in Figure 10, where the distribution patterns of (FEM-1 + FEM-2) along the X–Y transect are shown for loess unit L1-1 and paleosol unit S1. In loess unit L1-1, a sharp (quasi-exponential) north-to-south decreasing and overall high flux rate is observed, whereas in paleosol unit S1 the flux rates are very low and the trend less evident. The WB section is again somewhat exceptional as it is characterized by relatively high flux rates in unit L1-1 and, to a lesser extent, unit L1-2 (Figures 9a, 9b, and 10).

Figure 9.

Contour plot of average summed EM-1 and EM-2 fluxes (FEM-1 + FEM-2) in major loess and paleosol units across the northeastern Tibetan Plateau and Loess Plateau. (a) L1-1, (b) L1-2, (c) L1-3, (d) S1, (e) L2-1, (f) L2-2, (g) L2-3, (h) S2-1. Triangles indicate the location of the studied loess sections (compare with Figure 1); small numbers indicate individual observations; large numbers and bold dashed lines indicate contoured end-member proportions; and the Huanghe and Weihe rivers are indicated by the thin lines.

Figure 10.

Spatial distribution of the fractionated dust flux of the coarse-grained loess components (FEM-1 + FEM-2) in loess unit L1-1 and paleosol S1 along a north-to-south transect across the Loess Plateau. The profile runs approximately between the Huanxian site in the north and the Weihe Bridge site in the south (profile X–Y, indicated in Figure 1). Note the exceptional high (FEM-1 + FEM-2) value in loess unit L1-1 at the WB site, probably due to the admixture of dust entrained in the Weihe floodplain.

Figure 11 illustrates the spatial distribution patterns of the fractionated flux estimates of end-member EM-3 (FEM-3). A clear glacial to interglacial variability in the distribution patterns can be seen with north-to-south increasing (i.e., south-to-north decreasing) flux trends recorded within loess units L1-1 and L1-3 (and less clearly visible or even absent in units L2-1 and L2-3), and less clear (west-to-east?) decreasing trends in units L1-2 and S1 (and no trend in units L2-2 and S2-1, which may be partly related to the poor geographical coverage of data points). No clear temporal, glacial-interglacial variability in the FEM-3 flux estimates are recorded at most of the individual sites, indicating that the flux of end-member EM-3 is relatively constant (and low) through time, as already indicated in the previous paragraph. It should also be noted that relatively high FEM-3 values are recorded at the LD site, both in unit L1-1 and L1-2, clearly different from what has been recorded at the TXD site.

Figure 11.

Contour plot of average EM-3 fluxes (FEM-3) in major loess and paleosol units across the northeastern Tibetan Plateau and Loess Plateau. (a) L1-1, (b) L1-2, (c) L1-3, (d) S1, (e) L2-1, (f) L2-2, (g) L2-3, (h) S2-1. Triangles indicate the location of the studied loess sections (compare with Figure 1); small numbers indicate individual observations; large numbers and bold dashed lines indicate contoured end-member proportions; and the Huanghe and Weihe rivers are indicated by the thin lines.

4. Discussion and Conclusions

4.1. Grain-Size Distribution of the Modeled Loess Components

“Typical” loess sediments are characterized by fine-skewed GSDs, with modal grain sizes in the silt to very-fine-sand range [e.g., Pye, 1995]. In general, one can state that there is consensus on the significance of sand- and silt-sized particles in loess sediments, namely that they represent primary dust particles supplied to the deposition site primarily by saltation and suspension processes. However, various interpretations of the origin of clay-size particles and the significance of the clay content in loess sediments can be found in the literature. Post-depositional weathering and pedogenesis is often held responsible for alteration of the loess GSDs [e.g., Xiao et al., 1995; Porter and An, 1995; Kemp, 2001; Sun et al., 2006]. Chemical weathering of unstable minerals (e.g., feldspar) will result in the transfer of grains from the silt and sand fractions to the clay fraction, as a result of breakdown of these minerals and in-situ clay mineral formation, eventually resulting in the enrichment of clay-size particles. Following this line of reasoning, various studies have used the clay content of loess deposits as a proxy for summer-monsoon rainfall intensity, which is thought to be the main driving mechanism of post-depositional weathering and pedogenesis [e.g., Feng and Wang, 2006, and references therein]. Others suggested therefore to use the GSD of the weathering-resistant quartz fraction as a more reliable proxy indicator of winter monsoon strength [e.g., Xiao et al., 1995; Porter and An, 1995; Sun et al., 2006]. Sun et al. [2006] determined the degree of pedogenic modification of two loess-paleosol and red clay sequences located on the Loess Plateau by comparing the bulk-sample GSDs and the quartz GSDs. Their results indicate that the pedogenic modification of the glacial loess deposits (e.g., L1 and L2) is negligible, and that the effect of pedogenic modification is only significant in highly weathered interglacial paleosol units (e.g., S1 and S2) found at the southern Loess Plateau. In general, however, their results suggest that the loess-paleosol sequences discussed here have been subjected to only minor pedogenic alteration and that the loess deposits are predominantly of primary eolian origin.

Others assumed that clay-size particles in loess sediments represent primary dust grains, often citing studies which showed that the dust component in Pacific marine sediments exists of very-fine silty to clay-size particles [e.g., Ono and Irino, 2004, and references therein]. Recently, Sun et al. [2002, 2004], Sun [2004] and Qin et al. [2005] proposed a decomposition of loess GSDs into two or three end-member (lognormal or Weibull) distributions by parametric curve-fitting procedures. Parametric decomposition assumes that the end-members that make up an observed GSD are continuous unimodal distributions, which can be adequately described by analytical functions (e.g., lognormal, Weibull distributions) with a small number of parameters. Standard curve-fitting techniques may be used to decompose a single observed GSD into proportional contributions of analytical distribution functions belonging to a predefined class. This method cannot be used to simultaneously decompose a series of GSDs. The latter studies distinguished a silty loess component (with a mode at generally >20 μm) and one or two loess components within the clay fraction. Sun [2004] distinguished one clay component referred to as the fine component, with a modal grain size varying between 2 and 10 μm. Qin et al. [2005] recognized two components within the fine fraction referred to as the medium and fine modes, which have modal grain sizes in general between ∼3.5–5.5 μm and ∼0.5–1 μm, respectively. Sun [2004] interpreted the fine component as the background dust load of the atmosphere which is thought to be mainly transported by high-altitude westerly airstreams, and used the modal grain size of the fine component as a proxy for high-level westerly air stream intensity. The coarse, most abundant component is interpreted to be the product of dust storms generated by low-altitude northwesterly winds. Qin et al. [2005] proposed a very complex model in which they relate the median grain size and the proportional contribution of the three modes to variations in aerodynamic forcing (lift force related to vertical wind and turbulence) during dust entrainment in the source area and turbulence intensity in the depositional area (in order to reveal the aerodynamic patterns and evolution of the dust source area and the dust depositional area).

As there is no compelling reason why end-members should fit into any particular class of parametric models, Vriend and Prins [2005], Prins et al. [2007] and Vriend et al. (submitted manuscript, 2007) applied a non-parametric inversion technique (the end-member modeling algorithm EMMA [Weltje, 1997]) to obtain mixing models of their loess grain-size records from the loess region in China. In contrast to curve-fitting algorithms, EMMA does not require any case-specific assumptions, i.e., the number of end-member GSDs and their shapes do not have to be specified. EMMA does not make use of parametric GSD models corresponding to continuous functions because it was designed to process categorical data (the order in which one places grain-size classes does not influence the modeling results). Non-parametric decomposition regards GSDs as spectra which record a combination of some initial state modified by various processes, notably mixing and selective transport. This method of decomposition needs an array of GSDs (it cannot operate on a single GSD), as EMMA has been designed to provide the simplest possible explanation of the observed variation among a set of compositions in terms of (un)mixing. EMMA exploits the covariance structure of grain-size classes across a series of GSDs which contains information on the mixing structure of the data set. End-member GSDs determined with EMMA are thought to represent an assemblage of grains referred to as “dynamic populations” that are likely to occur together because they respond in a similar way to the dynamics of sediment production and dispersal within the system [Weltje and Prins, 2003, 2007].

Not surprisingly, the non-parametric decompositions of loess GSDs presented by Vriend and Prins [2005], Prins et al. [2007] and Vriend et al. (submitted manuscript, 2007) resulted in very different end-member GSDs compared with the parametric decomposition methods described above. The mixing models presented in these studies have three end-members which are characterized by unimodal, fine-skewed GSDs with a pronounced tail of fine silt and clay particles. Pye [1995] already indicated that care must be taken in interpreting the clay component of loess sediments, since a significant proportion of the finest grains are transported and deposited as silt- or sand-size aggregates held together by electrostatic forces, salts, or organic matter. Hence it is likely that the fine silt and clay particles recorded in the tail of the fine-skewed loess GSDs (and modeled loess components) have been originally present partly as aggregates and partly as individual particles, some of which adhered to the surfaces of larger grains. These aggregates have been disintegrated during pre-treatment of the loess samples and during subsequent grain-size analysis. The observations made by Pye [1995] clearly indicate that care should also be taken when loess GSDs are decomposed by a parametric curve-fitting method.

4.2. Interpretation of Modeled Loess Components

Prins et al. [2007] compared their modeled end-members with modern dust samples (data taken from Sun et al. [2003]) in terms of their grain-size distributions and flux rates. This comparison strongly suggests that the modeled end-members, with their characteristic tails of fine particles, may indeed be regarded as primary, unaltered loess components. The clayey loess component (EM-3) appears to be very similar to the fine dust component supplied over the entire loess region, partly during major dust outbreaks in spring and early summer, but mainly as part of a background supply system active throughout the year. The silty loess component (EM-2) was found to be very similar to the dominant dust fraction supplied over the proximal parts of the Loess Plateau during major dust outbreaks in spring and early summer. No modern analogue was found for the sandy loess component (EM-1). However, from the composition of EM-1 and its temporal and spatial distribution pattern in the loess deposits it was concluded that this component reflects the dust fraction supplied during very strong dust outbreaks in the glacial periods and that it might have been partly transported in saltation, rather than in suspension as is the case for end-members EM-2 and EM-3. The spatial distribution patterns of the sandy loess component in the full glacial loess units L1-1 and L2-1, reflected by the contoured [EM1:(EM1 + EM2)] ratio plots presented here, indicate that it occurs in significant volumes only in a narrow zone on the northern Loess Plateau (ZJC, HX and YN sections), just south of the present-day desert margin of the Mu Us and Tengger Desert, and within the Huang Shui river valley on the Tibetan Plateau (TXD and LD sections). The proximity of the sandy deserts and the Huang Shui river valley, the two likely source areas of the sandy loess component (EM-1) found in the Loess Plateau and Tibetan Plateau sections, respectively, supports the idea that the transport mode of this component has been saltation, or more likely, a combination of saltation and short-term suspension.

4.3. Spatial and Temporal Distribution Patterns of the Modeled Loess Components

The grain-size records presented in this study are obtained from the loess region of the NE Tibetan Plateau and the Loess Plateau. The similarities between the mixing models presented by Prins et al. [2007] and Vriend et al. (submitted manuscript, 2007), which are based on two subsets from the series of loess grain-size records, indicate that the “average” mixing model presented here might be regarded as representative for the vast loess region in northern China. Despite the highly variable grain-size characteristics of the studied loess and paleosol sediments, both in space and through geological time, the grain-size distributions are well described by this relatively simple mixing model. The mixing model describes the observed spatiotemporal grain-size variations by variations in the mixing coefficients of three loess components (end-members). The loess deposited at the proximal sites (ZJC, HX, YN) are mixtures of predominant the sandy and silty loess components (EM-1 and EM-2), whereas the intermediate (XF, LC) and especially the distal sites (XY, WB, DJ) are dominated by binary mixing of the silty and clayey loess components (EM-2 and EM-3). The overall spatial trends in end-member contributions thus clearly mirrors the transition from sandy loess to clayey loess across the Loess Plateau [e.g., Liu, 1985; Nugteren and Vandenberghe, 2004; Yang and Ding, 2004].

The unmixing results presented by Prins et al. [2007] and Vriend et al. (submitted manuscript, 2007) in conjunction with loess accumulation rate estimates revealed that two contrasting dust supply patterns were active over the loess region of northern China during the last glacial-interglacial cycle (EM-1 and EM-2 versus EM-3). The fractionated dust flux results presented here support the existence of two dust-supply patterns as very similar results have now been obtained also for three additional loess-paleosol sequences (XF, WB, DJ) and for the penultimate glacial-interglacial cycle (recorded in HX, XF, XY, YN and LC sections). An episodic, highly variable sediment input pattern, dominant during glacial periods throughout the region and noticeable during interglacial periods only over the northern Loess Plateau, is reflected in the admixture of the silty and the sandy loess components (EM-1, EM-2). The distribution patterns of these two loess components, which mirror the spatial thinning and fining trends of the loess deposits, and to a lesser extent of the intercalated paleosols [e.g., Liu, 1985; Pye and Zhou, 1989; Lu and Sun, 2000; Nugteren and Vandenberghe, 2004; Yang and Ding, 2004; Ding et al., 2005], strongly suggests that a northwesterly wind system has been the dominant supplier. In northern China, prevailing near-surface winter monsoon winds from the northwest, generated by the Siberian anti-cyclone, play a crucial role in transporting dust from the desert areas of Mongolia and Gansu toward the Loess Plateau at present-day [e.g., Derbyshire et al., 1998; Sun et al., 2003; Ta et al., 2004]. End-members EM-1 and EM-2 thus represent the dust components which are transported by the northwesterly winter monsoon during major dust outbreaks in spring and early summer.

The coarse-grained loess components represent the main part of the dust supplied by the low-level winter monsoon during major dust outbreaks in spring and early summer. The proportional contribution of the coarse dust fraction to the loess sediment appears to be a first-order approximation of the flux rate of event dust, as the flux rate of the fine “background dust” fraction is approximately constant. The grain size of the coarse dust fraction is thought to be mainly a function of the strength of the winter monsoon circulation system and of the transport distance, i.e., the distance to the eolian dust source area. Ding et al. [1999, 2005] proposed to use the sand fraction content (>63 μm) of loess deposits as a proxy-indicator of the location of the southern margin of the sandy desert north of the Loess Plateau, i.e., an indicator of source-area proximity. The end-member modeling results suggest that this is largely correct, as EM-1 consist of ∼46% sand (>63 μm), and therefore is the dominant “sand carrier.” However, the results also indicate that a substantial part of the fine-sand grains (EM-2 consists of 16% “sand”) are transported in suspension, which may occur over a distance of several hundred kilometers. The EM distribution patterns indicate that the [EM-1:(EM-1 + EM-2)] ratio can be used on the Loess Plateau as a potentially more precise proxy-indicator of the proximity of the sandy-desert margin, i.e., can be used to reconstruct changes in the location of the boundary (“transitional zone”) between source (desert) and sink (loess deposits). Ding et al. [2005] suggested that the migration of the desert margin is essentially controlled by the amount of summer-monsoon precipitation in the desert. However, it should be realized that variations in winter-monsoon strength might also play a role in the variable input of the sandy loess component across the Loess Plateau.

According to Prins et al. [2007] and Vriend et al. (submitted manuscript, 2007), the relatively constant and low flux of the clayey loess component (EM-3) represents a background sedimentation pattern which has been dominant during interglacial periods, especially over the central and southern parts of the Loess Plateau. As mentioned earlier, the clayey loess component (EM-3) resembles the fine dust fraction which is supplied over the entire loess region in modern times, partly during major dust outbreaks in spring and early summer, but mainly as part of a background supply system active throughout the year. The “background character” of EM-3 is clearly seen in the relatively uniform flux-distribution patterns (Figure 11) reconstructed for the interstadial (e.g., L1-2) and interglacial periods (e.g., S1). These observations suggest that non-dust storm processes may have been the dominant supplier during these time periods [Zhang et al., 1999; Prins et al., 2007]. Ding et al. [2000] noticed that the Tertiary fine-grained red clay deposits of the Chinese Loess Plateau, which underlie the Quaternary loess sequences, do not show a southward decrease in grain size along north–south transects. The relatively uniform sedimentation pattern of the red clay let them suggest that the fine dust may have been transported and deposited by westerly air streams. Similarly, the uniform flux pattern recorded in the interstadials/interglacials MIS 3 and 5 suggest that the high-level subtropical jet stream (westerly winds) might, at least partly, be responsible for the input of the fine-grained loess component.

The “background character” of EM-3 is less clear during the glacial periods (e.g., L1-1 and L1-3), where a systematic north-to-south increase in the flux of the clayey loess component is observed. This suggests that during glacial periods a significant portion of this component has been supplied by major dust outbreaks, which dominantly deposited the EM-1 and EM-2 components on the northern and central parts of the Loess Plateau, and the finer-grained EM-3 component to the southern part of the loess region (thereby largely by-passing the northern part of the Loess Plateau). However, the “exceptional high” flux rates of EM-3 on the southern part of the Loess Plateau may have been favored by some additional geological factors which are explained below.

A factor which may have enhanced the deposition of the clayey loess component is the effect of the Qin Ling and Luliang Shan mountains, acting as a topographic barrier enhancing loess accumulation on the windward, northward side of the mountain range [Pye, 1995]. Moreover, local dust sources may also have contributed in the increased dust fluxes recorded at the southern sites. Exposed fluvial sediments in the floodplains of the Wei He and its tributaries may have significantly enhanced dust flux rates at sites located proximal to these river systems, like for instance the WB site which is located on a fluvial terrace of the Wei He [Nugteren and Vandenberghe, 2004]. Exceptional high flux rates at this site of the silty loess component during MIS 2 (L1-1), and to a lesser degree during MIS 3 (L1-2), support this hypothesis. Additional sediment supply from local dust sources may also partly explain the enhanced flux rates of the clayey component during MIS 2 and 4 (L1-1 and L1-3, respectively) at the southern sites (XY, WB, DJ). Finally, deposition of the fine dust particles may have been enhanced by a dense vegetation cover on the southern plateau which acted as an effective dust trap. The transition from a sparse (or even lack of) vegetation cover in the north toward a denser cover in the south is the result of the northwest-to-southeast transition from (semi-) arid to more humid conditions. At this stage, however, it is very difficult to quantify the relative importance of these factors. Independent dust provenance data of the fine loess fraction are needed to determine the exact origin and further underpin the paleoclimatic significance of the clayey loess component.

Acknowledgments

Govert Nugteren, Jef Vandenberghe, and Martin Konert are kindly thanked for providing part of the grain-size data, and Michel Boos and Eveline Vermeer are thanked for grain-size analysis of the Duanjiapo section. Zheng Hongbo, Lu Huayu, Kay Beets, Simon Troelstra, Xianyan Wang, Jan-Pieter Buylaert, Eveline Vermeer, Michel Boos, and Douwe van Leverink are thanked for their support and company in the field. Financial support is provided by the Royal Netherlands Academy of Arts and Sciences (grant number 04-PSA-E-02) and the Ministry of Science and Technology of China (grant number 2004CB720506) as part of the Programme Strategic Scientific Alliances between China and the Netherlands.

Ancillary