Fieldwork and site description

We sampled five loess sections from a c. 200 km E–W transect along the Zailisky Alatau range front in SE Kazakhstan (Fig. S1). Since all sites were exposed as vertical cliff sections, we undertook fieldwork by abseil to ensure continuous down-profile observations and sampling. Prior to sampling, we cleared back at least 50 cm of the surface sediment to prevent contamination by recent sediment relocation and expose undisturbed sections. A brief description of our site locations is given below.

Remizovka (hereafter REM, 43° 13.2' N, 76° 51′ E; 1070 m a.s.l.) is a >25 m thick section exposed on a hillslope, originally excavated for the construction of ski jump facilities on the southern margins of Almaty city. Loess at the site has previously been described and dated by luminescence (Machalett et al., 2006, 2008; Fitzsimmons et al., 2018; Sprafke et al., 2018). A radiocarbon-dated subsection, which has since been removed during road construction (referred to as Tramplin, Feng et al., 2011), was recently placed in stratigraphic context to clarify ambiguity between studies (Sprafke et al., 2018). Our study here focusses on the uppermost 7 m of the main section.

Panfilov (PAN, 43° 22.295′ N, 77° 07.670′ E; 710 m a.s.l.) is a c. 5 m thick site located just southwest of the village of Panfilovo, c. 40 km northeast of Almaty. The loess section has been exposed as a result of erosion by the Tsyganski Creek, which flows northward from the Tien Shan.

Ashubulak (ASH, 43° 28.671′ N, 77° 47.379′ E; 760 m a.s.l.) is located at the southern edge of the village of the same name, c. 50 km east of the PAN section. The exposed outcrop at this site is 5.1 m thick and is entirely composed of pale yellowish primary loess. Like PAN, the ASH site is also located along the peripheral edge of the piedmont loess, where the loess deposits taper out northwards onto alluvial fans.

Taukaraturyk (TAU, 43° 29.445′ N, 78° 01.509′ E; 769 m a.s.l.) is a 7.5 m thick profile located on the southern edge of the village of Taukaraturyuk, c. 20 km and c. 100 km east of ASH and Almaty, respectively. Unlike the loess-marginal sites of PAN and ASH, which are located in more open, exposed sections c. 10 km out from the range front, the TAU site is located much closer to the mountain ranges (c. 1 km north of the first ridge of the Zailisky Alatau).

Malubai (MAL, 43° 26.312′ N, 78° 19.763′ E; 815 m a.s.l.) is located c. 20 km south of the city of Chilik (Shelek) and c. 25 km east of TAU. The exposed outcrop at MAL is 6 m thick and lies on the northern slope of a northeast–southwest bifurcating flank of the Zailisky Alatau, and is consequently more sheltered than the other sites.

Proxy indices

Samples for micromorphology, grain-size analyses and magnetic susceptibility were collected from four sites (PAN, ASH, TAU and MAL) at 10 cm intervals. We collected these samples in 8 cm³ plastic boxes hammered into the cleaned-back profiles. The same sample material was used for all three analyses in the order described below. Grain size and magnetic susceptibility data for the site of REM was already available from previous work (Fitzsimmons et al., 2018; Schulte et al., 2018).

Luminescence dating

Samples for luminescence dating were collected at 1 m intervals from all five profiles (REM, PAN, ASH, TAU and MAL) by hammering 3.5 cm diameter, 10 cm long steel tubes into freshly cleaned loess sections. A total of 27 samples were collected from five sites. Additional sediment for dose rate analysis was collected from the material immediately surrounding the tubes. The exposed outer material from the ends of the steel tubes was used for determining water content; sediment from the inner parts of the tubes was processed for equivalent dose analysis. All samples were processed under subdued red-light conditions at the Institute of Geosciences, Johannes Gutenberg University, and measured at the Max Planck Institute for Chemistry, respectively (Mainz, Germany). Wet-sieving of sediment yielded insufficient coarse grains (>63 µm) for measurement; we therefore prepared fine-grained (4–11 μm) quartz and polymineral samples using established protocols (Frechen et al., 1996; Timar et al., 2010).

We measured the equivalent dose (D_e) using optically stimulated luminescence (OSL) based on the single aliquot regenerative dose (SAR) protocol (Murray and Wintle, 2000, 2003) for the quartz fraction, and elevated temperature post-infrared infrared stimulated luminescence (pIRIR; Thiel et al., 2011) for the polymineral fine-grained (K-feldspar-bearing) material. In order to negate the suspicion of any feldspar contamination of the OSL signal from the quartz-rich fine grains, we applied the double-SAR approach (DSAR, Banerjee et al., 2001; Jain and Singhvi, 2001; Dave et al., 2019), which includes an additional IR stimulation prior to all blue stimulation steps within the SAR protocol. In samples where saturation of the quartz OSL signal was suspected (after Timar-Gabor et al., 2017), elevated pIRIR signals from fine-grained polymineral samples was used for D_e determination.

Radionuclide concentrations for dose rate determination were analysed using high-resolution germanium gamma spectrometry, measured at the Felsenkeller, VKTA Dresden. Dose rates were calculated from the radionuclide concentrations using published conversion factors (Guérin et al., 2011), combined with measured moisture content and published equations for cosmic-ray dose rate contributions (Prescott and Hutton, 1994). A detailed account of sample preparation, luminescence measurements and protocols, and dose rate calculations can be found in the SI.

Age–depth modelling and mass accumulation rates

Based on our investigations of fine-grained quartz as discussed below and the reliability of quartz-based OSL ages, we limited our assessment of loess MARs to the past 60 ka. The presence of a substantial number of dated loess sites within this age range allows for a good representation of loess sedimentation patterns across the Ili Basin, elsewhere in ACA and across the CLP.

We undertook age–depth modelling and calculated sediment accumulation rates for 30 additional sites based on published luminescence ages (based on quartz OSL and feldspar pIRIR ages) that span the time period 0–60 ka. Our analysis was based on high-resolution luminescence ages taken from 14 sites across the Ili Basin (including sites from the present study), eight sites from neighbouring sedimentary basins in ACA, and 11 representative sites from the CLP. Age–depth modelling for all sites was performed using the R package Bacon (Blaauw and Christen, 2011). The age–depth models derived using R-Bacon were used to estimate the corresponding sedimentation rates (SR, cm ka^-1), and then to evaluate MARs (g cm⁻²ka⁻¹) using the equation: MAR = SR × ρ_dry × f_eol, where ρ_dry is the dry bulk density (g cm^-3) and f_eol refers to the sediment fraction that is aeolian in nature (Kohfeld and Harrison, 2000). We used a value of f_eol = 1 for all calculations, since we assume that loess is entirely aeolian in nature. Based on published literature, we utilised mean bulk density values of 1.5 g cm^-3 for ACA (based on values obtained by Jia et al., 2018; Wang et al., 2019a; Li et al., 2019a) and 1.48 g cm^-3 for sites in the CLP (Kohfeld and Harrison, 2003; Kang et al., 2015). Details of the sites analysed in this study (and the selection thereof), criteria for the selection of luminescence ages, age–depth model parameters for R-Bacon, the constraints for MAR calculations and subsequent interpretation of different sites (where relevant) are described in the SI.

Stratigraphy and sediment characteristics

The stratigraphy of our five investigated sections, including grain size and magnetic susceptibility can be seen in Fig. 2. A detailed account of the micromorphological characteristics are summarised in Table S1 of the SI.

We focused on the uppermost 7.0 m of the REM loess profile, which spans c. 35–15 ka (based on pIRIR₂₉₀ dating; Fitzsimmons et al., 2018). The top 0.8 m of this section is characterised by recent soil formation, including penetration by living plant roots. A carbonate-enriched subhorizon is observed at 0.8–1.5 m depth, below which massive, homogeneous primary loess is observed down to 7.2 m. The magnetic susceptibility record shows relatively little variation down the profile below the recent soil. The grain-size record similarly yields minimal variations, with the exception of a slight increase in the coarser fraction at 3.9–4.3 m and 5.9–6.1 m depths (Schulte et al., 2018). Additional observations relating to colour indices and pedology are described in Fitzsimmons et al. (2018).

The PAN section is 5.1 m thick and overlies a >3.0 m thick fluvial gravel bed; the imbrication of cobbles indicates northward flow. The uppermost 0.5 m of the section comprises modern soil, penetrated by present-day roots. At 0.5–0.7 m, we observe a slightly darker humic horizon, which is intercalated with colluvial loess-like material. Below this layer and down to 0.9 m, colluvial angular clasts (2–4 mm) are observed which grade downwards into loess-like sediment. The upper c. 1.2 m yields higher magnetic susceptibility values (c. 130–90 × 10^-8 m³kg^-1) that support our observations of weathering and soil formation. From c. 0.9–5.1 m, the section is composed of pale yellowish primary loess, with a carbonate-enriched layer at 0.9–1.2 m. Reduced magnetic susceptibility values (c. 60 × 10^-8 m³kg^-1) below c. 1.2 m are consistent with unweathered, primary loess deposits. Micromorphological analyses of four samples (Table S1) from the uppermost primary loess (0.9–2 m depth) shows dominant well-sorted, fine-grained aeolian silt mixed with coarser, angular colluvial material (<4 mm); these observations are supported by the grain-size results (Fig. 2). The total dispersion method, which dissociates aggregates, yields a higher proportion of very fine clasts (<4 µm) than the KOH method, which indicates a high proportion of aggregates from 0.5 to 2.0 m. The aggregates are most likely derived from a combination of colluvial clasts, disintegration of carbonate-coated or carbonate-rich clasts, and aggregation by organic matter present in the sediment. We observe an abrupt increase in magnetic susceptibility at the base of the loess just above the gravels, although the reason for this is unclear.

The outcrop at ASH is 5.1 m thick and is entirely composed of pale yellowish primary loess. The modern soil is c. 30 cm thick with modern rootlets. While the geographic situation of ASH is similar to PAN, located along the northern margin of the piedmont loess deposits, the ASH sediments are more consistent with homogeneous, fine-grained primary loess and no colluvial input was observed. Magnetic susceptibility yields no major variations down the profile, which is consistent with primary loess deposition with no evidence of pedogenesis. Micromorphology of five selected samples (Table S1) provides further evidence of minimal disturbance of the sediments following aeolian deposition.

The TAU section is c. 7.5 m thick. The uppermost 0.3 m is composed of humic soil with modern rootlets and bioturbation, below which a blocky, carbonate-rich C horizon extends down to 0.7 m. From 0.7 to 7.5 m depth, the section comprises pale yellowish primary loess, with local occurrences of very coarse sand at depths of 1.5–2.0 m, 2.4–2.8 m and 3.8–4.9 m. Carbonate mottling was observed at 1.8–2.9 m and 5.5–7.1 m, and minor manganese concretions occur at 2.4–2.8 m and 3.0–3.8 m. The magnetic susceptibility values show subtle variations along the profile. Higher values (c. 90–120 × 10⁻⁸ m³kg^-1) occur in the uppermost 2.5 m. We interpret these peaks as incipient pedogenesis which was not observed in the field. There is considerable variation in the GSI in the upper 2.5 m, as well as three distinct increases in mean grain size and the GSI at 3.4–3.8 m, 4.0–5.0 m and 6.0–7.5 m (Fig. 2) which occur as subangular very coarse sand (Table S1). We interpret these peaks as evidence of colluvial input from the nearby bedrock outcrops, which are c. 0.7 km south of the site. Micromorphological analysis on six selected samples throughout the profile also identified persistent occurrence of angular gypsum crystals below 1.6 m depth (Table S1), which should represent in situ precipitation under arid conditions.

The exposed outcrop of MAL is 6.0 m thick. The uppermost c. 0.4 m is composed of modern soil with roots developed within colluvial loess-like material. At 0.4–0.9 m depth, we observe fewer angular clasts within a loess matrix. A discontinuous very coarse sand to gravel layer with silt matrix was observed at 1.1–1.3 m, below which carbonate-rich loess dominates down to c. 2 m. Below 2.0 m depth, the section is composed of pale yellowish-beige primary loess, with occasional carbonate mottling and Fe–Mn concretions observed at 2.5–5.0 m. The MAL section yields higher magnetic susceptibility (c. 70 × 10⁻⁸ m³kg⁻¹) in the upper 1.5 m, most likely reflecting recent pedogenesis. Mean grain size and GSI vary substantially at 0.5–2.5 m; below 5.0 m, a sudden decrease in grain size is observed. The grain-size variations in the upper 2.5 m most likely reflect variable colluvial input to the silt-dominated aeolian component. Below 2.5 m, the mean grain size, as well as GSI, does not show significant variation and is likely to reflect purely aeolian input to the site. The abrupt decrease of GSI in the primary loess below 5 m most likely indicates a decrease in wind strength (Fig. 2).

Chronology of the Zailisky Alatau loess sections, Central Tien Shan piedmont

The samples in this study were dated using quartz OSL and polymineral pIRIR dating and a detailed record of the luminescence characteristics, equivalent dose and dose-rate estimations and age calculations can be found in the SI. The final age estimates from all five loess profiles in this study are summarised in Table 1 and Fig. 3. Fine-grained quartz OSL ages were derived for 19 samples from four sites (REM, PAN, ASH and the upper 1 m of TAU; Fig. 3) The upper 7 m of the profile REM spans 10–40 ka and falls within the uncertainties of previously published pIRIR ages (Fitzsimmons et al., 2018), while profiles PAN, ASH and the upper 1 m of TAU date to between c. 5 and 17 ka. Polymineral fine-grain pIRIR dating on eight selected samples from below 1.5 m at TAU and the entire section at MAL yielded minimum age estimates (Table S2), as these samples exhibited saturation of the pIRIR₂₉₀ signal. Hence, we consider that the deposition below 1 m at TAU and at MAL occurred >180 ka.

Age–depth modelling and mass accumulation rates

Figure 4 shows the location of our sites, as well as of published sites from ACA and the CLP, for which we calculated MARs from luminescence-based chronologies (quartz OSL and feldspar pIRIR ages). We limited our calculations of MARs to the last 60 ka since (i) our investigations suggest an upper limit to reliable dating of quartz in this region (based on saturation of the quartz OSL signal) of c. 70 ka, and (ii) most published loess sites in ACA and the CLP span this time period, therefore allowing for a representative evaluation of loess depositional dynamics across these regions.

The MARs of sites in the Ili Basin, elsewhere in ACA and across the CLP are illustrated with respect to concentrations of dust within the Greenland NGRIP ice core (Ruth et al., 2007), stable oxygen isotope records from NGRIP (Rasmussen et al., 2014) and the global marine stack (Lisiecki and Raymo, 2005), and June solar insolation at 65°N (Berger and Loutre, 1991) in Fig. 5. We observe substantial differences in absolute accumulation as well as the timing of peaks in accumulation between sites in the Ili Basin and across ACA more widely (Fig. 5a, b). There are similar differences across the CLP (Fig. 5c). Furthermore, based on our calculations of MARs from 11 sites across the CLP, we observe a distinct difference between sites located in the northwest (NW) and those in the southeast (SE). The geographical difference in loess accumulation in the CLP has previously been reported for various time periods (Lu and Sun, 2000; Kohfeld and Harrison, 2003; Xu et al., 2018; Liu et al., 2020).

Spatio-temporal variation in loess deposition along the Central Tien Shan (Zailisky Alatau region)

Our new dataset adds chronological constraints for five loess sites (including new data from REM) from the virtually unexplored Zailisky Alatau piedmont in the western portion of the Ili Basin. Our sites bridge the geographical gap between the relatively intensely studied eastern Ili Basin (ChongYi et al., 2012; Kang et al., 2015; Song et al., 2012, 2015; Li et al., 2016b, 2018a; Wang et al., 2019a,b; Li et al., 2020) and several dated sites further west at Maibulak (MBK) along the Zailisky Alatau (Fitzsimmons et al., 2017), Bishkek (BSK) in the Chu River valley (Youn et al., 2014) and Valikhanova (VAL) on the eastern slopes of the Karatau Range (Fitzsimmons et al., 2017) (Fig. 4a).

We identified phases of loess accumulation and illustrated these with respect to altitude, deposit thickness and distance from the range front in a schematic diagram in Fig. 6. Geographically, the four sites of MBK, REM, PAN and ASH lie west of the Ili Gate and in the more open part of the basin. TAU and MAL lie at approximately the narrowest part of the basin (the ‘Gate′) and are comparatively sheltered from northerly winds. The sites of PAN, ASH, TAU and MAL lie at similar elevations (c. 700-800 m a.s.l.); both MBK and REM are situated at higher altitudes (1070 m a.s.l). The distances of the sites from the range front are also variable. MBK in the west overlies an alluvial fan c. 350 m from the range front and c. 2 km northward of a steeper break in slope. REM is situated atop a spur of the foothills which rises c. 200 vertical metres above the plain, and c. 5 km from the major break in slope representing the transition to bedrock ranges. PAN and ASH lie the greatest distance from the range front, c. 7.5–8 km to its north. TAU is located c. 1 km northward of the first ridge of the Zailisky Alatau, and MAL lies c. 150 m north of a bedrock spur and c. 2.3 km north of the main range. With the exception of REM, where the range front is oriented approximately SSW–NNE, all sites are situated northward of an east–west trending break in slope. We hypothesise that (1) location within the basin, (2) distance from the range front and (3) its strike are all likely to have played a role in the potential of individual sites to trap aeolian sediment through time.

The westernmost sites, MBK and REM, range in age between 12 and 45 ka (this study; Fitzsimmons et al., 2017, 2018). Although the loess at REM extends downward toward older ages (Machalett et al., 2006; Sprafke et al., 2018), the lack of high-resolution dating >45 ka prevents our consideration of accumulation during earlier periods. By contrast, the two central sites, PAN and ASH, are substantially younger in age and span much shorter periods of time; 17–5 ka and 15–12 ka, respectively. The site of TAU, situated c. 20 km east of ASH, preserves c. 1 m of loess accumulation during a similar time period to ASH (c. 11–13 ka) and overlies an unconformity with substantial hiatus, below which the loess exceeds 180 ka. The easternmost Zailisky Alatau site of MAL predates 180 ka and yielded no late Pleistocene deposits. We hypothesise that the location of TAU and MAL within the more sheltered Ili Gate area led to reduced loess accumulation over the late Pleistocene compared with the more exposed western sites. Increased wind strength related to a ‘funnelling' effect by the enclosing mountain ranges may also have led to the erosional unconformities observed at TAU and MAL.

While loess grain size is often used as a proxy for wind intensity through time (An et al., 1991a; Porter and An, 1995; Sun and An, 2005; Vandenberghe, 2013), our observations from the Ili Basin loess deposits strongly suggest that additional controls, such as geomorphic setting, sediment availability and supply to individual sites, also influence aeolian flux and grain size. We compared changes in GSI from three study sites (REM, PAN and ASH) with those from the published site of NLK in the eastern part of the basin (Li et al., 2018c), focusing on the time period c. 12–16 ka (Fig. 7). GSI values during this interval range from 0.30–0.40 at REM, 0.40–0.65 at PAN, 0.25–0.45 at ASH, and 0.30–0.60 at NLK. We observe that not only the magnitude of variability in GSI but also the mean value differs between sites. Sediments at PAN are coarser than at ASH, despite the relative proximity of these two sites. GSI values are less variable at REM in the west than at NLK in the east (Fig. 7). These differences within the same time interval along the piedmont suggest spatial variability in sediment supply, transport to and deposition at an individual site. For example, the proximity of both NLK and PAN to active fluvial channels may account for relatively greater amounts of proximal transport of coarser grains (Li et al., 2018c) than can be transported to the other sites. While the site REM, located away from fluvial channels and at a higher elevation relative to the other sites, shows overall finer GSI values and a smaller range of GSI at the site, which may reflect a greater reliance on distal transport and sorting.

The geomorphic context of individual sites is likely to influence not only grain size characteristics and variability through time, but also aeolian flux, expressed here as MARs (Fig. 7). For the 12–16 ka time period, aeolian flux at ASH far exceeds that at any other site, with PAN experiencing the next highest rate of accumulation. We suggest that its relatively low elevation, distance from alluvial fans and range front, and a strengthened westerly trajectory of dust-laden winds at the Ili Gate resulted in higher accumulation here than at any other site along the piedmont for which we have data. On the other hand, while MARs at PAN are lower than at nearby ASH, its GSI range is higher and may reflect greater immixing of proximal fluvial material with distal accumulation. This observation raises questions regarding hitherto popular correlation of proxies such as grain size and aeolian flux with wind strength in loess deposits generally. The older deposits at MAL and TAU yield GSIs in the 0.4–0.9 range, which is greater than those observed at the other sites. This may relate to the proximity of TAU and MAL to a seismically active range front (Chilik fault), which can introduce coarser clasts to the loess via slope transport. These observations highlight the need to first interrogate the geomorphic setting of individual sites for their potential to reflect local or larger scale processes according to proposed research questions.

We observe along the Zailisky Alatau transect, a ‘patchwork' loess piedmont of spatially variable timing of peaks in deposition, flux and grain size. This most likely reflects the result of complex interactions between local and continental wind regimes and associated dust transport, topographic (including palaeotopographic) context, and local sediment supply.

Loess sedimentation dynamics across the Ili Basin: Overcoming individual site bias to reconstruct the interplay between the Westerlies and the Siberian High Pressure system

The ACA piedmonts, especially the Ili Basin, lie at a pivotal topographic point which exposes the region both to the mid-latitude westerly winds and the Siberian High Pressure, the latter manifesting here in the form of northerly winds associated with the seasonal migration of the high-altitude polar jet (Fitzsimmons et al., 2020). The Westerlies rarely penetrate eastward of the Tien Shan range; loess deposited in this part of ACA therefore represents a strategic record of aeolian transport and deposition associated with both climate subsystems. Published correlations from the eastern Ili Basin loess linking grain size with penetration of westerly air flow suggest that the influence of the Westerlies was neither consistent nor strong over the Late Pleistocene (Li et al., 2018c). The loess deposits of the Ili Gate/Zailisky Alatau region investigated in this study, coupled with data from the eastern Ili Basin, can provide more meaningful information relating to the interplay between the Westerlies and Siberian High climate systems through time.

Here we examine the degree to which loess profiles across the entire Ili Basin record past environmental conditions in the form of accumulation rates. In Fig. 5a, we show MARs from our new datasets from REM, PAN and ASH, in addition to those from 11 published sites (ChongYi et al., 2012; Kang et al., 2015; Song et al., 2012, 2015; Li et al., 2016b, 2018a, 2020; Fitzsimmons et al., 2017, 2018; Wang et al., 2019a,b), presenting a 60 ka record from the eastern headwaters of the Basin to the ‘Ili Gate'.

We observe substantial differences in absolute MARs between sites, not only over the 60 ka range but also over shorter timescales such as the deglacial and marine isotope stage (MIS) 3 interstadials. We hypothesise that absolute MARs are likely to reflect differences in topographic settings between individual sites. Absolute sedimentation rates at NLK, for example, are approximately five times higher than those observed at the sites KS15 and XY17 during late MIS 3 (40–27 ka). NLK is located on the banks of the Kax River (Kashi; Song et al., 2015), whereas KS15 is located in the easternmost Ili Basin and on the upper terraces of the Kunes River (Li et al., 2018a). While both these rivers, tributaries of the Ili, derive from glacial sources, the closer proximity of NLK to the river banks implies a higher sediment availability during MIS 3, when glaciers in the Tien Shan expanded in response to increased moisture transport to ACA (Koppes et al., 2008). The topographic setting of KS15 is also likely to have resulted in relatively low aeolian flux during other time periods. The site KS15-05 (Wang et al., 2019a), located c. 3 km east of KS15 (Figs. 4a and 5a), yielded MARs three times higher than at KS15 during late MIS 2 and early MIS 1 and is likely to relate to the closer proximity of the former to the active river channel.

There is also considerable variability in the timing of peaks in loess accumulation across the basin. For example, despite the proximity of the two eastern Ili Basin sites NLK and ZKT (Fig. 4a and Fig. 5a), the former yields peak MARs during late MIS 3 (c. 45–30 ka) and the latter peaks during early MIS 3 (c. 60–47 ka). These site-specific variations in loess sedimentation rates raise questions regarding how representative the MARs of individual sites may be for understanding regional climatic variations. We propose instead that aggregated trends in loess MARs from multiple sites across a basin are more likely to reflect changes in climate dynamics, thus providing the best means to overcome the bias of individual sites based on their local setting.

We observe a general increase in sedimentation across the entire Ili Basin during interstadial MIS 3 (29–57 ka; Lisiecki and Raymo, 2005); this builds on findings from previous studies which noted primary loess accumulation at this time in the Zailisky Alatau area (Fitzsimmons et al., 2018). Fitzsimmons et al. (2018) hypothesised that increased loess flux during late MIS 3 (c. 40–27 ka) was due to a general increase in sediment availability and wind strength at this time. Increased sediment supply was likely to be linked to glacial advances in northern and eastern Tien Shan (Kong et al., 2009; Li et al., 2011, 2014; Chen et al., 2015) driven by increased moisture transport by the Westerlies, which is also responsible for the weak pedogenesis observed within the loess during this time (Song et al., 2012, 2015; Fitzsimmons et al., 2018). Furthermore, it has been proposed that the Siberian High, which intensified during late MIS 3 (Ding et al., 1995; Hao et al., 2012), would have compressed the Westerlies against the Tien Shan ranges and increased wind strength into the Ili valley (Fitzsimmons et al., 2018). The combination of increased sediment availability and wind strength in the Ili Basin would therefore have contributed to increased loess accumulation observed in aggregate during this period (Fig. 5a). Local katabatic winds might also have played an important role during this phase, resulting in localised effects on the timing and peaks of loess sedimentation.

We also observe generally increased accumulation rates in aggregate during MIS 2 (29–14 ka, Lisiecki and Raymo, 2005) (Fig. 5a). These conditions coincide with globally cold climates prevailing during the Last Glacial Maximum (26–19 ka LGM; Clark et al., 2009). In the Ili Basin the conditions were not only cooler, but also more arid: stable carbon isotopic (δ₁₃C_org) reconstructions for palaeovegetation in the valley indicate an increase in more arid-adapted C₄ vegetation (Ran and Feng, 2013). A cold, dry, windy climate appears to have resulted in coarser grain sizes within a number of loess sequences across the Ili Basin (Song et al., 2015; Li et al., 2016b, 2018c). Climate simulation models indicate windier LGM conditions, due either to stronger mid-latitude Westerlies (Sun et al., 2012) or intensified Siberian high pressure (Cheng et al., 2021). Glacial expansion during MIS 2 in the Central and Eastern Tien Shan (Kong et al., 2009; Li et al., 2011, 2014; Lifton et al., 2014; Blomdin et al., 2016), also resulted in increased production of silt, therefore according availability within the Ili catchments for dust entrainment and transport.

Our analyses also suggest a relatively widespread increase in MARs across most Ili Basin sites during the warming conditions of the global deglacial period (19–11 ka, Clark et al., 2012) (Fig. 5a). Those sites which do not show any increase (KS15 and ZSP) are likely to be more influenced by their topographic setting; both are located in the easternmost or upper reaches of the basin with minimal exposure to Westerly air flow (and in general record very low accumulation rates). Elsewhere, short-lived phases of increased flux of coarser grain size fractions have been correlated to abrupt climatic events such as the Younger Dryas (c. 12–11 ka) at several sites during the deglacial period (e.g. NLK: Li et al., 2018c; XEBLK: Li et al., 2016b), and have been hypothesised to associate with increased Westerly penetration eastward into the Ili Basin. Climate simulations (Wyrwoll et al., 2016) suggest that the strength of the Siberian High was reduced at this time. Such conditions would have increased the influence of the Westerlies on the Ili Basin, including increasing rainfall; increased deglacial moisture availability is supported by more dominant C₃ vegetation signatures in the basin (Ran and Feng, 2013). We suggest that sediment availability was greater than during the LGM due to increased runoff promoted both by glacial melt under warmer conditions and increased precipitation transported by the Westerlies.

Variability in loess accumulation rates across mid-latitude Asia: Implications for interpreting loess archives from Central Asia to the Chinese Loess Plateau

Continental mid-latitude Asia, with its extensive deserts and widespread loess deposits, is believed to be one of the major contributors to the global dust cycle (Narisma et al., 2007; Kok et al., 2021). It is therefore imperative that we understand the processes driving dust flux and their contribution to Northern Hemisphere climate dynamics. One major step towards our understanding of these processes involves quantifying Quaternary dust accumulation. Loess MARs provide the closest approximation of the ‘dust parameter' over Pleistocene timescales. Given the issues raised by our analysis of the Ili Basin piedmont deposits, we compared loess MARs elsewhere in continental mid-latitude Asia over the same time period (0–60 ka), with a focus on the CLP and published sites across ACA more widely. In doing so we further assess spatial and temporal inhomogeneity in loess MARs and the degree to which regional characteristics such as topography and climate influence MARs.

The CLP of central northern China was the first loess region to be recognised as a significant terrestrial palaeoclimate archive (Liu, 1962), and is now one of the most intensively studied loess areas in the world (An et al., 1991a,b; Porter and An, 1995; Ding et al., 2002; Kohfeld and Harrison, 2003; Sun and An, 2005; Stevens et al., 2008, 2013, 2018). Unlike the Central Asian piedmonts, the CLP forms an extensive, continuous plateau exceeding 200 m thickness in parts and centred on the southern half of the Ordos Loop of the Yellow River (Yang and Ding, 2010). The loess is believed to derive from denudation associated with uplift of the Tibetan highlands (Sun and Liu, 2000; Sun, 2002; Smalley et al., 2014); its connection with the ice sheets of the Tibetan ‘third pole' underpins the assumption linking primary loess with glacial phases. The dominant sources of loess to the CLP are sediments derived from the Tibetan plateau, transported by the Yellow River prior to entrainment as dust (Stevens et al., 2013; Nie et al., 2015, 2018; Bird et al., 2015, 2020); a significant proportion is recycled along the length of the river system as it flows along the plateau (Licht et al., 2016). The climate of the CLP is strongly influenced by the East Asian winter and summer monsoons (EAWM and EASM, respectively; Lu et al., 2022). Northwesterly winds associated with the Siberian High Pressure system produce cold, dry, dust-bearing EAWM winter conditions (Liu and Ding, 1998; Maher, 2016), which alternate with rain-bearing southeasterly EASM summer monsoonal air flow (An et al., 1991b; Yang et al., 2015). Grain-size analyses suggest that loess particles coarsen with strengthening EAWM during glacial conditions (Sun et al., 2012; Stevens et al., 2018). Several studies report a southeastward decrease in loess flux which correlates with decreasing grain size, and has been linked to the strength of the EAWM (Lu and Sun, 2000; Vriend et al., 2011; Liu et al., 2020). However, several high-resolution luminescence dating studies challenge the prevailing assumption that CLP loess accumulation is uniformly driven by the strength of the EAWM (Stevens et al., 2006, 2007; Xu et al., 2018).

Fig. 5c shows MARs from 11 selected sites (Lai and Wintle, 2006; Lai et al., 2007; Lu et al., 2007 Stevens et al., 2008, 2016; Buylaert et al., 2008; Sun et al., 2012; Kang et al., 2013; Qiu and Zhou, 2015; Fig. 4b) across the CLP, following a northwest–southeast (NW–SE) transect over the period c. 0–60 ka. We observe a distinct difference in net loess accumulation, as well as in the timing in peak MARs, between the northwestern and southeastern sites. MARs in the northwestern sites are approximately three times higher than those in the southeast.

As we observed for the sites in the Ili Basin, the geomorphic context in the CLP appears to influence both absolute accumulation and the timing of peaks in accumulation at individual sites. For example, among northwestern sites, JY shows higher accumulation rates than ZJC and GL, and likely relates to the proximity to the Yellow River; JY is located on the upper terraces of the river, whereas ZJC and GL are both situated further from fluvial source, south and north of the river, respectively. In contrast to the generally low accumulation rates of the southeastern sector, the site of MN, located on the banks of the Yellow River, has net accumulation comparable to the northwestern sites (Qiu and Zhou, 2015; see Figs. 4b and 5c). In general, sites in close proximity to the Yellow River, as well as those close to the desert deflation zones, record higher MARs than those more distal to these likely sources. It appears that despite the role of the EAWM as a dust transport vector (Vriend et al., 2011; Liu et al., 2020), net accumulation of loess across the CLP is also locally influenced by proximity to source sediment, whether fluvial or aeolian. Nevertheless, desert marginal sites, such as Jingbian (Stevens et al., 2018), also record erosional unconformities; this effect has been proposed to result from increased erosional capacity of the EAWM in those regions. Not all records of MARs can be easily explained by climate mechanisms or by geomorphic context, however; the extremely high MAR at BGY during LGM (Fig. 5c) may relate to local preservation potential associated with expansion and contraction of the EASM (Buylaert et al., 2008).

The timing of MAR peaks in aggregate across the CLP gives a more nuanced insight into the climate dynamics acting in the region than previously obtained. We observe a divergence in the timing of peaks in loess accumulation between the northwest and southeast. In the northwest, loess accumulation increases during c. 18–11 ka and decreases at c. 11–6 ka (Fig. 5c), whereas the opposite is observed in the southeastern sites. This alternating pattern has previously been attributed to a persistent weakening of the EAWM and gradual strengthening of the EASM during the early Holocene, with the opposite proposed for the deglacial (Xu et al., 2018; Kang et al., 2020). We suggest that the increase in MARs of the CLP southeastern sites during early Holocene may relate to their proximity to the Wei River, which drains the Qinling Mountains and is more strongly influenced by the EASM; aeolian flux in this region would therefore respond more closely to sediment availability relating to increased runoff during strengthened monsoon conditions. Overall, the CLP experienced an increase in MARs during the LGM and is consistent with widespread glaciation on the Tibetan Plateau (Owen et al., 2003a,b, 2006; Owen and Dortsch, 2014), although relative changes in the southeastern sites was less pronounced, likely due to distance from the Yellow River and northern deserts. Loess MARs decreased altogether during MIS 3 (c. 57–29 ka, Lisiecki and Raymo, 2005), most likely due to relatively weakened EAWM wind strength, although sediment supply was sustained by suspended load fluvial outwash linked to glaciations on the Tibetan plateau (Owen et al., 2003a; Owen and Dortch, 2014; Rother et al., 2017) in response to increased monsoonal precipitation. Local preservation effects of vegetation during warmer interglacial periods may have also contributed to the decrease in sedimentation rates. Nonetheless, the sustained supply of sediment to the CLP can be observed in the generally higher MARs at the northwestern sites proximal to the river.

In addition to our analyses of MARs across the Ili Basin and CLP, we also calculated MARs from eight other published loess sites across ACA (Figs. 4a and 5b). These sites not only represent piedmont settings similar to those found in the Ili Basin but are also influenced by similar climatic contexts. The westernmost sites of BSK (Youn et al., 2014) and VAL (Fitzsimmons et al., 2017) lie in the foothills of the Kyrgyz Tien Shan and Karatau Range, respectively (Fig. 4a). The sites of LJW10 (Li et al., 2015), SCZ17 (Duan et al., 2020) and BYH10 (Li et al., 2016a) are located along the northern piedmonts of the eastern Tian Shan, south of the Junggar Basin (and the Gurbantunggut desert), and TC and YM are situated in the westward opening Tacheng Basin, bound by the northern edge of the Dzungarian Alatau and the Tarbagatai Range to the north (Fig. 4a; Li et al., 2019b).

As observed elsewhere, both absolute MAR values and the timing of peak accumulation are highly variable across the greater ACA. In the west, BSK experienced increased accumulation during late MIS 3, continuing into the LGM, while VAL yields substantially lower MARs which remained consistent from late MIS 3 into the LGM. The higher accumulation rates at BSK are most likely due to its proximity to the glacially derived Chu River. Glacial expansion in the Kyrgyz Tien Shan during MIS 3 (Koppes et al., 2008) likely generated the fine-grained sediment for loess deposition at BSK, whereas the Karatau Range, from which the VAL sediments were sourced, was never glaciated and is much more arid, resulting in substantially less sediment supply. Both Tacheng Basin sites yield low, almost constant MARs (c. 30-40 g cm⁻²ka⁻¹) throughout the past c. 60 ky (Li et al., 2019b). It is likely that these records relate to the lack of glaciation in the surrounding mountains, resulting in a dependence on the desert to the west as a source of dust (Li et al., 2019b). By contrast, the sites located in the Northern Tien Shan foothills yield variable depositional characteristics (Figs. 4a and 5b). The site of QS-16 yielded an LGM peak in accumulation; by comparison, the site of BYH-10, located further east, experienced peak MARs from early to mid-MIS 3 an order of magnitude greater than at QS-16, followed by a hiatus and subsequent LGM MAR values comparable with QS-16. LGM deposition in the northern Tien Shan area has been linked to regionally arid conditions and strengthened Westerlies which increased aeolian flux (Li et al., 2016a, 2020). The substantial difference in absolute MAR values at BYH-10 and QS-16 is most likely linked to their individual topographic and geomorphic setting; the former is located on the exposed northern piedmont of the Bogda Shan (eastern Tien Shan) which drains a number of glacially fed rivers, whereas QS-16 is sheltered from the prevailing westerly winds by the Borohoro mountains and is at a greater distance from the northern dust source regions.

It is clear from our MAR review of the Ili Basin, wider ACA and CLP that absolute accumulation rates and timing at a given site may be influenced by both synoptic-scale climate and local geomorphic settings in the form of proximity to dust sources, topography and sediment availability. Here, it is important to emphasise the influence of post-depositional erosion on sedimentation rates at a site; which in an aeolian context, is implicitly controlled by the topographic setting as well as local climatic conditions at the site. Notwithstanding, our review suggests that the timing of peaks in loess accumulation as an aggregate of multiple sites represents a reliable response to climate.