Implications of the loess record for Holocene climate and human settlement in Heye Catchment, Jiuzhaigou, eastern Tibetan Plateau, Sichuan, China

Abstract We examine the Holocene loess record in the Heye Catchment on the margins of the Tibetan Plateau (TP) and China Loess Plateau (CLP) to determine: the region to which the Heye Catchment climate is more similar; temporal change in wind strength; and modification of the loess record by mass wasting and human activity. Luminescence and radiocarbon dating demonstrate loess deposited in two periods: >11–8.6 ka and <5.1 ka. The 8.6–5.1 ka depositional hiatus, which coincides with the Mid-Holocene Climatic Optimum, is more similar to the loess deposition cessation in the TP than to the loess deposition deceleration in the CLP. Grain-size analysis suggests the Heye loess is a mixture of at least three different grain-size distributions and that it may derive from multiple sources. A greater proportion of coarse sediments in the older loess may indicate stronger winds compared with the more recent depositional period. Gravel incorporated into younger loess most likely comes from bedrock exposed in slump scarps. Human occupation of the catchment, for which the earliest evidence is 3.4 ka, postdates the onset of slumping; thus the slumps may have created a livable environment for humans.


INTRODUCTION
Situated on the margins of both the Tibetan Plateau (TP) and the China Loess Plateau (CLP), Holocene loess deposits in Jiuzhaigou, northern Sichuan ( Fig. 1), provide an opportunity to add to the understanding of the spatial variation in Holocene climate and loess provenance in western China. In addition, the region's oldest archaeological site (d'Alpoim Guedes et al., 2015) is in Jiuzhaigou loess deposits on hillslopes subject to landsliding and co-seismic faulting. Previous investigations of loess in Jiuzhaigou (Wen et al., 2014(Wen et al., , 2017 left unanswered questions about the depositional history and provenance of the loess and also the role of hillslope processes and seismic activity in modifying the loess and hillslope topography. Thus, in addition to enhancing understanding of Holocene climate and loess provenance in western China, analysis of the Jiuzhaigou loess record can also provide insight into causal connections between climate, loess deposition, hillslope deformation by landslides and co-seismic faulting, and human inhabitance. Analysis of the age of loess and loess-paleosol sequences in the CLP and TP has been widely used to examine regional paleoclimate (e.g., Derbyshire, 1982;An et al., 1991;Lehmkuhl et al., 2000Lehmkuhl et al., , 2014Porter, 2001;Lehmkuhl and Owen, 2005;Sun et al., 2007;Pullen et al., 2011;Stauch et al., 2012), because loess generally deposits during cool and dry periods, whereas paleosols generally form on loess deposits during intervening wet and warm periods (Porter, 2001). The grain size of loess may also provide information on wind speed (e.g., An et al., 1991;Xiao et al., 1995;Jiang and Ding, 2010), aridity and frequency of storms , loess sources and paleowind directions (Sun et al., 2004;Prins et al., 2009;Yang et al., 2010a), and the relative importance of low-altitude and high-altitude eolian transport of loess (Sun et al., 2004;Nottebaum et al., 2015). While loess-paleosol sequences up to 400 m thick in the CLP provide valuable paleoclimate data for at least the last 2.4 Ma (Porter, 2001), there has been less study of either Holocene loess in China (Maher et al., 2003;Kuster et al., 2006) or loess in China outside the CLP (Yang et al., 2010a(Yang et al., , 2010bLehmkuhl et al., 2014).
In western China, Holocene loess deposits and their associated soils, which span the TP, western Xinjiang Province, the CLP, and south to the Sichuan basin, and climate proxies from cave formations and lake and bog cores generally record a cool and dry climate in the Early and Late Holocene, with an intervening warmer and wetter period during the Mid-Holocene Climatic Optimum (MHCO; Maher et al., 2003;Kuster et al., 2006;Nottebaum et al., 2015), but the loess record also shows within-region variation in the depositional response to climate . For example, in the CLP, loess deposited continuously in both cool, dry periods and warm, wet periods, but deposition slowed and soil formation increased during wetter and warmer conditions (Huang et al., 2003). In other locations, loess accumulated sporadically: in some places, loess accumulated only during cool and dry periods, while in other places, loess accumulated in wet and warm periods, when increased vegetation enhanced the efficacy of dust trapping, with a hiatus in deposition during cool and dry periods (Lehmkuhl et al., 2000Kuster et al., 2006;Sun et al., 2007).
Loess tends to become finer the farther it travels, and thus fines downwind (Liu and Ding, 1998;Nugteren and Vandenberghe, 2004;Yang et al., 2010a). West-to-east fining of loess on the CLP Porter, 2001) implies a source from the west; various studies suggest a source or mix of sources in the northern TP, the Qaidam basin, the Gobi Desert, and/or the Taklamakan Desert (Derbyshire et al., 1998;Sun et al., 2000;Maher et al., 2009;Stevens et al., 2010;Wu et al., 2010;Pullen et al., 2011), and some argue that loess sources are different during glacial and interglacial times, with deserts dominating during interglacial periods and the TP and Qaidam basin dominating during glacial periods (Pullen et al., 2011;Fig. 2). Similarly, downwind fining has been used to determine that the loess in the Sichuan basin of northern Sichuan comes from the eastern TP rather than the CLP (Yang et al., 2010a;Wen et al., 2014). Jiuzhaigou is located on the margins of the CLP, TP, and Sichuan basin, and prior studies suggest that the loess is sourced from the eastern TP (e.g., Wen et al., 2014). However, it is unknown if loess source or wind speed changed during the period of loess deposition.
As early as 1977, scientists identified three dominant modes of airborne dust and interpreted these modes as having different transportation mechanisms and source materials (Patterson and Gillette, 1977). Although descriptive statistics such as mean, standard deviation, and skewness (Folk, 1980) can be calculated for multimodal samples, these statistical metrics are not particularly meaningful for mixed materials . Prior work determining components of mixed grain-size samples relies on end-member modeling analysis (e.g., Dietze et al., 2012) or single-specimen unmixing (Sun et al., 2004;Qin et al., 2005;Xiao et al., 2009Xiao et al., , 2013Pye, 2015), typically using either Weibull (Sun et al., 2004) or lognormal (Qin et al., 2005;Xiao et al., 2009;Pye, 2015) distributions in the modeling. For multimodal samples, different peaks are attributed to different dust sources (Vandenberghe, 2013). Most studies identify at least two fractions: a coarser fraction interpreted to be from the local area and transported up to a few kilometers during dust storms and a finer fraction interpreted to be non-dust storm dust sourced from locations more than a few kilometers away (Patterson and Gillette, 1977;Sun et al., 2004;Vandenberghe, 2013;Pye, 2015). Some studies identify multiple intermediate modes that are interpreted as either a mixture of other modes or additional transport distances (Vandenberghe, 2013). Occasionally submicron-sized dust is observed, and this is interpreted as either anthropogenic particles, weathering materials, or long-range transported dust (Patterson and Gillette, 1977;Zhang et al., 1994). CLP loess has two modes that are interpreted as non-dust storm dust transported long distances on the high westerly winds and more locally sourced material that is coarser and transported during dust storms (Sun et al., 1998(Sun et al., , 2004. The northeastern TP loess has two to three modes that are fine sandy, coarse silt, and medium silt in size (Vriend and Prins, 2005;Nottebaum et al., 2015), and the ratio of the medium silt to coarse silt modes may indicate the relative strength of local winds as opposed to high westerlies transporting finer particle sizes (Nottebaum et al., 2014). The grain size of loess in Jiuzhaigou has not been analyzed in this way, and thus the potential for different contributions from different locations beyond the eastern TP source area previously identified (Wen et al., 2014) has not been explored, possibly because accessible software to do these calculations only recently became available (Liu et al., 2021).
The loess record can also contribute to an understanding of the Holocene history of human inhabitance of western China. For example, the timing of loess deposition, and implied climatic fluctuations, can correspond to changes to human habitation in loess regions of western China d'Alpoim Guedes et al., 2015;Wen et al., 2017), because agriculturalists can preferentially inhabit areas of loess soils Schmidt et al., 2017). However, the widespread landslide hazards reported in loess regions (Dijkstra et al., 1994;Derbyshire, 2001;Wen et al., 2005;Zhang and Wang, 2007;Schmidt et al., 2017) also compromise safety . Conversely, human inhabitance can modify the loess record; for example, farming can inflate the rate of loess deposition (Roberts et al., Figure 2. Distribution of grain-size data from Jiuzhaigou (marked on map with bold name; data from Wen et al. [2014]) and previous studies on loess grain size in the region (Lei, 1998;Wang et al., 2003;Sheng, 2010;Yang et al., 2010a;Stauch et al., 2012;Lehmkuhl et al., 2014); size of point indicates reported mean grain size. Background elevation is from World Wildlife Fund (2008).
Here we analyze the loess record in Heye Catchment, Jiuzhaigou, Sichuan, to add to the understanding of how loess deposition varies temporally within western China in response Sketches of terrace risers are available in Supplementary Figs. S1-S10. to regional Holocene climate; determine the provenance of loess deposited in the Jiuzhaigou area; understand the mechanisms through which the loess record may have been altered during or following deposition; and better understand Holocene relations between loess and human inhabitance in northwestern Sichuan Province.

STUDY SITE
Our study site is located in Heye Catchment, Jiuzhaigou National Nature Reserve (33°13 ′ N, 103°55 ′ E), a World Heritage Site in the Minshan Mountains (Fig. 1). Mean annual temperature is 6°C and mean annual precipitation is 650 mm, with the majority of precipitation falling during the summer monsoon (Qiao et al., 2016). The elevation of the Heye Catchment ranges from 2200 m at its outlet to the main Jiuzhaigou valley to 4000 m on the catchment's southern boundary.
The catchment is primarily formed in limestone and dolomite with lesser amounts of phyllite and slate and is locally mantled by up to 10 m of loess (Guo et al., 2006). The region is tectonically active and was subject to shaking from both the 2008 Wenchuan earthquake and the 2017 Jiuzhaigou earthquake (Sun et al., 2018). The Heye and Longkang Faults are near, or cross through, Heye Catchment, but contradictory locations on available, published maps and/or the small scale of those maps (Zhang and Xu, 1993;Guo et al., 2006;Burchfiel and Chen, 2012) leaves their precise locations uncertain. The timing of previous movement on the faults is unknown (Zhang and Xu, 1993). Spatially extensive slump earthflows on east-, south-, and west-facing loess-covered slopes are characterized by terraces with 1-to 10-m-tall risers (Figs. 1 and 2, Supplementary Figs. S1-S10). One of these slump features, on which most of our observations are concentrated, consists of a series of slump blocks having near-horizontal surfaces, or terrace treads, and nearvertical failure planes, or terrace risers; has an area of 10.5 hectares; is covered in loess up to 10 m thick; and may have been destabilized by the Heye Fault (Zhang and Xu, 1993).
Deformed phyllite and slate are exposed at the base of some of these terrace risers. The exposed bedrock is overlain by either the older of two distinct loess units or by the younger of the two loess units. A red horizon caps the older loess in most exposures. The older loess sometimes includes mollusks, but not charcoal. At one exposure (HY49), measured levels of soil carbon and nitrogen in the red layer are similar to levels in the modern soil, indicating the red layer is likely a paleosol . Because this stratigraphy is similar throughout the Heye Catchment, and the red layers have the same appearance as the paleosol at HY49, we interpret red layers elsewhere in the catchment to be paleosols. A paleosol is visible in most, but not all, terrace scarps.
A younger loess overlies the paleosol or in some locations directly overlies the exposed bedrock. The younger loess consists of a silty matrix supporting angular, lithic clasts ranging in size from 2 mm particles to 1 m blocks. The paleosol and gravelly loess in some exposures contain archaeological artifacts, usually charcoal, and do not contain mollusks. In one terrace riser exposure (HY49), the paleosol is topped by loess instead of gravelly loess; this is the only location where the paleosol is covered by loess without gravel. All terrace risers contain outcrops of younger loess (typically gravelly loess), and more than half (∼60%) contain older loess.
Geologic hazards in Jiuzhaigou, and in Heye Catchment in particular, have been studied due to high levels of tourism and possible risk to visitors. Debris flows in the reserve, which reportedly have increased in frequency in recent years due to human activity (Zhang, 2019), are common. Slope stabilization structures for diverting flows away from tourist areas have been used to mitigate hazards, as in national parks elsewhere in China (Cui et al., 2003(Cui et al., , 2007. In our study area, Heye Catchment, debris-flow hazard management has focused on dams and channels to divert flows away from the relatively large village near the mouth of the valley (Cui et al., 2003). In addition to debris flows, the slump earthflow (Fig. 3A) could be a major hazard in the valley if it were to become more active. Some mass-wasting, including slump earthflows and debris flows, may be related to earthquakes in the region (Zhang and Xu, 1993), an interpretation supported by increases in landslide frequency after both the 2008 Wenchuan and 2017 Jiuzhaigou earthquakes (Han et al., 2018;Xie et al., 2018). Despite the fault running through or near Heye Catchment and long-recognized debris-flow hazards, a GIS-based landslide susceptibility analysis of Jiuzhaigou completed after the 2017 earthquake found that Heye Catchment is at low risk for landslides due to relatively low slope steepness as calculated using remotely sensed digital topographic data (Yi et al., 2019(Yi et al., , 2020. Terraces formed by slumping in the Heye Catchment contain the only known archaeological sites in Jiuzhaigou and the oldest site in the region, Ashaonao (d'Alpoim Guedes et al., 2015). Archaeological evidence from Ashaonao indicates that people occupied the region as early as 3.4 ka (d'Alpoim Guedes et al., 2015). Artifacts are concentrated in the paleosol and slightly above the paleosol in gravelly loess but are not found in the loess underlying the paleosol Urgenson et al., 2014;d'Alpoim Guedes et al., 2015;Figs. 3 and 4).
Previous studies leave open several questions about the loess. There is uncertainty about when the younger loess deposited, whether and how it may have been modified syn-or postdeposition, and what role landsliding and human occupation may have played in any modification. Prior work on the younger, gravelly loess used radiocarbon dates of bulk charcoal, grain-size analyses of the loess (i.e., excluding the coarse lithic component), and rare earth element analysis at the Ashaonao archaeological site to determine that the loess is likely derived from fine-grained glaciofluvial deposits in eastern Tibet during the Late Holocene and was deposited continuously over the last ∼5000 years (Wen et al., 2014(Wen et al., , 2017, but considers neither the older loess nor the source of gravel in the younger loess. Henck et al. (2010) interpret the older loess below the paleosol as primary loess deposition, which they date to the late Pleistocene/Early Holocene (11.1 ± 0.7 and 18 ± 1.4 ka), and younger gravelly loess above the paleosol as material remobilized by Late Holocene landslides, with the gravel as having been incorporated during landsliding. Thus, three questions remain about the Heye Catchment loess: (1) What is the source of gravel in the younger loess? (2) Is the younger loess remobilized or primary deposition? (3) When were the two loesses deposited, and what inferences about Holocene climate can they support? Answers would provide insight to Holocene climate in the Jiuzhaigou region; how human habitation may have influenced the character of loess deposits; and how the loess, and the possible role of landsliding in modifying the loess, may have influenced human inhabitance.

METHODS
We conducted fieldwork in the summers of 2007, 2008, 2010, 2011, and 2016. In summer 2008, an archaeological team excavated the Ashaonao archaeological site Urgenson et al., 2014;Harrell et al., 2017). In summer 2016, stratigraphy and detailed sections were drawn for seven terrace faces. In summer 2011, we described stratigraphy of 50 terrace scarps and collected charcoal samples for radiocarbon dating and sediment samples for grain-size analysis, thin sections, and optically stimulated luminescence (OSL) dating from 19 of those scarps. Twelve sampled scarps had loess exposed under a paleosol or gravelly loess; eight of these have <1.5 m of exposed loess. The remaining four scarps have between 3.5 and 5 m of exposed loess below a paleosol. While we collected samples from the entire section of the younger loess, we confined our geochronology sampling to the upper 2.8 m of the older loess and grain-size sampling to the upper 3.5 m of the older loess, which is locally up to 5 m thick. The one OSL sample collected deeper than 2.8 m in the older loess was damaged during shipment from China.
We used petrographic analysis to characterize mineralogy of the older loess and the younger, gravelly loess. We embedded in epoxy the sediment from seven samples from terraces HY49 and HY50 collected in the field in 2011, made thin sections from the epoxy-embedded sediment, and analyzed the polished thin sections using a petrographic microscope. To determine percentages of quartz, calcite, and mica, we point counted at least 475 grains from each sample and identified the grains as quartz, calcite, mica, or other.
We used radiocarbon dating of organic material and luminescence dating of sherds and sediments to determine the timing of loess deposition. Charcoal, grain, wood, or bone samples for radiocarbon dating (n = 32) were taken in 2011 and 2016 from anthropogenic layers, paleosols, and the younger loess unit and were analyzed at the University of California Irvine Keck Carbon Cycle Accelerator Mass Spectrometry Laboratory (2011 samples) Santos, 2004, 2007; and at the Oxford Radiocarbon Accelerator Unit (2016 samples) (Ramsey et al., 2004;Brock et al., 2010). After pretreatment with weak HCl, carbon dioxide was extracted and then graphitized on an iron catalyst. We used AMS to measure the graphitized carbon dioxide for 14 C/ 12 C ratios. In addition to new radiocarbon ages, we include analysis of 27 previously published radiocarbon ages d'Alpoim Guedes et al., 2015;Wen et al., 2017). We calibrated new and previously published radiocarbon ages to calendar years with CALIB v. 8.2 using the IntCal20 data set (Reimer et al., 2020;Stuiver et al., 2022). To facilitate comparison with luminescence ages, we report the probability weighted mean age in thousands of years (ka) before 1950 rather than a calendar date; error bars for these dates span the range reported by OxCal.
Sherds were collected in 2007 and bagged in the field along with small samples of surrounding sediment for radiation measurements. Sediment was sampled in 2011 in metal pipes, the ends of which were then stuffed with paper and sealed with dark tape. We dated sherds using thermoluminescence, infrared stimulated luminescence, and OSL at the University of Washington Luminescence Lab (see Fig. 4 for a photo of a representative sherd sample). We dated sediments using OSL on 4 mm aliquots of 63-90 μm quartz at the Desert Research Institute Luminescence Lab in Nevada using a modified single-aliquot regenerative dose approach (Murray and Wintle, 2000; see text in Supplementary Material and Supplementary Tables S1-S8 for more details). We also include previously reported luminescence ages for sediments in the Heye Catchment Feathers et al., 2012). The ages reported by Henck et al. (2010) were revised in Feathers et al. (2012). The latter paper used fine grains (1-8 μm) for measurement and tried to isolate the quartz signal both by chemical treatment and by application of pulsing methods. Both methods produced similar ages. For this paper, we used the ages from the chemical treatment, because that is the same procedure used by the Desert Research Institute (except on larger grain sizes) for the new ages we report. However, it would not make a significant difference if we used the pulsed ages. The chemical treatment was not applied to three samples from Feathers et al. (2012), so we report the pulsed ages for these. To facilitate comparison with radiocarbon data, all OSL ages are reported in thousands of years (ka) before 1950.
To determine the grain-size distributions of the loess and gravelly loess in Heye Catchment, we collected 50 samples of sediment from 17 terraces, 35 samples of loess below the paleosol and 15 of gravelly loess above the paleosol. After pretreating sediment with hydrogen peroxide to remove organics, we used a Mastersizer laser diffractometer at Whitman College, Walla Walla, Washington, USA, to quantify the grain-size distribution of the <2 mm fraction of samples.
Because all samples had multiple distinct modes, we used QGrain (Liu et al., 2021)-software that implements the methods described in Sun et al. (2004)-to separate grain-size distributions of mixed samples into component parts. QGrain implements a single-specimen unmixing (SSU) model that reports the mean, standard deviation, and fraction of the sample for a defined number of unimodal parametrically distributed component sources. This allows us to determine similarities and differences among samples without having to propose and measure grain-size distributions of possible source regions (Sun et al., 2004). Thus, while we are unable to identify specific source locations using the SSU model, we are able to determine whether different samples are mixtures of material from the same or different sources. QGrain allows the user to choose between normal and Weibull curve fits and to determine the number of components in the mixture. Prior work on CLP loess grain sizes suggests that Weibull distributions may be more appropriate for loess areas, because modern wind speeds have a Weibull distribution (Sun et al., 2004). However, all samples had good, statistically significant fits with normal distributions, and thus we used normal distributions rather than Weibull, which requires more parameters to be fit and therefore has the potential to overfit the data.

Mineralogy and petrography
In thin section, loess samples from both the older and younger loess units are composed of individual grains that grade from rounded to subangular. The younger, gravelly loess also contains aggregates of loess particles, lithics, and needle-fiber calcite (Fig. 5C). The older loess samples only contain individual rounded particles and no soil aggregates (Fig. 5D). The composition of older loess is >60% quartz, 8-15% mica, 8-13% calcite, and up to 24% other minerals. The composition of younger gravelly loess is ∼50% quartz, ∼20% metamorphic lithics, 7-10% mica (muscovite and biotite), 5-13% calcite, and up to 18% Figure 7. All ages as a function of depth below stratigraphic surface (i.e., terrace top for gravelly loess and either terrace top or paleosol for loess, depending on stratigraphy of the terrace). 2σ error bars are shown; some are smaller than the symbols used and not visible. Dates with ages >5.1 ka and samples taken from the paleosol were not used in the correlation analysis for date as a function of depth. Data shown includes previously published OSL and 14 C data for terraces in Heye Catchment Feathers et al., 2012;Wen et al., 2014Wen et al., , 2017d'Alpoim Guedes et al., 2015). Details of all ages shown are in Supplementary Tables S9-S12. **p < 0.01. Figure 6. Overview of stratigraphy, sample locations, and geochronology. (A) Longitudinal profile across the main study area with terraces labeled. HY7 is uphill of HY34, but profile is not available. T1 and T2 are not captured in the longitudinal profile, but approximate location is noted. (B) Stratigraphy of terraces and locations of samples shown. Selected dates are provided (in ka). Details on all data included are in Supplementary Tables S9 and S10. The Ashaonao excavation was at T6. Location of terraces is in Fig. 1. other minerals. Pedogenic aggregates are common in the younger, gravelly loess samples taken from all depths of the terrace scarps. Both lithics and aggregates are coarser than grains of primary minerals. Lithics consist of angular fragments of metamorphic rock that have lithologies consistent with having been derived from the bedrock exposed in road cuts and landslide scarps in the Heye Catchment. Younger gravelly loess from samples taken in or slightly above the paleosols have fragments of needle-fiber calcite, a distinctive texture composed of porous masses of acicular calcite crystals in subparallel orientations. In thin sections made of a sample of the paleosol (HY-C), needle-fiber calcite is found in voids that appear to be macropores formed by roots ( Fig. 5A and B).

Geochronology
Radiocarbon and luminescence ages of sediment and pottery sherds, including those from previous studies of Heye loess and terraces Feathers et al., 2012;d'Alpoim Guedes et al., 2015;Wen et al., 2017), fall into two distinct age groups: those from the younger loess are younger than 5.1 ka, and those from the older loess are older than 8 ka (Figs. 6 and 7, Supplementary Fig. S11, Supplementary Tables S9-S12). The paleosol, stratigraphically located between the older and younger loess, yielded dates that fall within the age ranges of both the older and younger loess units, reflecting that the paleosols consist of older loess parent material and younger organics, sediments, and artifacts incorporated over the period of paleosol formation. Thus, we group ages from the paleosol separately from those from the two loess units. The single anomalously old age of paleosol material is likely the age of the older loess parent material, while the younger ages are likely the age of charcoal and sediments incorporated into the older loess during the hiatus in loess deposition.
Ages of material sampled from young loess linearly increase with depth below the current terrace surface (Fig. 7) and indicate a deposition rate of 0.39 ± 0.05 m/ka. Four sherd dates fall within this trend, implying that the sherds were buried relatively quickly after the pottery was fired. In contrast to the younger loess, the large uncertainties of OSL ages of the older loess preclude defining an age-depth relationship; five samples have statistically indistinguishable ages from 11.1 to 8.58 ka (Fig. 7); the error-weighted mean deposition age of the older loess (excluding the outlier paleosol sample) is 9.6 ± 1.6 ka. The 11-8.6 ka age range for a 2.8 m section of loess could suggest relatively rapid loess deposition, but we cannot determine this with the data available. Because our older loess ages extend to no more than 2.8 m below the surface on deposits up to 5 m thick, we may only have ages for the most recent ∼50% of the period of loess deposition and cannot determine the date for onset of loess deposition.

Grain size
Analysis of grain-size data indicates there are multiple components of the Heye loess but that the number of components and relative dominance differs between the younger loess, the upper part of the older loess, and the lower part of the older loess (Figs. 8 and 9, Supplementary Fig. S12, Supplementary Tables S13-S15). The coarse (33-49 μm), medium (6.4-10.5 μm), and fine (0.58-0.71 μm) components are the first, second, and third most dominant components of the younger loess, upper part of the older loess, and lower part of the older loess.
However, the lower older loess has more coarse material and less medium and fine material than the upper older loess and younger loess. The younger loess and the upper older loess additionally have a very coarse (316-405 μm) component, and the younger loess also has a very fine (0.038-0.042 μm) component. The mean grain size and fraction of the sample contained in each component are consistent with depth for each of the stratigraphic units (younger loess, upper older loess, and lower older loess). However, the grain size and fraction of sample for a given component differ in the lower older loess compared with the younger loess and upper older loess (Fig. 9).

DISCUSSION
We characterized the petrography and grain size for an older and a younger Holocene loess from Heye Catchment, Jiuzhaigou, and used radiocarbon to age charcoal and luminescence to age loess and sherds. In this section, we use the timing of deposition to make inferences about the regional Holocene climate, use grainsize data to infer paleowind conditions, and propose an explanation for the gravel component of the younger loess unit in Heye Catchment. Finally, we explore the environmental-social linkages suggested by the timing of loess deposition, landsliding, and human settlement in the Heye Catchment.

Timing of loess deposition and regional climate implications
Prior studies restricted their attention to the younger of the two loess units (Wen et al., 2014) or provided a provisional estimate for the depositional period for an older loess Wen et al., 2017); dates in this study indicate that loess in Heye Catchment deposited in two distinct periods in the Holocene, >11-8.6 ka and 5.1-0.3 ka (Fig. 7), with a break in deposition and commensurate soil formation that corresponds to the MHCO. Because our samples of the older loess unit were restricted to the upper 2.8 m, we cannot define the onset of deposition. Prior studies also left unresolved whether, or how, the younger loess had been subject to syn-or postdepositional modification ; the linear age-depth relationship derived from the younger loess across 11 terraces in the study area suggests the younger loess represents primary loess deposition that started ∼5 ka. This conclusion is also supported by an exposure of young loess, above the paleosol, that does not contain gravel (HY49). Thin-section analysis of angular rock fragments contained within the loess show that these rocks have lithologies consistent with the local bedrock; later, we propose a mechanism by which these rock fragments could have been added to the depositing loess rather than being incorporated into remobilized loess.
The older loess (>11-8.6 ka) below the paleosol was deposited during a regional cool and dry period when loess was being deposited elsewhere in western China at least as early as 12 ka (Porter, 2001). Closer to Heye, pollen data from the Zoige area, 150 km northwest of Heye Catchment, indicate that the area alternated between woodland and grassland catchment between 10.3 and 7.3 ka (Zhao et al., 2011), suggesting intermittently colder conditions. In addition, several sites in the eastern TP record cool periods around 10 to 9.5 ka (Mischke and Zhang, 2010;Wang et al., 2010) and a temporal minimum in moisture on the TP around 9 ka (Ran and Feng, 2013) that corresponds to increased fire frequency (Sun et al., 2016;Zhao et al., 2017). Thus, loess in the Heye Catchment from >11-8.6 ka deposited during a regionally cool and dry period, as in the TP and the CLP. The timing of the cool and dry period on the CLP is no earlier than 12 ka, suggesting that we capture most of the period of Early Holocene deposition with the samples we measured.
There is a hiatus in the loess record in the Heye Catchment from 8.6-5.1 ka, more similar to the loess record in the TP, where loess deposition stops, than in the CLP, where loess deposition continues, albeit at a slower rate, through this period. The hiatus in loess deposition in the Heye Catchment corresponds to the MHCO, a warm period marked in western China by a slowing or hiatus in loess deposition; increased rates of soil formation, depending on the location Porter, 2001); higher lake levels (Zhou et al., 1991); and reduced fire frequency (Sun et al., 1998;Huang et al., 2006). There is regional variability in MHCO timing, with the onset of the MHCO in western China and the TP occurring between 10 and 7.5 ka and the end occurring between 5 and 4 ka Lehmkuhl et al., 2000;Feng et al., 2006;Wu et al., 2007;Zhu et al., 2009;Schütt et al., 2010;Bird et al., 2014). It is unclear whether the variability is systematic or due to different responses recorded in different climate proxies. Meta-analyses of regional data suggest that the MHCO is associated with an expansion of the Indian summer monsoon . The gradual expansion of the range of the Indian summer monsoon could well explain variations in timing of the MHCO around the region. Although the onset of the MHCO seems to vary somewhat, all of these core data suggest that the warm and wet period ended after 5 ka.
Closer to our study area, pollen and charcoal data from cores in two peat bogs 150 km northwest of Jiuzhaigou on the eastern edge of the TP indicate a warm and wet period from 10.8-5.7 ka (Hong et al., 2003;An et al., 2006;Zhao et al., 2011). These data suggest that the warm and wet period started as much as 2 ka earlier on the eastern TP than in Jiuzhaigou, but as with the regional data, it is possible that different climate proxies have different response times to the changing climate. Overall, the period of paleosol formation associated with a hiatus in loess deposition that we identify as dating from 8.6-5.1 ka suggest that climate conditions in the Heye Catchment may have been similar to those of the TP.
Resumption of loess deposition in the Heye Catchment corresponds to a post-MHCO cooling and drying regionally Feng et al., 2006) and locally (Hong et al., 2003;An et al., 2006;Zhao et al., 2011;Wen et al., 2017); loess deposition in western China increased in 6-3 ka (Lehmkuhl et al., 2000Porter, 2001;Huang et al., 2002;Maher et al., 2003;An et al., 2006;Stauch et al., 2012). On the eastern TP, multiproxy analyses of lake and peat cores corroborate cooling and drying after 6-3.7 ka (Zhang and Mischke, 2009;Zhao et al., 2011;Chen et al., 2014), with a slight warming after 3.7 ka (Zhang and Mischke, 2009), which we cannot evaluate in our data, because we have only two dates in younger loess that are older than 3.7 ka. Closer to Jiuzhaigou, cores of peat bogs to the west indicate a cool and dry climate after 5 ka . Overall, the climatic evidence in the region suggests that cool and dry conditions prevailed after 5 ka, when the younger Jiuzhaigou loess was deposited.
The deposition rate of the younger loess in Heye Catchment (0.39 ± 0.05 m/ka) may be inflated by human activity, a phenomenon documented elsewhere in China, because agriculture and land clearing increase erosion on uphill slopes and thus mobilize material deposited on loess deposits (Roberts et al., 2001). Because archaeological evidence suggests farmers arrived in Heye Catchment relatively soon after the younger loess started to accumulate (d'Alpoim Guedes et al., 2015), it is possible that farming accelerated loess deposition by redistributing material, including loess as well as locally derived sand and gravel, that sloughed from terrace scarps.

Paleowind conditions and loess sources
Although grain-size analyses identify three to five unique components of loess, we interpret three distinct sources and associated transport mechanisms for this material. The coarse fraction, which is present in all samples, has a mean grain size (44 ± 4 μm) that corresponds to grain sizes previously described as locally sourced material transported tens of kilometers (20-65 μm) during discontinuous events such as dust storms (Patterson and Gillette, 1977;Sun et al., 2004;Vandenberghe, 2013).The larger proportion of this sized material in lower older loess implies that surface winds were stronger during the Early Holocene when the lower older loess was deposited. The high westerly winds seem to have transported sediment through the entire period of loess deposition. High westerly winds likely transported the fine and medium (< ∼10 μm) material, as has been Figure 8. Example of a grain-size distribution for one sample, the three components, and the sum of the components as determined from QGrain. Each component is a lognormal distribution with the reported mean and standard deviation. Because the distributions are lognormal, the standard deviation is not symmetric in linear space. The mixture of the three lognormal distributions in the proportions noted on the figure sum to best approximate the measured distribution of this sample. Similar curves are available for all grain-size distributions analyzed in Supplementary Fig. S12. Samples are grouped by stratigraphic unit (lower loess, upper loess in the top 30 cm, and gravelly loess). An asterisk (*) above a box plot indicates that distribution of fraction of sediment contained in the peak for the stratigraphic unit shown in the box plot is statistically significantly ( p < 0.05) from the unit with the same color as the star for the peak shown. (G-K) mean grain size for each peak shown as a box plot with overlying jitters for each stratigraphic unit. An asterisk (*) at bottom as described for B-F. Center line is the median for box plots. Box boundaries extend to 25th and 75th percentiles. Whiskers are 1.5 times the interquartile range and outliers (not shown) are beyond that range.
inferred from similar data elsewhere in China (Sun et al., 2004). Finally, the very coarse fraction is too coarse to be transported by wind, even during discontinuous events (Patterson and Gillette, 1977;Sun et al., 2004;Vandenberghe, 2013). Thus, we conclude that this fraction is sourced from local bedrock during the same events that incorporated gravel into the younger loess.
Where did the gravel come from?
The existence of slump-block topography during deposition of the younger loess provides a mechanism for the incorporation into the younger loess of gravel and smaller lithics. Bedrock fragments mobilized by ravel and other modes of failure of the bedrock exposed at the base of some slump-block scarps would have fallen onto the relatively horizontal surfaces of slump blocks. As the loess accumulated, these locally derived lithics would then have been incorporated into the loess through bioturbation, pedogenesis, and farming (Fig. 10). As far as we know, a gravelly, lithic-rich loess such as this has not been documented elsewhere.
In Heye Catchment, deep-seated slumping produces failure planes through both loess and bedrock. This manner of slumping has apparently not been documented in other loess areas (see review by Li and Mo, 2019). Loess elsewhere in China experiences a range of landslide types that are attributed to both natural (Dijkstra et al., 1994;Wen et al., 2005;Zhang and Wang, 2007) and anthropogenic causes Li and Mo, 2019). However, rotational slides previously documented in Chinese loess typically only involve the loess deposit itself, while documented landslides that also involve underlying bedrock are typically fast and have a long run-out (see review by Li and Mo, 2019).
That gravel is found incorporated into the paleosol at most exposures and paleosols formed during the period of nondeposition in Heye (8.6-5.1 ka) that corresponds with the MHCO indicates the slumping likely started during the MHCO. The MHCO's warmer and wetter climate  would have promoted slumping, including involvement of the weak bedrock underlying the Early Holocene loess. However, it is also possible that the development of slumping was driven by activity of local faults and by the weighting of the slope by the younger Holocene loess. Slumping that exposes bedrock is the only source of local gravel to incorporate into the loess, and thus the only explanation we have for gravel in the younger loess. Similarly, the absence of gravel in the older loess provides a limiting date of approximately 9 ka for the onset of slumping. Field evidence of active ground deformation indicates the slump feature is active; whether its activity has been continuous or episodic is unknown, but persistent activity over long time periods is characteristic of slump earthflow around the world (Bovis and Jones, 1992;Mackey et al., 2009;Rutter and Green, 2011;Coe, 2012;Handwerger et al., 2013).

Implications for the history of human inhabitance
The earliest evidence for occupation of the Ashaonao site dates to 3.4 ka (d'Alpoim Guedes et al., 2015), while slumping likely started earlier than 4.1 ka based on the presence of gravel in loess of that age. The occupants of the Ashaonao site arrived after the slump terraces formed and during the subsequent deposition of the younger loess; the oldest date of occupation postdates the oldest dates of gravelly loess. This raises the possibility that the occupants of the Ashaonao site settled there because of the topography created by the slumping as well as the preference for farming on loess soils Schmidt et al., 2017). The documented use of wheat and barley at the Ashaonao site (d'Alpoim Guedes et al., 2015) provides further evidence of the site having been viewed as a favorable location for agriculture. The upper older loess is closer in grain-size distribution to the younger loess than the lower older loess. However, both the large gap of time between the older and younger loess deposits and the statistically identical ages of the older loess make it unlikely that there was a smooth transition in climate from the time of the lower and upper older loess. Thus, it seems unlikely that the upper older loess was deposited under different environmental conditions than the lower older loess. However, agricultural activity mixes soils, and thus it is possible that tilling for agriculture mixed younger loess into the upper 30 cm of the upper older loess. Abandonment of the site by ∼2 ka (d'Alpoim Guedes et al., 2015) has not been explained; however, a reactivation of slump failure planes provides a possible explanation.

CONCLUSIONS
Loess in Heye Catchment deposited during cool and dry climate from >11 to 8.6 ka, was interrupted during the warmer and wetter MHCO (∼8.6-4.1 ka), and resumed in the cooler and drier Late Holocene (<4.1 ka). The loess is a mixture of material of three to five grain-size distributions. The fine and medium distributions (< ∼10 μm) likely derive from transport by high westerly winds, the coarse distribution (∼45 μm) may be windblown material sourced from the eastern TP, and the very coarse fraction is probably fragments of the local bedrock sourced from slump blocks and incorporated into the upper older loess and younger loess by gravity and farming. The relatively large fraction of the coarse grain-size fraction present in the older loess, exclusive of the upper 30 cm of that unit, suggests that the regional winds transporting material from the eastern TP were stronger during the Early Holocene loess deposition (>11-8.6 ka) than during the later Holocene deposition (<5.1 ka). The Late Holocene (<5.1 ka) loess was mixed with gravel sourced from local bedrock exposed in slump scarps; wetter conditions during the MHCO may have contributed to the slumping that formed the slumpblock risers that are the presumed source for gravel in the younger loess. Archaeologically important stratigraphic layers in the gravelly loess suggest that human occupation of the site started soon after the Late Holocene loess deposition began, possibly due to the attractive combination of low angle surfaces created by the slump terraces and the loess soils.