The Chinese Loess Plateau (CLP) has the best preserved and most continuous sequence of eolian deposits in the world (Liu, 1985) (Figure 1). A number of studies of Chinese loess have established reliable chronologies, described their pedogenesis, and examined relationships to the uplift of the Tibetan Plateau (e.g., Ding et al., 1994; Porter and An, 1995). Dust flux and grain size have been shown to reflect winter monsoon intensity; and a variety of chemical indicators, as well as magnetic susceptibility, have been shown to be related to weathering and summer monsoon intensity. Such studies have revealed variations in the ancient monsoon in northern China since at least 8 Ma (An et al., 2001; An, 2014). Previous research on climate change in the CLP has focused on monsoon intensity and mechanisms of cyclical climate change (Ding et al., 1994). However, methods of determining paleowind directions in different portions of the CLP remain scarce and controversial.

Heller et al. (1987) first introduced the anisotropy of magnetic susceptibility (AMS) into loess studies. They verified that the reliability of paleomagnetic remanence is based on the consistency of the magnetic fabric in the Luochuan section and the relatively stable sedimentation rates observed within the loess-paleosol sequence. With improvements in the accuracy of AMS measurements, research on magnetic fabrics has gradually begun to compare the dynamic index with monsoon direction; that is, establish a linkage between the azimuth of the maximum axis of magnetic susceptibility and wind direction (Thistlewood and Sun, 1991). By taking AMS measurements from extant mineral assumblages and using statistical and kinetic analyses, ancient wind directions and intensities can by determined (e.g., Lagroix and Banerjee, 2002; Zhu et al., 2004; Zhang et al., 2010).

Based on mechanistic models inferred from AMS orientations, several recent studies have determined paleowind variations in the CLP. For example, Zhu et al. (2004) estimated dominant wind directions by considering magnetic lineations in loess (represented by maximum axes of AMS), whose orientations were observed to be clustered in different orientations during warm and cold periods. Zhang et al. (2010) proposed that the summer monsoon could facilitate the consolidation of the eolian particles that preserve AMS orientations. Ge et al. (2014) suggested that spatial variations in paleowind direction can be explained by regional surface winds. Other workers have reached much different conclusions. Zhu et al. (2007) and Liu and Sun (2012) pointed out that the model of eolian consolidation under such weak dynamics could not record dynamic wind directions. In most cases however, researchers determined orientations based on their measurements alone, without reference to dynamics in the horizontal surface (Zhu et al., 2007; Liu and Sun, 2012; Ge et al., 2014; Peng et al., 2015; Xie et al., 2016). We attempted to test further whether paleowind directions could be recorded in Chinese loess, and to discern the underlying mechanisms that govern the paleowind signal.

For all sediments, each mineral grain (ferromagnetic, paramagnetic, and diamagnetic) contributes to the total (bulk) magnetic susceptibility (MS). Several studies have already shown that precipitation is the main factor causing differences in total bulk k (MS) during the formation of loess and paleosols (i.e., Maher and Thompson, 1991; Maher and Thompson, 1992; Maher and Possolo, 2013). Precipitation also plays a significant role in consolidating particles and affects AMS orientations in different loess and paleosol units (Zhu et al., 2004; Zhang et al., 2010). For this study, we chose three loess sections—Lingyuan (LY), Xunyi (XY), and the classic Luochuan (LC) section—in the southern and central CLP, where stratigraphy and paleomagnetic results have been well-studied (e.g., Ding et al., 1994; Porter and An, 1995; Xue et al., 2003)—and tested whether loess samples that formed in a humid environment retain a paleowind signal. The chosen sections all lie within areas with more than 600 mm present-day annual precipitation, and are among the wettest sections we could find in the CLP. Yang et al. (2015) studied the grain size and distribution pattern of organic carbon in the CLP from the last glaciation to the present, and proposed that the northwestward migration of the monsoon was closely related to the position of the rainfall belt during the last glacial period. We note that the present day 600 mm precipitation zone is higher than rainfall levels in the last glacial (L₁) loess. To assist our selection of CLP sites, we also reviewed and compared previous records to find those that satisfy our requirements for AMS orientations and from the same time interval (i.e., Zhang et al., 2010).

The study sections LY, XY, and LC, were chosen from the southern and central CLP (Figure 1), and receive present-day precipitation of more than 600 mm per year. We initially planed to collect the last glacial (top loess layer L₁) and last interglacial (paleosol layer S₁) from each section. After removing 1 m thick coverage, this was accomplished for the XY (11 m collected) and LC (10 m collected) sections, however due to local weathering and erosion, we were limited to a portion of L₁ and no S₁ for the LY section (5 m collected).

Porter and An (1995) and Porter (2001) demonstrated that L₁ and S₁ correspond to marine isotope stage (MIS) 2–5, spanning 130–10k years ago. L₁ consists of two primary loess units (L₁LL₁ and L₁LL₂) and one underdeveloped paleosol (L₁SS). All stratigraphic subdivisions of L₁ correlate with marine oxygen-isotope stages observed in the LC section (Porter, 2001). Only L₁SS (partly) and L₁LL₂ were recovered from the XY section and subdivisions in the LY section were absent, possibly because rainfall homogenized the weak paleosol. Oriented samples were taken at a spacing of 2–5 cm in three sections and cut into 2 cm thick cubes. The AMS of each sample was measured using a KLY-4S KappaBridge coupled with an automated sample handling system. Each sample was rotated through three orthogonal planes, and the susceptibility ellipsoid was calculated by the least-squares method. Lineations (L), foliation (F), degree of anisotropy (P), and orientations of the three principle axes were obtained with Anisoft software using the statistical method of Jelínek and Kropáček (1978). All samples were also weighed to normalize mass MS results and facilitate comparisons among sections.

AMS tensors are defined by maximum (K_max), intermediate (K_int), and minimum (K_min) principal susceptibility axes. Following an established technique (e.g., Lagroix and Banerjee 2004b; Zhu et al., 2004; Zhang et al., 2010), the area of uncertainty around the orientation of the principal axes is quantified by 95% confidence ellipses that have semi-axes, denoted herein by epsilon ε, oriented parallel to the orthogonal planes defined by the three principal axes. Here we used ε₂₃ and ε₁₂, which represent the 95% confidence interval for K_min or K_int in the plane joining K_min and K_int and the uncertainty of K_max or K_int in the plane joining K_max and K_int, respectively. Figure 2 demonstrates the relationship between these AMS parameters. An inverse relationship between L (F) and ε₁₂ (ε₂₃) was observed (Figures 2B,C), demonstrating that L-(F) decreases as ε₁₂ (ε₂₃) increases due to the increasing importance of random measurement errors for K_max (K_min) in a lineation (foliation) plane with weak lineations (foliations) (Zhang et al., 2010). For the majority of samples, ε₁₂ < 22.5°, with a statistically significant level (81.7% for Xunyi, 93.8% for Luochuan, 90.6% for Lingyuan), yielding a confidence ratio of 1.0 for maximum and intermediate susceptibility axes in the foliation plane (Lagroix and Banerjee 2004b).

K_min axis orientations are distributed along the vertical axis of a stereographic projection. The mean I-K_min was equal to 88.8° in LY, 87.8° in XY, and 88.4° in LC. Table 1 compares I-K_min in sections in the northern and western CLP (Zhang et al., 2010). The dip in K_min axes displays a noticeable steep upright among these sections in the CLP. I-K_min showed only 1°–2° deviation from the vertical plane in all three sections which means there are no clearly preferred tilted K_min axes directions (Tarling and Hrouda, 1993; Zhang et al., 2010).

The iron oxide magnetite (and maghemite), especially its pseudo-single domain particles are the main carriers of the AMS signal in loess-paleosol sequences (e.g., Zhu et al., 2004; Zhang et al., 2010; Ge et al., 2014). Spherical projection of K_max and K_min for three sections were plotted using the Stereonet3D program in Figure 3 (Cardozo and Allmendinger, 2013). The magnetic minerals in loess show strong foliation and weak lineation (Figures 2A, 3) because gravitational forces are dominant and cause a strong oblate fabric within the bedding plane (Rees and Woodall, 1975; Tarling and Hrouda, 1993; Zhang et al., 2010; Gong et al., 2015).

The variation of AMS parameters vs. depth in the LY section are shown in Figure 4. The highest MS values in L₁ observed among all of the sections was in LY, and fell between 1 × 10⁻⁶ and 3 × 10⁻⁶ m³/kg. Principal axis eigenvalues show clear bi-axial oblate distributions (K_max–K_int > K_min) throughout every section as has been seen in many other loess sections (e.g., Lagroix and Banerjee, 2004a; Lagroix and Banerjee, 2004b). The sections frequently show fluctuations in the uppermost 1.5 m, which could be affected by the agricultural activity. For the XY and LC sections, we obtained AMS records from both the loess (L₁) and paleosol (S₁) layers (Figures 5, 6). These show a trend of increased oblate distributions from top to bottom (a decreased eigenvalue τ_min indicates that the tensor in the vertical direction has been flattened), suggesting that increased compaction with depth influenced both the upper loess layer (L₁) and upper paleosol layer (S₁).

Following the division and subdivision of loess units, four distinct stages of MS variability are recognized in our sections (especially for LC and XY) (Figure 7A). Due to higher precipitation in the past, we find higher MS in our sections than others (e.g., Zhang et al., 2010) except in stage three of the Luochuan section, which has lower MS. We observe that sections with average higher MS often have lower AMS parameters (P, F, and L) which is also seen in the results given in Table 1, where we report mean P, F, and L values from our sites, along with several other sites from the literature (BCY, XF, and YC). As an example, the BCY section has the lowest observed MS among all of the sections, and it has the highest AMS parameter values (P, F, and L). It should also be noted that in order to compare all the sections with similar depth scales, we use only the data of L₁ in Figure 7.

Zhang et al. (2010) used the statistical azimuth of K_max (D-K_max) and K_min (D-K_min) to infer a pattern of summer paleomonsoon routes. Under relatively weak dynamic conditions, D-K_max indicates the direction of the source of the current/paleowind while under relatively strong current/paleowind, a component of D-K_max turns toward the perpendicular of the horizontal dynamic/flow direction (Tarling and Hrouda, 1993; Zhang et al., 2010). D-K_min is always tilted in the direction of the current/paleowind under weak or relatively strong current/paleowind (Tarling and Hrouda, 1993). The oblate AMS ellipsoid of grain distributions, gives a more reliable K_min orientation than K_max. In such cases, changes in I-K_min correspond to hydrodynamics, and can be more than 10° for a relatively strong current (Tarling and Hrouda, 1993), 5°–7° for a weak paleowind in the northern and western CLP (e.g., Zhang et al., 2010), and less than 3° for quiet conditions. Figure 3 shows that quiet model conditions were observed in all three sections. D-K_max was randomly distributed in all directions around the edges of the steronet (Figure 3A) and orientations of K_min were perpendicular to the horizontal plane (Figure 3B). Considering the eolian origin of the sediment grains, we expect that pedogenesis and neoformed magnetic minerals may produce chaotic AMS values. For example, rapid growth of summer vegetation and consequent plant root development can interfere with the alignment of detrital particles and obscure the paleowind AMS signal (Zhang et al., 2010). In this manner, relatively high rainfall may have influenced ellipsoid orientations at the LY site (Figure 3), where rainfall was relatively high. This is consistent with the steep I-K_max gradient we observe in the LY section, suggesting deep rainwater penetration. We summarize modern precipitation distribution with isopotential density maps (Figure 1), and ground surface wind vectors (Figure 9). The present-day 600 mm precipitation zone experiences more rainfall than during the last glacial, when the (L₁) loess was deposited (Yang et al., 2015). Although the distribution pattern might be similar, we expect that quantitative differences between the last glacial period and the present exist. Future paleowind investigations on the CLP would do well to choose a low rainfall area for study, to avoid the effects of reworked/water-lain sediments. We find that the orientation of K_min is be the most noteworthy feature, as it represents the placement of the dominant AMS parameter—foliation, in the loess.

For a single section, there is no obvious relation between MS variations and AMS parameters. In the LY section for example, some extremely high F, P or L values are randomly distributed, and are not concentrated on horizons with higher or lower MS (Figure 4). For the XY and LC sections (Figures 5, 6), we indicate with pink horizontal bars those with high P and F (and L in most cases) values. These stem from high eigenvalues of τ-max (and τ-int in most cases) and high MS values. Aside from these extreme cases, we are unable to establish a relationship between AMS fluctuations and MS values. In the LC section for example, we have observed two large MS peaks in S₁ with high AMS variability. However, the rest of S₁ has lower AMS values than the average loess horizon values. For the XY section, there is no distinct difference between the AMS values of the loess and paleosol. In such cases we infer that the relations between AMS parameters and MS values cannot be attributed to a single factor (for example, precipitation or paleowind intensity). On the CLP, during a strong monsoon rainfall season, water could remain longer in pores within the loess layer. Such extended saturated conditions could disturb the AMS signal and distort inferred paleowind directions. All of our sections (LY, XY, and LC) experienced relatively high precipitation as compared with sections from the northwest CLP (i.e., Zhang et al., 2010), and in most cases, we observe low values of L, F, and P (Table 1). Therefore, although precipitation might increase MS, it does not increase L, F, and P.

In Figure 8, we show K_max and K_min orientation distributions for the four stages seen in the sections. We find large K_min deviations from the vertical plane that could be random, and the spread of D-K_max values do not show a clear trend with which to evaluate the paleowind direction without ambiguity. We also found clustered major K_min in the projection (shadow ellipsoid in Figure 8), mainly in the NW-SE direction. This could suggest that the major horizontal dynamic variations correspond to two flow directions (NW and SE), with implied ground surface wind directions similar to the present—SE (summer) and NW (winter) (Figure 9). The only exception is the clustered K_min in the LC section, which shows a slight clockwise rotation from W-E (stage 4, S₁) to N-S (stage 1, L₁LL₁) in the stereo-graphic projection (Figure 8A). Under the present major winter wind patterns, winds can vary from NW, to N, to NE, and even to E, with a slight shift to the south (Figure 9). Such scenarios imply that paleowinds recorded in the LC section may have varied since the last interglacial. These results can also explain why a previous study of the LC section found no preferred paleowind directions (Liu and Sun, 2012).

Magnetic fabrics in eolian sediments can be measured by AMS to infer paleowind directions. However, high precipitation in the southern and central part of the CLP impart variability in the inferred paleowinds. We suggest that studies of paleowind would be more consistent in the northwest CLP, where there is less rainfall. In particular, the orientation of K_min may be used as a dual-level indicator of horizontal dynamics and direction. It should first be determined if there is an absence of horizontal force or tilt response, then the flow direction should be decided by studying the statistical trend of K_min with time. In future studies, we recommend that statistical parameters of AMS orientations, especially K_min, should be used to interpret whether the paleowind can be recorded.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

RZ is the corresponding author and he collected the AMS data. FX and RZ contributed equally to this work. FX and RZ performed the experiments, carried out data analysis, and wrote the manuscript.

This work was co-supported by the National Natural Science Foundation of China (41772027, 41930321, 41972035, and 41950410574) and State Key Laboratory of Loess and Quaternary Geology (SKLLQG1732).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer (XW) declared a past co-authorship with one of the authors (FX) to the handling editor.