The Northern Route for Human dispersal in Central and Northeast Asia: New evidence from the site of Tolbor-16, Mongolia

The fossil record suggests that at least two major human dispersals occurred across the Eurasian steppe during the Late Pleistocene. Neanderthals and Modern Humans moved eastward into Central Asia, a region intermittently occupied by the enigmatic Denisovans. Genetic data indicates that the Denisovans interbred with Neanderthals near the Altai Mountains (South Siberia) but where and when they met H. sapiens is yet to be determined. Here we present archaeological evidence that document the timing and environmental context of a third long-distance population movement in Central Asia, during a temperate climatic event around 45,000 years ago. The early occurrence of the Initial Upper Palaeolithic, a techno-complex whose sudden appearance coincides with the first occurrence of H. sapiens in the Eurasian steppes, establishes an essential archaeological link between the Siberian Altai and Northwestern China . Such connection between regions provides empirical ground to discuss contacts between local and exogenous populations in Central and Northeast Asia during the Late Pleistocene.


Sediment analyses
At Pit 1, bulk samples for analysis were collected at 5cm intervals in a column extending from the top to the base of the section. Within unit 3, careful sampling ensured that samples came from single laminae within the finely laminar sediment, and the sampling interval varied as a result. Blocks for micromorphological analysis were also collected from the section (see figure S5). At Pit 4, bulk samples for analysis were collected every 5cm along a column extending from the top to the base of the north wall of the excavation, including a test pit and a stepped trench cut down the side of the gully (see figure S6). The total depth of the sampled sediment is 255cm. Figure S3: Sampling locations for bulk sediment analyses in Pit 4, north wall (Samples 1-32). Geo-localized 3D photogrammetry in Agisoft Photoscan, data collected in New Plot (8,9) 3D plot generated in ISTI-CNR MeshLab.

Laboratory Methods
Screening and weighing, loss-on-ignition, magnetic susceptibility, and laser particle size analyses took place in the department of Geography, University of Cambridge. For loss-on-ignition analysis, samples were placed in a weighed crucible and successively heated to 400, 480, and 950 o C in a muffle furnace. Changes in sample mass after each heating indicate the proportion of uncarbonized and carbonized organic matter and calcium carbonate, respectively, in each sample. Magnetic susceptibility was measured on a Bartington Instruments magnetic susceptibility meter. Screened samples were dehydrated, weighed into pots, three measurements were taken and the results averaged. Laser particle size analysis was carried out using a Malvern Mastersizer 2000. Screened samples (<2mm) were disaggregated with sodium hexametaphosphate for 24 hours. A vortex mixer was used to agitate the samples before their introduction into the Mastersizer.
Blocks for micromorphological analysis were prepared in the McBurney Laboratory for Geoarchaeology, Department of Archaeology, University of Cambridge. The intact blocks of sediment were air dried, then oven dried for 48 hours to drive off any remaining moisture, whose presence might disrupt the subsequent impregnation of the blocks with clear resin. The dry blocks were immersed in a preparation of polyurethane resin thinned with acetone in the ratio 9:1, and with MEKP in a minute quantity (0.00002%), and were placed in a vacuum chamber to ensure that the resin penetrated to the center of the blocks and filled all of the voids and pore spaces within them. More resin was added as needed to ensure complete immersion of the blocks. The samples were then allowed to cure for two months at room temperature. Slices of the nowhardened blocks were cut using a rock saw, temporarily mounted on glass, lapped on a Brot thin section machine to create a smooth surface, permanently mounted on prepared glass slides using polyurethane resin, lapped to a thickness of around 30 μm and hand-finished by sanding, then cover slipped. Three of the thin sections were prepared by Julie Miller (Earthslides). The thin sections were analysed using petrographic microscopes in plane and cross-polarized light.
Tolbor-16 Pit 4: The sediments at Pit 4 can be divided into 4 main units (Figures S4 and S6a): the recent soil, a light brown kastanozem (unit 1; samples 1-3, 0-15cm depth), a thin (truncated) deposit of light brown loess with sparse gravel and humic patches toward its base (unit 2; samples 4-9, 16-49cm depth), a thick deposit of soliflucted laminar silt which extends to a depth of about 2.5 meters (unit 3; samples 10-32, 50-230cm depth), and a thin grain-supported deposit of sorted gravels (unit 4; samples 47-52, 231-255cm depth), possibly marking the top of a deeper sequence of fan-type sediments. Unit 3 is subdivided into solifluction lobes 3a (a stone-banked lobe to the west of the section, not sampled), 3b (samples 10-19; 50-99 cm depth), 3c (samples 20-23, 100-119cm depth), 3d (samples 24-26, depth 120-134cm) and 3e (samples 27-46, depth 135-230cm). As at Pit 1, the interface between units 2 and 3 is marked by an episode of slope changeresulting from the initiation of loess depositionand solifluction which truncates unit 3 and mixes material from units 3 and 2 along the contact. Section drawings for the main excavation are presented in Figure S4a and b; results of analyses are presented alongside a deeper section drawing (compiled from the steps cut into the hill) in Figure S6.

Sedimentological analyses
Loss-on-Ignition: The results of loss-on-ignition analysis at Pit 1 are presented in figure S5b (carbonate content, grey line), c (carbonized organic matter, blue line) and d (uncarbonized organic matter, orange line). The results of loss-on-ignition at Pit 4 are presented in figure S6b (carbonate content, grey line), c (carbonized organic matter, blue line), and d (uncarbonized organic matter, orange line). Particularly high organic matter and calcium carbonate contents are recorded for sub-units 3c, 3d, and 3e.
Magnetic susceptibility: The results of magnetic susceptibility analyses at Pit 1 are presented in figure S5h, and at Pit 4 are presented in figure S6h. Magnetic susceptibility varies through the sections but not systematically. The highest measured value at Pit 4 is within unit 2. A small peak occurs in the gravel horizon of unit 4, perhaps because of the lithology of the gravel.
Particle size analysis: Gravel contents (measured by screening and weighing the sediment samples) at Pit 1 are presented in figure S5g and at Pit 4 are presented in figure S6g. Figure S5d and e, and S6d and e are derived from laser particle size analyses of the fine fraction (>2mm). Figures S5d and S6d present the standard deviation within the analysed samples; primary loess is typically very well sorted with a low standard deviation, while weathering, soil development, and colluviation all raise the standard deviation within the samples for different reasons. Where standard deviation within the sections is low, the dominance of eolian processes is presumed. Figure S5e and S6e present the mean particle size of the silt fraction. This is intended as a measure of the energy of deposition; sand and clay are excluded because they may reflect post-depositional processes. Figure S7 presents the proportions of clay, silt, and sand (S7b) and particle size distribution curves for all of the sampled sediments at Pit 4 (S7c). All of the analysed sediments display a mixture of sand, silt, and clay (despite gravel contents as high as 22% by weight). The plots in figures S5d, e, and f, S6d, e, and f, and figure S7 confirm that multiple, generally low-energy sedimentary processes affect the sampled sediments. Figure S5. The Western wall at Pit 1 (a) and the results of sedimentological analyses. Rectangles: micromorphology. Shading indicates sediment color. b. Calcium carbonate content (grey line, %, bottom axis). c. Carbonized organic matter (blue line, %, top axis) and d. Uncarbonized organic matter (orange line, %, top axis) by loss-on-ignition. Gaps are samples which were removed by customs officials. e. Standard deviation within the fine (<2mm) fraction, by laser particle size analysis. The extent to which the sampled sediments are dominated by loess-like sedimentation (low standard deviation) or sheet erosion or soil development (higher standard deviation) may be a proxy for precipitation. f. Mean particle size of the silt fraction (µm, top axis). Coarser particles mean higher-energy deposition. g. Gravel content of the sampled sediments (weight %, bottom axis). Gravel represents gravitational input. Increased gravel content indicates longer surface exposure and thus slower sedimentation, and the distance between e and f indicates surface stability. h. Magnetic susceptibility (SI). Variation below the Holocene soil complex is difficult to explain, and may reflect the lithology of the local bedrock and its varying presence within the section. Figure S6: a. The sampled section at Pit 4, including a stepped trench below the main excavation, and the results of sedimentological analyses. Shading indicates sediment color. b. Calcium carbonate content (grey line; %, top axis), c. Carbonized organic matter (blue line, %, bottom axis). d. Uncarbonized organic matter (orange line, %, bottom axis) and in the sampled sediments, by loss-on-ignition. e. Standard deviation in the fine fraction (<2mm) by laser particle size analysis. The extent to which the sampled sediments are dominated by loess-like sedimentation (low standard deviation) or sheet erosion or soil development (higher standard deviation) may be a proxy for precipitation. f. Mean particle size of the silt fraction (µm, top axis). Coarser particles mean higher-energy deposition. g. Gravel content of the sampled sediments (weight %, bottom axis). Gravel represents gravitational input. Increased gravel content indicates longer surface exposure and thus slower sedimentation, and the distance between e and f indicates surface stability. h. Magnetic susceptibility (SI). Variation below the Holocene soil complex is difficult to explain, and may reflect the lithology of the local bedrock and its varying presence within the section.

Interpretation
The geomorphological situation The T16 site sits on a solifluction terrace, a relatively flat-topped landform with a steep scarp formed by the superposition of successive lobes of soliflucted material, and is considerably elevated above the floodplain of the Tolbor River. Although of Holocene origin, the present-day gully which dissects the landform follows the course of a drainageincised into the bedrockof considerable antiquity; before the Holocene a cone of sediments existed at its mouth. The material of the solifluction terrace seems to have been thickest by the opening of the drainage into the wider Tolbor valley and dips away from this point, and it is likely that the redeposited material of the sheet erosion deposits of unit 3 also moved along the drainage. Somewhere beneath these sediments lies a bajadavisible elsewhere along the valley and in roadcutsof which the gravels of unit 4 may mark the top. The rock-cut drainage, and others like it along the valley, probably supplied the coarse material for this earlier fan deposition as well. The geomorphological situation at T16 is important to the sedimentary and stratigraphic interpretation presented here because of the role of water in the deposition of unit 3. Wateras an agent of transport and in the form of soil moisture -is the cause of the sediments' laminarity, which is described in more detail below. The position of the sections adjacent to and evenin the lower excavated steps of Pit 4within the drainage should probably be borne in mind when considering the hydrological regime at the time of site formation.
The loess and loess-like sediments of units 1 and 2 at Pit 4 Particle size analyses confirm eolian sedimentation as a dominant process during the time periods represented by the deposition of units 1 and 2. Even in the Holocene soil of unit 1 the organic matter content is relatively low, and the particle size distribution curve is consistent with loess (possibly with some removal of the very fine fraction by washing). The explanation for this is that Holocene gully incision truncates the loess deposit of unit 2 (substantially thicker in Pit 1 just a few meters away), and a relatively thin, weak soil forms at the surface. Unit 2 contains sparse gravel (increasing with depth). Particle size distribution curves (figure S7c), while still indicating the dominance of eolian sedimentation, suggest that the role of colluviation increases with depth in the sediments of unit 2. The contact with unit 3 is erosional and, in the sampled section, thick owing to solifluction. A slope change is evident from the dip of the sediments; loess accumulation against the hillside likely resulted in a steeper profile. The development of a weak soil (samples 10-13) and its subsequent reworking by solifluction (following this new, steeper slope) result in local truncation of the underlying sediments and the admixture of material from unit 3.
The textural laminae of unit 3 at Pit 4 The laminar silt of unit 3 is very poorly sorted, and clay and very fine silt are included in the same deposit as cobbles and even boulders. In thin section (figure S8), the mineralogy of the coarse (>approx. 100µm) and fine fractions within the laminar sediment are different, with the former comprising fragments of the local igneous bedrock, and the latter dominated by silt-sized grains of quartz, which represent eolian sedimentation (loess); a loess parent material for the sediments is likewise indicated by the predominance of silt within the section (figure S7b and c). The mineral fraction of the laminar sediment has therefore at least two sources (one proximal and one distal). Within unit 3 the quartz silt is mostly incorporated into inhomogeneous aggregates of more and less humic material ( Figure S8a and b). Sand stringers (again of the local bedrock; Figure S8a and b) and fragments of the calcium carbonate pendants which form under larger clasts are all suggestive of reworking of the sediments, particularly by flowing water. With a few exceptions, most of the samples from unit 3 show particle size distribution curves (figure S7c) consistent with reworking (more gently peaked curves, multiple peaks) or soil development (enhanced clay content). Although loess is the parent material, the dominant depositional process within unit 3 seems to be sheet erosion.
Gravel, cobbles, and boulders are too big to have been transported in what appears to have been a low energy reworking. Some of the sand and gravel seem to have actually been generated post-depositionally by frost action and mechanical weathering during solifluction (larger cobbles are physically crushed to lenses of gravel), but the arrival of the larger clasts within a sandy silt matrix is presumably due to gravitational input from the bedrock outcrops just upslope. The coarser clasts are not uniformly distributed through the section; instead particular layers seem to play host to lines of cobbles or gravel. Although the bedrock source may have experienced climate-dependent changes in the rate of clast production (likely through frost shatter), in general the presence of larger clasts seems to indicate a decrease in the rate of sedimentation (surface stabilization, reduced colluviation and sheet erosion) resulting in longer surface exposure. Lines of cobbles sit at the surfaces of some the solifluction lobes identified in the section drawing; the gravel content of the sampled sediments (figures S5g and S6g) is here used as a proxy for sedimentation rate, with lower gravel contentbecause of the multiple sedimentary processes involved in the deposition of the studied sedimentsindicating more rapid burial, and higher gravel contents indicating prolonged surface exposure and slower sedimentation (an exception is unit 4 at pit 4, the gravel deposit). The mean silt particle size (figures S5f and S6f) varies inversely with gravel content, indicating two separate depositional processes for the two size fractions.
The compositional laminae of unit 3 at Pit 4 The color variation within the laminar deposit, from dark brown to white, is due to variations in calcium carbonate content and, to a lesser extent, organic matter (figure S8c and d). While the source of the calcium carbonate within the sediments is probably loess, certain laminae within unit 3 have carbonate contents well above that of the primary loess of unit 2. There doesn't seem to be a very strong relationship between the particle size distribution of the fine fraction and calcium carbonate content, and these two variables probably reflect different processes; increased carbonate content is not simply a product of loess accumulation. In thin section, calcium carbonate enrichment within lighter laminae includes the precipitation of calcium carbonate pendants on lithics and coarser clasts (these are almost always on the bottom), as well as micritic calcite precipitation on the groundmass of aggregates, and is therefore likely post-depositional. There is some evidence that the enrichment forms at or very near to the ground surface (pronounced white laminae frequently co-occur with lines of cobbles, and often display higher organic matter contents than the over-and underlying sediments). The calcium carbonate enrichment is laminar in character but does not seem to form nodules or coat smaller grains in thin section as is usually expected for a k-horizon. Apart from disaggregated grains of root cell calcite, which can be abundant, calcium carbonate does not seem to be related to roots, pores, or voids (although, since these are subsequently affected by frost action, which merges them to a lenticular microstructure, it can be hard to tell); it exists as micrite precipitated directly on the groundmass from the soil solution. Its finely laminar character, alternating with less carbonate-rich laminae, is most consistent with crust formation. In the sedimentary regime represented by most of unit 3, crusts are buried by colluvium, creating compositional laminae. Figure S7. a. The sampled section. b. Proportion of clay (blue), silt (orange), and sand (grey) in the sampled sediments. c. Particle size distribution curves for each of the sampled sediments. The best-sorted sediments (sharply peaked) represent primary loess or reworked loess. Although loess is the parent material for most of the sampled sediments, secondary sorting events (reworking, colluviation, redeposition by flowing water) or weathering and soil development (the breakdown of coarser silt particles, clay enrichment) change the shape of the curves. Blue shading is intended to distinguish stratigraphic units.
The question of what climatic situation is represented by the combination of calcium carbonate enrichment, stone or gravel lines, and increased organic matter is important, since it relates to the sediments which house AH6. The enhanced organic matter content alone is indicative of soil formation, and it looks like the reworked soil material in unit 3 shows various degrees of soil development through the section according to the climate at the time of pedogenesis. The repeated episodes of sheet erosion which form unit 3 are linked with precipitation, and the alternation of sheet erosion and the development of thin surface crusts of evaporites (the fact that strong evaporation frequently follows precipitation) tells us something about the seasonality of that precipitation. Either the very thick white laminae at the surfaces of solifluction lobes 3c,d, and e form during soil development, growing more pronounced as colluviation slows with longerterm surface stability (and therefore the soil itself, which in some cases is even more strongly humic than the present one, reflects hydrological conditions very different than the present-day), or they form in a cold arid period just after the end of soil development. One possible scenario is a moist active layer sitting over permafrost or seasonally frozen ground, with arid conditions in particular seasons bringing about strong evaporation.

The climatic record
The variations in organic matter and carbonate content, sedimentation rate (for which gravel content is used here as a proxy), and in the predominance of high and low energy processes and eolian or water-mediated sedimentation in the deposition of the section can be put together into a climatic record for the sediments of Pit 4. This record is augmentedbut also disruptedby the episodes of solifluction which took place intermittently until the initiation of loess deposition in unit 2. We should probably assume a lacuna where solifluction lobes meet, and there may be some repetition or thickening of the sediments; owing to the small area of exposure below 3c, it is difficult to say whether 3d is a repetition of 3e, and within the column of samples it appears that top of lobe 3c is caught up in the base of solifluction lobe 3b, locally doubling the sediment. Lobe 3a is not sampled (see figures S3 and S4), but radiocarbon dates make it the same age as 3c. Finally, the solifluction event at the base of unit 2 truncates unit 3 (probably to a different depth than at Pit 1), which means that some of the climatic record is missing. These are the interpretive challenges the section presents, but bearing them in mind, it nevertheless preserves a fairly high-resolution climatic record.
Carbonate content is generally lower in the bottom half of the section (as yet undated). Together with particle size distribution curves consistent with sorting by flowing water (figure S7c), and relatively low organic matter contents, we can infer cool and wet conditions during the deposition of unit 4 and most of lobe 3e (bearing in mind that because the trench was stepped, these samples are closer to the mouth of the bedrock drainage which supplied sediment and moisture). Within unit 3, precipitation (as evidenced by sheet erosion; figure S7c) seems roughly to covary with organic matter content; areas of clay enrichment may reflect the deposition of reworked soils or in situ clay development and translocation. Both organic matter and carbonate content peak within these sediments; the combination of sheet erosion, soil development, high organic matter content, and high carbonate content with crust formation collectively suggest a different hydrological regime than today. Changes in the seasonality of precipitation may be invoked to account for the covariation of seemingly opposed processes and constituents. While organic matter content declines above lobe 3c, carbonate content remains generally high, suggesting overall cooler and more arid conditions; a change from sheet erosion to eolian deposition takes place above lobe 3b in the sampled section.
The climatic cyclicity implied by repeated episodes of solifluction is confirmed by other proxiesapparent warm events in sub-units 3c, d, and e see increased organic matter content, sheet erosion, evidence for soil development including increased clay content, and a reduction in the overall rate of sedimentation, followed by climatic deterioration and a change toward eolian sedimentation. Despite possible gaps and repetitions in the record, the sediments within unit 3 seem broadly to confirm a series of (at least two) warmer climatic episodes interspersed with relatively serious deteriorations. That a climatic deterioration takes place at the top of sub-unit 3c, which houses AH6, is supported by a change to loess-type sedimentation and a reduction in organic matter content in the uppermost part of this lobe. Regardless of any hiatus or doubling between the top of 3c and the overlying sub-unit 3b, we can associate this climate deterioration chronologically with end of the occupations of AH6. Aside from solifluction, evidence for this climate crash at Pit 1 includes the formation of a thick white (calcium carbonate-rich) lamina, most likely connected with arid conditions and a reduction in the number of heavy rainfall (sheet erosion) events.

Luminescence Dating
The chronologic framework for human occupation at Tolbor 16, and for contextual landscape change in the surrounding valley, was determined by both radiocarbon and luminescence techniques. 16 luminescence samples were collected from Tolbor.
Luminescence dating determines the antiquity of the most recent exposure of sediment grains to sunlight, and hence the depositional age (10). Luminescence dating is a highly suitable method for determining the timing of human activity at archaeological sites such as Tolbor 16, where archaeological traces are clearly directly associated with the sediments in which they are found. Quartz is generally the preferred material for luminescence dating due to its more stable luminescence signal and negligible internal dosimetry; finegrained wind-blown quartz has successfully been used for dating in comparable archaeological sites in the Eurasian landscape (11,12). However, only 12 of the 16 samples collected yielded sufficient quartz for dating with luminescence. For these samples, both quartz and polymineral (feldspar-dominated) aliquots were measured for independent age determination and comparison. For all polymineral samples we used the post-infrared-infrared (pIR-IR) protocol at elevated temperature (290°C), which overcomes the problems of time-dependent fading of the feldspar luminescence signal (13). These protocols have been shown to be accurate with independent age control in Eurasian loess contexts (11,14).

Sample collection
16 luminescence samples were collected from the Tolbor 16 profiles in Pits 1, 2 and 3 (Table S1, Fig. S9). Sampling was undertaken by driving stainless steel tubes horizontally into the cleaned profile. Approximately 300 g of additional sediment was collected from the same position as the luminescence samples, for analyses of moisture content and laboratory dosimetry measurements using gamma spectrometry (Fig. S9A). The sampling procedure aimed at getting ages for the three main sedimentary units (Unit1-3) and check for their consistency.   (14)(15)(16). The finegrained polymineral (feldspar-bearing) fraction (4-11 µm) was first isolated from other components of the sediment for measurement of the equivalent dose (De). This involved sediment settling to obtain the preferred grain-size fraction, removal of carbonates by digestion in HCl and of organic matter by digestion in H2O2. Fine-grained quartz was extracted from subsamples of the 4-11 µm polymineral fraction by etching in fluorosilicic acid, however not all samples yielded sufficient quartz for measurement following this process and since no further sample could be spared, only the polymineral fraction was measured for those samples. For each sample, 18 aliquots of 1cm diameter were prepared by pipette in water solution onto stainless steel discs for measurement.
Equivalent dose (De) measurements were undertaken using an automated Risø TL-DA-20 reader equipped with infrared light-emitting diodes for light stimulation of single aliquots (17), and with U340 and D410 filters inserted in front of the photomultiplier tube to detect the quartz and feldspar luminescence emissions respectively. Irradiation was provided by calibrated 90 Sr/ 90 Y beta sources (17).
Quartz aliquots were measured using the single aliquot regenerative dose (SAR) protocol of Murray and Wintle (18,19), including an additional repeat dose at the end of the protocol which provided an infrared depletion ratio as a means to assess potential contamination of the quartz OSL signal by feldspar. Preheat tests were conducted on three samples (Fig. S10) and indicated optimal preheat and cut heat temperatures of 220ºC and 200ºC respectively. Since the calculated dose distributions for each set of aliquots yielded Gaussian distributions (Fig. S11), the Central Age Model (20) was used to calculate the De for each sample. The pIR-IR290 protocol of Buylaert et al. (13) was applied to 18 polymineral aliquots of each sample, since higher preheat and measurement temperatures have been shown to reduce feldspar signal fading in loess (13; 21-23.

Luminescence characteristics
Both quartz and feldspar-bearing polymineral fine-grained aliquots yield very bright luminescence signals, which decay at rates typical for quartz and feldspar respectively (Fig. S11). The samples record variable, although acceptable, sensitivity change throughout the regenerative dose steps of the measurement protocol, remaining below a magnitude of two ( Figure S12).   The dose-response curves of all samples could be fit to a single saturating exponential or exponential-pluslinear function ( Figure S13). Almost all aliquots, both quartz and polymineral, passed selection criteria for equivalent dose determination (Table S2). Mean recycling ratios lie within 5% of unity and overdispersion across dose distributions lies below 20% for all samples (and below 15% for all but one sample).  Dose recovery ratios were calculated for representative quartz sample L-EVA-1430 (Unit 1) and two representative polymineral samples, L-EVA1433 (Unit 2-3) and L-EVA1438 (Unit 4-3). All fall within 10% of unity, indicating acceptable recovery of dose and suitability for dating ( Figures S14 and S15).

Dose rate calculations
Dose rates were determined based on high resolution germanium gamma spectrometric analysis of the radioactivity of uranium, thorium, potassium, and their daughter isotopes, undertaken on the bulk sediment samples at the "Felsenkeller" laboratory (VKTA) in Dresden. Gamma dose rates were calculated using published attenuation factors (24). The beta component of the dose rate was measured using an in-house beta counter from Risø. Dose rate attenuation by moisture was accounted for using average measured water content values of each sample, and incorporating an uncertainty value reflecting the saturation potential of the sediments based on laboratory measurements (10 ± 5%). The cosmic ray component of the dose rate was calculated based on published formulae (25).

Age calculations
The age calculations are summarized in Table S3. Alpha-values of 4 ± 2% and 8 ± 2% (26,27) were used to correct for internal dosimetry within the quartz and polymineral fine grains, respectively. The quartz and polymineral ages generally lie within 2σ error of one another, which supports their accuracy. In one instance, the polymineral age differed from the quartz age by more than 2σ (sample 3071/ L-EVA-1439, Pit 3). In this case, the quartz age was assumed to be the more likely accurate value. However, since this sample was collected from an archaeologically sterile unit and falls into stratigraphic order, the age of this sample has little bearing on our overall interpretation of the chronology of occupation of the Tolbor 16 site.

Samples selection and location
In Tolbor-16, most of the cortical surface of the bones is gone. Hence, it was not possible to identify anthropogenic or animal post-depositional modifications (e.g. cut marks). Overall, 20 % (N=13) of the bone fragments larger than 2cm have been dated. Our selection of samples was limited by the preservation issues and by the complexity of the stratigraphy. In Pit 4, MAMS-31815 was collected at the interface of unit 2 and unit 3. Although there are possible intrusions of the material from unit 3 (Figure 1), we note that the age is consistent with the dates obtained on samples MAMS-24089 and MAMS-24090 but also with OSL age-estimates for the top of 3a in Pit 1 and the unit 2/3 interface in Pit 2 and Pit 3 (SI3). MAMS-240873 is associated with 3b and at 95.4% confidence interval; it is consistent with OSL measurements from unit 3 polymineral samples. The same applies to the rest of the samples collected in solifluction lobe 3c. The ages obtained for the Pit 4 -unit 3c (archaeological component referred to as IUP) is further confirmed by the sample MAMS-20981, obtained in unit 3c in Pit 1. It is associated with the same archaeological material stratigraphically below a thick white lamina.

Pre-treatment and dating
Samples were selected for dating when showing a satisfying amount of collagen yield (>1%) (Ambrose, 1990;Weber et al., 2005;Hublin et al., 2012) and C:N between 2.9 and 3.5 (Klinken, 1999). The samples were pretreated at MPI-EVA Leipzig using the method described in Talamo and Richards (2011). Approximately 500 mg of samples were first cleaned and then demineralized in 0.5 M HCl at room temperature until no CO2 effervescence could be observed. 0.1 M NaOH were then added for 30 min to remove humics. The NaOH step was followed by further rinsing with 0.5 M HCl for 15 min. The sample was then gelatinized, following Longin (1971), in a pH3 solution at 75 C for 20 h. The resulting gelatin was first filtered in an Eeze-Filter™ (Elkay Laboratory Products (UK) Ltd.) to remove small (<8 mm) particles and then through a 30 kDa ultrafilter (Sartorius "Vivaspin 15") (Brown et al., 1988). Prior to use, the filter was cleaned to remove carbon containing humectants (Brock et al., 2007). The sample was then lyophilized for 48 h. The selected samples were dated by AMS at the Klaus-Tschira-Labor für Physikalische Altersbestimmung (Curt-Engelhorn-Zentrums für Arch€aometrie), Mannheim, Germany (Kromer et al., 2013).   Although all geological evidence points to low-energy deposition, slopewash, shearing or doubling of sediments may have altered the spatial organization of the remains. Erosion and slope wash may have led to depleted occupation horizons or alternatively, post-depositional processes may artificially inflate the thickness of the archaeological layer (time averaging). Such a situation is best illustrated by the eastern end of the north section where the assemblage at the interface of sedimentary units 2 and 3 sits directly on the top of the assemblage of unit 3c (Fig. S23). Because the solifluction challenges the laws of superposition, we sampled the lithic material that is clearly attributed to solifluction lobe 3c. AH6 corresponds to the lowest archaeological deposit in an excavated surface of 15 m 2 . To draw the upper limit of lobe 3c, we used observations made during the excavations and projection of the piece-plotted material on to the western and northern walls.
In the western wall, lobes 3a and 3c dip northward; hence, vertical projections on a wall from a greater distance than 1m are likely to be inaccurate. Instead, we used 3D projections of pieceplotted artefacts (<2cm) on a 3D model of Pit 4 generated by photogrammetry. Fig. S24A illustrates the distribution of the AH6 and the distribution of AH1-AH5. Fig. S24B shows that bone distribution is in close association with AH6, artefact from AH6 and the sediment bulk samples. Note that bone fragments are in close association with AH6 and AH5. Figure S24: 3D models of the Western and Northern wall on Pit 4 with projection of piece-plotted objects (<2cm, here excluding the two-points). A. Piece-plotted artefacts from AH1-AH5 (light blue) and the sample studied here, AH6 (dark blue) in solifluction lobe c, Unit 3. B. Piece-plotted artefacts from AH6 (dark blue), piece plotted bones (yellow) and bulk sediment samples (red). Geo-localized 3D photogrammetry in Agisoft Photoscan, data collected in New Plot (8,9) 3D plot generated in ISTI-CNR MeshLab.

Fabric analyses
We performed a standard fabric analysis (39,40). 'Orientation of clast' refers to an alignment of an elongated object in three-dimensional space in terms of horizontal and vertical displacement. 'Bearing' refers to a horizontal angle of object's long axis relative to the arbitrary north. 'Plunge' or dip is a vertical angle relative to a horizontal plane. Rose and Schmidt diagrams best represent these data (Fig. S25). Statistical procedure for orientation data applies uniformity tests for bearing and plunge values. Bearing data are captured in 360 degrees so the use of circular instead of linear statistics is more appropriate (e.g. Rayleigh test of uniformity). Plunge, on the other hand, follows regular statistical testing (e.g. t-test). If uniformity is confirmed for bearing angles, a preferential orientation of clasts is indicated suggesting post-depositional processes that could have aligned the objects. On the other hand, a non-random pattern of plunge values suggest that object dip angles correspond to the surface on which they were deposited and didn't go through significant disturbance. Figure S25: Schmidt diagram and plunge for Unit 2 (above) and Unit 3c (below) Results are presented on a Benn diagram (41) that considers both bearing and plunge angles simultaneously. Following Benn's method, Lenoble and Bertran (40) provided a comparative data from experimentally and archaeologically derived data for debris flow, runoff on steep and shallow slopes, and solifluction. The range of Benn values for these processes are used as a reference. Figure S26. Benn-diagram for unit 3 (left) and unit 2 (right).
We recorded two points for elongated with the total station (42). Due to scarcity of bone remains and natural clasts, two points are mainly recorded for elongated lithic artefacts (blades). Pit 4 provided a sample of 83 objects for unit 3 and 45 objects for unit 2. To reach an acceptable sample size (ca. 50) (39,43,44) in Pit 4, we considered Unit 3 as one single unit (lumping 3a, 3b and 3c) and we used Unit 2 as a control sample. Analyses and figures have been produced in R (45) using the protocols and the codes described by McPherron (44).
Unit 3 shows overlap with three processes: runoff, both shallow and steep, and solifluction, while unit 2 plots in a planar portion of Benn diagram ( Figure S26). Permutation test is used to assess whether two assemblages have different orientations. This test allows for assessing and analyzing orientations in three-dimensional space (treated together as a vector that describes the orientation of an object) rather than bearing and plunge aspects separately (44). The test assesses the probability that two assemblages come from the same population (the null that they have the same orientation cannot be rejected). Permutation test between Benn ratios of unit 3 and unit 2, based on 10000 resampling, shows that we cannot reject the null hypothesis (p = .92) that the samples are the same. This result likely reflects the overlap between confidence intervals in the planar range of the diagram. Overall, the results are in line with the geological descriptions (SI section 2) that noted the role of run-off in the formation of unit 3 and identified post-depositional events such as solifluction (3a-3c). We interpret the large standard deviation from Unit 2 as reflecting the impact of an interface unit 2/unit 3 (which features solifluction and intrusion of material from unit 3 into unit 2) on the overall fabric signatures. We also note that by lumping together the 3 solifluction lobes into one unit, we may have increased the standard deviation of unit 3, hence resulting in the slight overlap observed between the two units.

Technological description
The lithic assemblage is mainly produced on a medium-to fine-grain, local, cryptocrystalline raw material for which the primary sources occur in the whole valley as sub-vertical uplifted bands. Eroded clasts of this material along streams and canyons around the site, or buried under alluvium along the Tolbor River, constitute a possible secondary source. Although the use of fine-grained chert, flint or jasper-like material (possibly exotic) is reported during the later phases of the Upper Palaeolithic, it is absent in the coarse fraction of the AH6 sample (>2cm). A light blue patina, semiextensive with root-like patterns, alters most of the archaeological material. On the opposite side of the artefacts, we note the frequent occurrence of calcium-carbonate concretions usually located on the lower face (in contact with the sediments). This phenomenon is consistent with the macroand microscopic geological observations and supports the idea that the concretion formed by ground water exposure after the re-deposition by solifluction. Among piece-plotted blanks, flake and blade frequencies are balanced, with blades being slightly better represented (N=107; 55%). When calculating a minimum number of individuals (MNI= sum of platform blanks), blades frequency decreases sharply (N=42; 36%). The same way, retouched blades are prominent within the total sample (N=59; 65%) but less so in the MNI (N=25; 37% of flakes). The core to blank and tools ratio is 1/15, but the latter is strongly affected by the fragmentation of blades. Considering the MNIs for blades and tools on blades drops the ratio to 1/5. Retouched flake frequencies are low in the whole sample (N=32; 36%) but high in the MNI (N=21; 65%) reflecting the difficulty to differentiate flake fragments from shatters. The flake category is mainly composed of platform and complete flakes (N=74; 88%). Blades, retouched or not, are represented by 90% of fragments. The general structure of the blade assemblage is presented in Table S6; we highlight several features relevant to the present case. First, the blade dorsal faces include both unidirectional and bidirectional patterns. Explaining the variations in frequencies between these two categories is beyond the scope of this paper, but it is notable that the high degree of fragmentation influences the visibility of the dorsal patterns; hence there are numerous undetermined specimens. In general terms, the co-existence of bi-and unidirectional dorsal patterns suggests a core reduction system involving two-opposed platforms and alternate short sequences of flaking (Table S7). When the whole sample is considered, bidirectional blades are particularly well-represented among the retouched tools. Second, the good representation of crested blades demonstrates that this method is part of the behavioural repertoire. It also underlines efforts to produce blades as opposed to an opportunistic production and illustrates core-management operations. The lack of initial crests as opposed to the second crests could indicate the import of semi-prepared cores from an unknown distance (arguably, the initial flaking could occur not further than a few meters away). This corroborates the low frequency of primary or secondary cortex on both blade and flake blanks.
Another observation is that all categories of blades are retouchedincluding technical elements such as crested blades. Looking at the advanced stage of exhaustion of blade cores, a ratio of 1/5 is too low, and part of the blade production might be missing. The frequent use of technical pieces as retouched tools corroborates the idea of an expedient use.
The platform preparations show substantial variations between object categories (Table S8). For blades, plain, facetted and dihedral convex platforms are the most common. With retouched blades, dihedral and faceting preparations are the most frequent whereas plain platforms rise among unretouched blades. With flakes have very little faceting and plain, or unprepared platforms are the most frequents. This is consistent with the little investment observed in the flake production and a more careful preparation of blade platforms, some of which will be retouched into formal tools. The external platform angle (EPA) is prepared using various methods but notable are the marginal faceting and the pecking (Fig. S28-2), that occur in high frequencies in the Initial Upper Palaeolithic (IUP) from the Altai and the Baikal region (46,47). Defined respectively as a faceting or a hammering of the platform edge, it is often combined with a prominent plain or dihedral platform. It mostly consists in preparation of the EPA (and not the platform per se) and therefore in Table  S8, it is grouped here with the plain or dihedral platforms. Most platform thickness means are above 4mm, which for blades, would fall into the expected range of direct percussion with mineral hammer (Table S9) (48). Considering the large standard deviations and the lack of controlled experiments on the specifics of the raw material, direct percussion with hard hammer is not exclusive of the use of other techniques (e.g. soft stone hammer). The platform thickness of the retouched blades stands out as the thickest in the sample. Figure S27: AH6, blade core reduction. Asymmetrical reduction pattern (A1) that produces large-medium size blades (Ab; Fig. S28-2, 3 and 5) and thick debordant/crests blades (Ac, Fig. S28-4). The latter (Ac) are snapped and turned into Burin-Cores (B1, B2, B4) or more rarely into truncated-facetted (B3). Mode A cores (Fig. S27Aa) are reduced on both broad and narrow face (Fig. S27Ab) (Fig. S28), with neo-crest or thick debordant blades removed to manage lateral convexities (Fig. S27Ac). Mode A illustrates the production medium to large blade following a specific asymmetric pattern (49)(50)(51). These blades are of three different kinds based of their overall contour: with convergent ( Fig. S30-2,3), sub-parallel (Fig. S30-5) and parallel edges (Fig. S30-2). Some of the thickest blades are further reduced using the Burin-Core method as defined by Zwyns et al. (52) (Fig.S29) 2). The latter show a flaking surface parallel to the long axis of the blade. Mode A cores are exhausted in different ways and at the end of the reduction process, they can either produce blades or flakes. It is worth noting that the method is relatively systematic and does not deviate until the end of the process. No specific production of flakes can be identified based on the cores and the production seems to be primarily oriented toward the production of blades. Half of the tools are retouched blades and retouched flakes and overall, there is a lack of formal tools. Among the standard types, notch and denticulate are the most numerous. The presence of Upper Palaeolithic tool types such as endscrapers and truncation is noted but except for a single dihedral burin, classic burin types are absent from the studied sample. This typological signature is typical of the IUP but it is notable that tools on flakes (including Middle Palaeolithic types such as notch and denticulate) are not produced on standardized blanks. Instead, they appear to be linked with management operations of Mode A blade core. These include core tablets and crest-shaping flakes. Anecdotally, the toolkit includes a small fragmented bifacial tool/preform. The assemblage AH6 from the Unit 3c is a blade-based assemblage. The main characters of the blade production fall into the definition of the Upper Palaeolithic (sub-volumetric reduction and volumetric reduction of the core, reliance on crest method, management of striking platform by core tablet removal). The technique used to produce blades likely involves a mineral hammer although this aspect of the blade technology would require further attention. Careful faceting of platforms and bevel-shaped EPA (marginal faceting) indicate an interest to produce robust blades whereas the alternate bidirectional system seems well adapted to the production of straight blanks. Mode A core shows a specific conception of the volume (50) described here as asymmetrical reduction.
The thickest blanks are selected for two distinct purposes. First, they are used as tools with expedient modifications. The selection of technical elements and the lack of formal blades, the exhaustion of the Mode A cores are elements indicative of a specific economic behaviour. Second, the thickest blades are further reduced using the Burin-Core method, or as truncated-facetted. There is no evidence for a specific production of flakes except for the exhausted blade cores (same reduction method, smaller size).

5.4
Relevance for the regional record This exceptional combination of features assigns these assemblages to a united Initial Upper Palaeolithic (IUP) techno-complex and not only to a generic Upper Palaeolithic classification. Both Mode A and Mode B methods are documented in the Altai and in the Baikal regions and are typical of the Asian IUP (48,51,53) but seem absent from the European variants such as the Bohunician (54,55).
In the Altai, the IUP has been recognized at Ust-Karakol 1 (sector 1, OH5.5 and OH5.4), Kara-Tenesh, Maloyalomanskaya Cave (for a review see the Kara-Bom variant in 56). It can be traced eastward in the Cis-Baikal (e.g., Makarovo-4) (53, 57) Trans-Baikal (e.g. Khotyk, Kamenka A and C and Podzvonkaya) (58-60) Mongolia (e.g. Tolbor 4) (3, 6, 61) and China (Shuiddongou) (62)(63)(64)(65). The IUP complex also marks the first appearance of the use of pigments, body ornaments, formal bone tools and even musical instruments in Siberia (53,66). The quantity and quality of chronological data associated with the IUP vary depending on the region (for a review of the labs and methods see 67). Large standard deviation and occasional infinite ages are frequently associated with estimation close to the limit of the method. Younger results are not reported here as they may reflect differences between measurements (Conventional/AMS) or pre-treatment methods, contaminations by younger material. They could also indicate that in some regions the IUP may have lasted a few thousands of years.