Supporting Online Materials A. Excavation history and site summary for the FAY-NE1 rockshelter at Jebel Faya, Sharjah, UAE. The site FAY-NE1 (Fig. S1) at the north-eastern end of Jebel Faya in the Central Region of the Emirate of Sharjah (UAE) has been excavated by a Joint Project between the Directorate of Antiquities of the Department of Culture and Information, Government of Sharjah (UAE) and the Institute of Early Prehistory of the University of T?bingen (Germany) since 2003. Paleolithic levels were first reached in 2006. They have since been discovered in several excavation trenches covering an area of more than 150 m2. FAY-NE1 is located in an ecotone-situation with access to the comparatively well vegetated gravel-plain in the east, the steep and deeply incised slopes of Jebel Faya in the south-west, a paleolake basin to the north and the sand sea of the Rub' al Khali to the north-west. Water was available for at least part of the year from potholes and small temporary springs along the slopes of Jebel Faya. These diverse landscapes provided habitats for a variety of edible plants and animals. The site itself is a rockshelter (Fig. S2), large enough for several families, which opens to the east. Together with the cliff to the south it provides shade from late morning through the rest of the day. Being sheltered in the mouth of a valley, neither the dust-storms from the south nor the cold northerly winds have a direct impact on the site. An important feature of the site relates to a steep gully, collecting the runoff from the adjacent mountain-slopes, which debouches over the cliff just south of the rockshelter (Fig. S1). A dejection-cone accumulation has built up in front of the cliff, which also extends north in front of the overhang. Large eboulis blocks fallen from the cliff and the roof of the shelter hold these sediments in place (Fig. S2). They consist of rocky debris in a sandy matrix. Inside the rockshelter and behind a barrier of large fallen blocks there is a more fine-grained fill at the bottom of the sheltered cavity consisting of aeolian materials and small splinters of the limestone into which the site is carved. The sedimentary conditions have led to the accumulation of the stratified sequence including the lithic occupation debris. In 2003 the first exploration trench was cut into the talus sediments immediately in front of the rockshelter, where some pottery and other artifacts from the early historic Mleiha period were found just below the surface, followed in stratigraphic order by Iron Age and Bronze Age artifacts. Neolithic stone implements occur from a depth of c. 50 cm downwards. They indicated stratification within the Neolithic period, a rare regional occurrence. In 2004 and 2005, larger trenches were opened and again enlarged, concentrating on the Neolithic levels. In 2006 pre-Neolithic levels were reached. Paleolithic artifacts were separated by 25-100 cm of sterile sediments from the overlaying Neolithic sediments. In 2007 the barrier of fallen blocks under the former drip-line of the shelter was cut, connecting the earlier exploration trenches inside and outside the shelter. The result was a continuous section of almost 24 m length (Tr1, Tr4, Tr27, Tr19, Fig. S3, Fig. S4). In 2008 bedrock was reached inside the shelter and below most of the rockfall in front of it. An area of heavily cemented rock-debris, probably under earlier drip-lines of the overhang, could not be excavated. While the higher levels of the rockshelter-sediments were comparatively poor in flints, the bottom part was rich in Paleolithic artifacts. North and south of the 24 m long transverse section, two wider trenches were excavated from 2005 onwards (Tr4, Tr28, Fig. S3). Together with the transverse section they covered an area of more than 150 m2 by the end of the 2008 campaign. Outside the rockshelter, Paleolithic levels were also reached both south and north of the main section. Fig S4 shows the distribution and varying density of stone artifacts in the main section, as well as the clear separation of the lithic assemblages. Fig. S4 shows a detailed section from the main trench where the lithics are in primary contexts. Neolithic artifacts (light blue dashes in Fig. S4) are well separated stratigraphically from the uppermost Paleolithic Assemblage A (green squares), which is also clearly separated from the deeper Assemblage B (blue-green triangles). Assemblage B and C (pink squares) are tightly superimposed and not easily separated stratigraphically, but are different in technology and typology. Assemblage C might be typologically related to the early East African MSA (S1). It includes bifacially worked flints, which are diagnostic and only found in this assemblage. Either side of this section Holocene gullying has disturbed the sediments. Three OSL dates were obtained from sediments in Tr19 containing Assemblage C lithics (see Section C) yielding dates of c. 94.8?13.0 ka (OSL Tr19-04), 123?10 ka (OSL Tr1909) and 127?16 (OSL Tr19-03). In Tr4 and Tr28, the Neolithic and Paleolithic Assemblage A are separated by sterile sands up to 40 cm thick. Two OSL dates from Assemblage A yielded ages of 40.2?3.0 ka (OSL Tr4-10) and 38.6?3.1 ka (OSL Tr28-08). The overlying sterile sands were dated to 38.6?3.2 (OSL Tr4-02) and 34.1?2.8 ka (OSL Tr28-06). An Early Neolithic component of the Holocene flint industries is found in the form of Fasad-points. Two radiocarbon dates were obtained for these layers on samples of marine shells (Turbo sp.) found c. 9 m apart from each other (Table S1). Both ages indicate an early (c. 10 ka) re-arrival of humans in the area after the end of the Pleistocene (S2). The dates also indicate a period of some 25,000 years without human presence at the site during late Marine Isotope Stage (MIS) 3 and MIS 2. The described artifact-bearing levels give physical evidence for human presence at FAYNE1 from MIS 5e to the present. An interruption in occupation of FAY-NE1 during late MIS 3/MIS 2 is indicated by the sterile sand layer overlying Assemblage A. Since southeastern Arabia experienced pronounced aridity during this period (Section D), it is probable that this hiatus represents an absence of humans from the entire region. A second hiatus in human occupation of FAY-NE1 is suggested by the sterile sediments separating Assemblages A and B. Although no absolute ages are available for either these sediments or those bearing Assemblage B, the site stratigraphy and regional climatic history make it likely that the sterile sediments were deposited during the relatively arid MIS 4. However, the lack of technological affinity between Assemblage A and contemporaneous assemblages from outside southern Arabia (Section B), makes it probable that this assemblage was produced by a population which survived the dry period in a local refugium, probably along the coast and watercourses of the largely dry Persian Gulf area, rather than by migrants re-occupying FAY-NE1 from more distant regions. Consequently, we suggest that the sterile sediments between Assemblages A and B represent the absence of human from FAY-NE1, but not from south-eastern Arabia as a whole. The excavations in spring 2010 added indications for an even longer sequence: The cemented layers become thicker towards the south of the site, where the blocks fallen from the former roof are firmly connected to the cemented sediments. Behind them, inside the shelter, are older archaeological layers with artifacts representing another technological tradition which still needs to be studied in detail. Thus, the site of FAY-NE1 holds a potential for further observations on human presence in this part of the world. B. Characterization of the Paleolithic Assemblages Assemblages C through A have sufficient samples to permit some technological and typological description. These assemblages are all either exclusively (A) or mostly (B and C) dominated by the use of local flints. Assemblage C has fewer than a dozen artifacts made on red chert, which is only available some 20 km to the east, as dike exposures. Assemblage B has fewer than six artifacts of the same material. Another shared trait is the variability in edge condition. Much post-depositional edge abrasion is evident, suggesting modest movement but many artifacts exhibit pristine edges. All Paleolithic artifacts at FAY-NE1 exhibit some degree of chemical weathering, often with white and highly brittle edges. Many artifacts are heavily desilicified. Present sample sizes for lithics clearly attributable to Assemblages C through A run from 500 in C to over 2,000 in B. Assemblage C: Technologically, this assemblage is characterized by three quite distinct reduction strategies: bifacial, Levallois and volumetric blade. Bifacial reduction by hard hammer is present on pre-forms, foliates, and small hand axes. The Levallois method is rare; a single core is not typical (Fig. 2), although the few Levallois flakes are well prepared. While no true blade cores have been recovered, there are two blades clearly struck from volumetric cores. Retouched tools include sidescrapers, endscrapers, denticulates, as well as bifacial foliates, small handaxes, and a range of bifacial pre-forms (Fig. 2 and S5). The number of bifacial pre-forms suggests this assemblage represents a specialized workshop, with little evidence for maintenance activities. Assemblage B: Technologically, this assemblage may be characterized by the production of flakes and a few blades, using a number of different reduction strategies. As reflected in the abandoned cores, blank production came from Kombewa cores, simple single unfaceted platformed cores, some volumetric, from 90? opposite side cores, from flat cores with converging to 90? removals on one core surface, as well as from rather rough radial cores (Fig. S6). The tendency toward convergent removals is clearly seen in blades with converging edges. A few examples might be called Levallois points, although no Levallois point cores have been recovered. Yet, convergent cores are not numerous. Platforms are mainly unfaceted but a reasonable number are either dihedral or multiple faceted, reflecting the faceting seen on the flat cores with convergent removals. Compared to Assemblage A, there is a greater range in the size of artifacts, with a number in excess of 100 mm, although flakes still average about 35 mm in greatest dimension. Retouched tools are not abundant but include well made sidescrapers, endscrapers and denticulates, as well as a few poor burins and truncated-facetted pieces (Fig. S6). There are no tools with backing and no evidence for bifacial retouch. Assemblage A: In the collection, parallel, Kombewa, convergent and radial cores are present (Fig. S7). Technologically, this assemblage is characterized by the production of flakes mainly from multiple platform, orthogonally shaped cores. The cores are abundant and reflect a consistent reduction strategy. Relatively small blocks of flint were used, with an initial flaking surface formed normally with an unfaceted platform perpendicular to the flaking surface. After the removal of some flakes, this flaking surface was used as a platform for the removal of additional flakes, again perpendicular to the old flaking surface. This new flaking surface may be formed at the distal end or along a lateral edge of the first flaking surface. The process was repeated from two to four times, so that the abandoned cores tend to have multiple flaking surfaces, each perpendicular to their platform. In all cases, the platforms are unfaceted. This is reflected in the flakes, which mainly have unfaceted platforms perpendicular to the dorsal surface of the flake. Bulbs of percussion are marked, there is no regularization of the striking platforms, and there is no evidence for soft hammer percussion. Owing to the repeated changing of the flaking surface, there are a good number of debordant flakes. Although there are some artifacts that are twice as long as they are wide, there is little evidence for purposeful blade or bladelet production. Both cores and flakes tend to be small. There are a few artifacts exceeding 70 mm in greatest dimension but the average artifact length is around 35 mm. In spite of a relatively large artifact sample, there are few retouched tools. Denticulates, endscrapers and a single burin are present (Fig. S7). Simple retouched pieces occur but there is no evidence for backing or bifacial retouch. The paucity of retouched tools and the relatively large number of cores indicates that the Assemblage A represents a primary workshop area. The FAY-NE1 assemblages provide a picture of what industries were present near the Gulf of Hormuz during the period of postulated African migration through the area. Deposition of Assemblage C began during the Last Interglacial. Technologically, this assemblage has general links to East Africa (S3 S4) while showing none of the technological traits characteristic in the contemporaneous Levantine Mousterian (S5). As in the early Middle Stone Age (MSA) of East Africa, Assemblage C exhibits three profoundly different reduction strategies: bifacial, volumetric blade, and radial Levallois. This combination is unknown in the Levant after about 200 ka, where there is no bifacial reduction and the Levallois method is largely limited to unidirectional converging. The latter produced large numbers of Levallois points, which are absent from Assemblage C. Although pre-AMH archaeological material is present in Arabia (S6), direct connections between Assemblage C and the Arabian Acheulean are highly unlikely, since the latter is not found in eastern Arabia, and it has disappeared in Africa and the Near East by c.500 ka (S7). Although Assemblage B has no absolute date, its stratigraphic position between the dated Assemblages C and A suggests an age between c.90 and c.40 ka. Although imprecise, this age range is broadly contemporaneous with the previously proposed period of migration out of Africa (S8, S9). This age range makes Assemblage B contemporaneous with the later MSA of East Africa, represented by such industries as those from upper Unit VI from Mumba Cave (S10) in Tanzania, and the Aduma sites in Ethiopia (S11), among others. While there is some regional variability in East Africa during the later Middle Stone Age (S12), all share broad technological and typological characteristics. Most notable is the dominance of the classic Levallois method for flake production (S10-12), the presence of both unifacial and bifacial points (S13-15), and moderate to low occurrences of bipolar reduction. Since these three characteristics are totally absent from Assemblage B, there appears to be little to connect it to any original base in East Africa. Similarly, the absence of a dominant wide Levallois point production (S16) and the presence of a mixture of Upper Paleolithic and Middle Paleolithic tool types in Assemblage B, make it quite distinct from the contemporaneous Late Levantine Mousterian (S16). Again, there is little to suggest a connection between the populations which produced Assemblage B and contemporaneous populations in the Levant. Levallois technology also dominates the Middle Paleolithic of Western Arabia, showing connections to the Levant (S17). Virtually nothing is known of contemporaneous southern Iranian industries, since no absolute dates are yet available for the recently excavated "Middle Paleolithic" sites (S18) but the descriptions of the recovered Iranian materials might suggest dates earlier than that of Assemblage B. Assemblage A is technologically totally unrelated to any published late MSA or early LSA industry. At 40 ka in East Africa, two different technological patterns have been described. The first combines three reduction strategies: Levallois, bipolar and single platform parallel (e.g. S10, S11, S19). The second essentially uses single platform cores to produce small flakes and blades, as well as having significant bipolar reduction. While there are a very few single platform cores in Assemblage A, there is not the slightest hint of the Levallois method, bipolar reduction, or purposeful bladelet production. Typologically, in East Africa, backed microliths and unifacial or bifacial points are characteristic almost everywhere (S12, S20). These tools are totally missing from Assemblage A and, as importantly, there is no indication of any use of backing in tool production. Assemblage A also shows no technological or typological affinities with contemporaneous industries in the Levant, where elegant blade/bladelet production typifies the Ahmarian (S21, S22) and tools are mainly made on long blades and bladelets using semi-steep retouch. Slightly later, Iran has assemblages that are Aurignacian in character, with carinated reduction typically resulting in minute twisted microblades (S23). Assemblage A shows no technological tendency toward carinated reduction, or the production of any microblades. The only similarities between Assemblage A and contemporaneous industries from adjacent regions lies in a range of generic tools, such as denticulates, endscrapers, and sidescrapers. Since these tool classes occur in virtually all late Pleistocene assemblages throughout the Old World, their presence in Assemblage A has little diagnostic value. C. Single-grain optically stimulated luminescence dating Optically stimulated luminescence dating (OSL) is an absolute dating technique which allows determination of the time elapsed since a sediment was last exposed to sunlight (S24, S25). OSL has been applied successfully to a wide range of archaeological and Quaternary deposits (S26-28). The method is based on the time dependant increase in the number of trapped electrons present in mineral grains (usually quartz or feldspar). These electrons become trapped due primarily to the effects of ionising radiation from naturally occurring radioisotopes within the sediment body (the 238U, 235U, 232Th decay series and also 40K) and to cosmic rays. On exposure to light in antiquity, electrons are released from their traps, resetting the trapped electron population to zero. Similarly, exposure to light in the laboratory (optical stimulation) releases the trapped electrons, a proportion of which dissipate energy as photons (luminescence). The intensity of this luminescence signal is proportional to the trapped electron population, which is in turn proportional to the total ionising radiation dose to which the mineral grain has been exposed. The equivalent dose (De), which is a laboratory estimate of the total radiation dose to which a mineral grain has been exposed since burial, is determined by measuring the OSL signal. Typically, the ionising radiation flux (dose rate, Dr) is assumed to be constant and hence the burial age of the mineral grain may be determined by dividing De by Dr. The resulting OSL age is obtained in calendar years prior to measurement. The datum used for calculating the ages reported here is 2010. All measurements, analysis and interpretation of data necessary to generate OSL ages were performed by SJA at Royal Holloway University of London, UK. C.1 Sampling strategy Ideally, a sampling strategy for an archaeological site would aim to bracket the earliest and latest evidence for each phase of occupation/lithic technology present. The nature of the sediments at FAY-NE1 precludes such an approach since most deposits are clast supported, with limited sandy matrix. Consequently, OSL samples were taken from the small number of sandy pockets present within sediments containing the assemblages A (two samples) and C (three samples). In addition, two OSL samples were extracted from an archaeologically sterile sand horizon overlying Assemblage A. Details of the association between the OSL samples analysed in this study and the artifact assemblages are presented in Table S2. The depositional processes responsible for the different sampled sediment bodies vary. The archaeologically sterile sand horizon overlying Assemblage A (samples Tr4-02 and Tr2806) is interpreted as aeolian in origin, due to its homogenous particle size distribution and lateral extent. However, aeolian bedding structures are not preserved in this sediment suggesting that some post-depositional mixing, possibly bioturbation, has occurred. We envisage a similar origin for the sediments sampled within Assemblage A (samples Tr4-10 and Tr28-08). The predominantly clast supported sediments containing Assemblage C are interpreted as local slope-wash deposits. It is not possible to determine whether the sandy pockets within these sediments were deposited during the slope wash events, or as subsequent aeolian infill of voids at the surface. In either case, deposition of the sand is effectively contemporaneous with deposition of the slope-wash, and hence OSL ages for these samples (Tr19-03, Tr19-04 and Tr19-09) represent the burial age of Assemblage A artifacts. C.2 Sample collection and preparation Samples were collected during the excavation of FAY-NE1 by hammering opaque plastic or metal tubes into cleaned section faces, or by scraping material into opaque plastic bags. In the latter case, the unit of interest was cleaned whilst shielded from sunlight by black tarpaulin. Fresh, unexposed sediment was then collected in an opaque plastic bag. Both sample types were subsequently sealed with adhesive tape and wrapped in two opaque plastic bags for transportation. In the laboratory, samples were processed under subdued orange light. Sunlight exposed material was removed from each end of the tube samples and retained for radioisotope concentration measurement (see Section C.3.f). Samples collected in opaque bags were also sub-sampled for radioisotope concentration measurements. Quartz was extracted from the portion of each sample which had not been exposed to sunlight since burial. Because the bedrock at FAY-NE1 is composed primarily of carbonate material, samples were initially wet-sieved to isolate the 212-180 um size fraction. This approach removed large bedrock clasts from the dating sample prior to acid treatment. Consequently, the risk of incorporating any unbleached grains liberated by dissolution of the bedrock was minimised. Carbonates and organic matter were subsequently removed from the 212-180 um fraction using 1M HCl and H2O2 respectively. The samples were then re-sieved at 180 um and quartz was extracted from the >180 um fraction using density separations at 2.62 and 2.70 g/cm3 and a subsequent HF acid etch (23M HF for 60 min followed by an 10M HCl rinse). Etched samples were again sieved at 180 um, to remove partially dissolved grains, and stored in opaque containers prior to measurement. C.3 Luminescence measurements C.3.a Equipment All OSL measurements presented in this study were carried out using a Ris? TL/OSL-DA15 automated dating system (S29), fitted with a single-grain OSL attachment (S30, S31). Optical stimulation of single aliquots was carried out using a blue (470 nm) light emitting diode (LED) array with a power density of 33 mW/cm2, while single-grains were stimulated using a 10 mW Nd : YVO4 solid-state diode-pumped green laser (532 nm) focussed to yield a nominal power density of 50 W/cm2 (S29). All infra-red (IR) stimulation was carried out using an IR (870 nm) laser diode array yielding a power density of 132 mW/cm2. OSL passed through 7.5 mm of Hoya U-340 filter and was detected using an Electron Tubes Ltd 9235QB15 photomultiplier tube. Irradiation was carried out using a 40 mCi 90Sr/90Y beta source giving ~6 Gy/min. This source is calibrated relative to the National Physical Laboratory, Teddington 60Co ?-source (Hotspot 800) (S32). Due to the spatial inhomogeneity of beta emitters across the active face of our 90Sr/90Y beta source, it was necessary to calibrate the dose rate to each individual grain position on a single-grain disc (S33). 4,800 grains of calibration quartz were measured, yielding 1,815 dose rate estimates (18?3 per grain position). Dose rates for individual grain positions differ from the mean single-grain disc calibration by up to 20%. All single-grain discs were placed into the Ris? instrument at the same angle relative to the 90Sr/90Y beta source. For the instrument used in this study, the sample discs rotated 0.04?0.03? per measurement cycle within the single-aliquot regenerative-dose method (i.e. 0.04? rotation from L1 to L2 etc, Section C.3.b). Consequently, it is possible to accurately correct for the effects of inhomogeneity in our 90Sr/90Y beta source by applying a grain position correction to each equivalent dose. Position-corrected equivalent doses were used in all age calculations. C.3.b The single-aliquot regenerative-dose method Equivalent doses were determined using the single-aliquot regenerative-dose (SAR) method (S34). This method allows an independent estimate of De to be generated for each aliquot or mineral grain measured, though in practice single-grains suffer a high rejection rate (Section C.3.c). The SAR technique involves making a series of paired measurements of OSL intensity. The first measurement in each pair gives the natural (Ln) or regenerated (Lx) OSL intensity, while the second measurement (Tx) gives the OSL intensity in response to a fixed test dose. Tx is used to monitor changes in sensitivity (luminescence per unit dose) during the measurement sequence. By dividing the natural or regenerated luminescence intensities by their respective test dose intensity (Ln/Tn or Lx/Tx), a sensitivity-corrected luminescence response is determined. The magnitude of sensitivity change during an SAR measurement sequence varies with measurement conditions, notably preheating regime, and also between samples (e.g. S35). While the SAR sensitivity correction often performs well across a wide range of preheating conditions (e.g. S35, S36), it is frequently observed that optimum conditions vary between samples (e.g. S34, S37). Consequently, the performance of the SAR method was tested prior to adopting measurement conditions for the dating study. The most robust test of the internal consistency of the SAR technique is the dose recovery test (S38, S39). Dose recovery experiments involve controlled bleaching of the natural OSL signal from a group of aliquots, followed by the application of a known laboratory dose. The equivalent dose for these aliquots is then determined using the SAR method. Where the applied dose and the measured equivalent dose (recovered dose) are indistinguishable, there are grounds for assuming that the measurement conditions used are appropriate for that sample. Dose recovery experiments were performed on five samples from FAY-NE1, using a range of eight commonly adopted preheating regimes (PH1 temperatures of 160-280 ?C at 20 ?C intervals, held for 10 s, all with a 160 ?C, 0 s PH2, and also a 260 ?C PH1 and a 220 ?C PH2, both held at temperature for 10 s). 24 aliquots of each sample were mounted on aluminium discs using Silkospray silicone oil applied via a 5mm mask. Three aliquots were measured using each of eight preheating regimes. Aliquots were bleached for 40 s at room temperature (c.20?C) using blue LEDs, followed by a 10 ks pause and a second 40 s bleach (S40). A beta dose similar to the natural dose was applied to bleached aliquots prior to measurement using the standard SAR method. Optical stimulation was carried out for 40 s at 125 ?C using a blue LED array. Aliquots were heated at 5 ?C/s during all heating steps, and a 10 s pause at 125 ?C was incorporated prior to optical stimulation to allow for thermal lag between the sample and heater plate. The OSL signal was that recorded during the first 0.32 s of stimulation, with a background signal from the last 4 s of stimulation subtracted. Several preheating regimes yielded measured:given dose ratios within two standard errors of unity for all samples, demonstrating the overall robustness of the SAR method. Of these regimes, we adopted a 260 ?C, 10 s PH1 and 220 ?C, 10 s PH2 for subsequent single-grain measurements (Table S3). All ages were produced using single-grain datasets. Measurement conditions, with the exception of regeneration doses, were held constant for all samples. Single-grains were preheated to 260 ?C for 10 s (PH1) prior to measurement of the natural or regenerated luminescence intensity, and to 220 ?C for 10 s (PH2) prior to measurement of the test dose (20 Gy) luminescence intensity. Optical stimulation was carried out at 125 ?C for 1 s using a green laser. The OSL signal was that recorded during the first 0.3 s of stimulation, with a background signal from the last 0.3 s of stimulation subtracted (S41, S42). Dose response curves were constructed by plotting sensitivity-corrected luminescence response (Lx/Tx) against regeneration dose. Regeneration doses were chosen to bracket the full range of expected equivalent dose values for each sample. A number of additional regeneration points were also included to monitor the quality of the data generated (Section C.3.c). These were: (1) a zero dose point, to measure "recuperation"; (2) a repeat measurement of the initial regeneration dose, to calculate the "recycling ratio", which tests the internal consistency of the growth curve; (3) a second repeat of the initial regeneration dose followed by a room temperature IR bleach and subsequent OSL measurement, to calculate the "IR depletion ratio", which allows contaminating feldspar grains to be detected. Dose response curves were fitted with a saturating-exponential-plus-linear function. De values were calculated for individual grains by projecting the sensitivity-corrected natural luminescence intensity (Ln/Tn) onto the dose response curve. The standard error associated with each individual De determination was estimated by Monte Carlo simulation. Curve fitting, De determination and Monte Carlo simulation were performed using version 3.24 of the Luminescence Analyst software (S43). A selection of OSL decay and growth curves is are presented in Fig. S8. C.3.c Single-grain rejection criteria It has been observed widely that the majority of quartz grains from unheated sedimentary deposits do not yield a measureable OSL signal (e.g. S44-47). Similarly, a large number of grains display luminescence characteristics which indicate that they are unsuitable for age determination. Consequently, single-grain dating studies must adopt criteria for rejecting uninformative grains (S45). In this study, we adopted rejection criteria similar to those used by (S37). Specifically, grains were rejected where one or more of the following conditions are met: (1) the natural signal from the grain is too low to distinguish it from the variability in the background signal (the difference between Tn and the average count from the background interval is less than three times the standard deviation of the counts in the background interval, determined using the "sig. >3 sigma above BG" rejection criterion in Luminescence Analyst, S43); (2) the recycling ratio (S34) differs from unity by greater than 10 %; (3) the IR-depletion ratio (S48) is greater than two standard errors below unity; (4) recuperation is high (i.e. Lx/Tx for the 0 Gy regeneration point is greater than 5 % of Ln/Tn, S34); (5) Ln/Tn does not intercept the growth curve (S35, S49, S50) or (6) the dose response curve shape precludes the generation of a meaningful equivalent dose. Unlike the other five criteria, this last criterion cannot be applied mechanistically and requires user-judgement. It was used where the dose response curve showed no pattern, or where Ln/Tn intercepted the dose response curve at the point where growth ceased (Fig S8.c). The latter case was more common, and was assumed to indicate sample saturation, which precludes the measurement of a meaningful finite equivalent dose. These rejection criteria were applied to grains in the order listed above and only one cause for rejection was noted per grain. The results of this analysis are presented in Table S4. Of the 25,600 grains measured, only 700 displayed acceptable luminescence characteristics, a yield of 2.7 %. This value is relatively low, with yields from individual samples ranging from 1.3 % to 3.7 %, while comparable values of 1.7 % and 12.7 % were obtained for samples from Blombos Cave, South Africa (S45, S51). The majority of grains were rejected due to either low natural sensitivity (67 %) or a poor recycling ratio (26 %). The remaining criteria account for the rejection of c.1 % of grains each. However, the majority of grains rejected because Ln/Tn does not intercept the growth curve came from the Tr-19 samples. Nearly as many grains from sample Tr19-04 were rejected (96 grains) for this reason as were accepted (107 grains). At present, the physical mechanism which causes Ln/Tn to fail to intercept the growth curve is not well understood, though it is likely to relate to growth curve characteristics at high doses. High dose grains are thought to be present in Tr-19 due to partial-bleaching during deposition (Section C.3.e). It is possible that a small number of high dose grains may be also be derived from in-situ decomposition of the limestone bedrock, though laboratory dissolution of a 2.4 kg block of bedrock only yielded 2.5g of >180 um sand. While variations in both the rejection criteria applied and the format in which the results are reported preclude direct comparison, the results presented in Table S4 are not inconsistent with those found in the published literature. All samples yielded sufficient acceptable De values (S52) to allow the application of appropriate numerical age models (Section C.3.d). C.3.d Estimation of the sample burial dose using statistical models To determine the age of a sample which has been measured using single-grain OSL techniques, it is first necessary to analyse the distribution of individual De values in order to generate a single burial dose (Db). Hereafter, Db is used to signify the dose estimate which is appropriate for use in age calculation. The statistical models most frequently used to calculate Db from single-grain De distributions are the central age model (CAM, S53), minimum age model (MAM, S54) and the finite mixture model (FMM, S55). These models were first applied to single-grain datasets by S53, S56 and S57 respectively. Each model makes different assumptions about the dataset and hence it is vital that the correct model is chosen. The most appropriate model for a particular dataset is determined by the depositional and post-depositional processes to which the sample has been exposed. The CAM generates Db on the assumption that the dispersion of measured De values is explained by the overdispersion parameter (?d, the relative standard deviation of the true paleodoses) and measurement uncertainties (S53). Consequently, it is appropriate for samples where the luminescence signal of all constituent grains was fully bleached prior to deposition, and where no post-depositional mixing has occurred. Such samples represent the ideal for OSL dating but in practice many naturally occurring sediments are more complex. The MAM, as applied to single-grain datasets, is intended to obtain an accurate value for Db from a sample containing both grains which were fully bleached on deposition and grains which were partially bleached. The MAM fits a truncated normal distribution to log De values, with the truncation point giving Db. Application of this model is successful where Db is equal to the paleodose of the fully bleached grains. The MAM has been advocated for application to samples from a wide range of depositional environments (S58, S59). Finally, the FMM was developed to identify separate, fully bleached, populations within a single-grain dataset. This approach treats log De values as a random sample from a mixture of normal populations with identical overdispersion (S57). Consequently, it is appropriate for samples containing more than one population of grains (e.g. samples mixed by bioturbation during burial). Of the statistical models described above, only the CAM and FMM were considered for application to single-grain datasets generated in this study. The MAM was not used since it is very sensitive to the presence of a small proportion of outlying low De values, leading to underestimation of Db (S46, S53, S56, S60). Anomalously low De values may be obtained where (a) younger grains are intruded into older levels (S53, S56); (b) material from a younger deposit is incorporated into the sample by inadvertent cross-cutting of different age layers (S61) and (c) grains experience unusually low beta dose rates (S53, S61, S62). Where anomalously low De values exist in a dataset, the FMM has been shown to yield Db values which are more consistent with independent age control than values produced using the MAM (S46, S60). The choice of age model for each sample was determined by the distribution of equivalent doses within the dataset. Specifically, the overdispersion parameter was used to discriminate between datasets which are consistent with a single-component De distribution, and those where more than one component is present. Based on the reported range of overdispersion values (9-22 %) from samples which are thought to have been well bleached prior to deposition, several authors have concluded that ?d values under 20 % are consistent with single-component De distributions (S58). We adopted this criterion, applying the CAM to datasets with less than 20 % overdispersion and the FMM to datasets with greater than 20 % overdispersion. This approach has been demonstrated to yield stratigraphically and archaeologically consistent ages within and between individual archaeological sites (e.g. S37, S62). All overdispersion values for the FAY-NE1 samples exceed 30 % (Table S5) indicating that several populations of grains are present. This is also apparent from the radial plots presented in Fig. S9. Consequently, all datasets were analysed using the FMM. The FMM was run using two to six De components and overdispersion values between 10 and 20 %. The minimum number of statistically supported De components, and the most appropriate overdispersion value, were determined for each dataset by optimising the Bayesian Information Criterion (BIC) and the maximum log likelihood (llik) (S62). The results of this analysis are presented in Table S6. For six of the seven samples, the FMM indicates the presence of a single dominant population. The equivalent dose derived from this population has been used in subsequent age calculations, with minor components being attributed to post-depositional mixing or partial bleaching (Section C.3.e). In the case of sample Tr19-03, two distinct populations contain 84 % of the grains. However, neither of these components contains more than half of the grains measured. The largest population identified yields an age which is consistent with samples from the same assemblage (Tr19-04 and Tr19-09), and this population is used in subsequent analysis. However, in the absence of a clear dominant population, the age for Tr19-03 must be treated with caution. C.3.e Discussion of the equivalent dose distributions The analysis of the single-grain dose distributions described above follows well established published methods (e.g. S37, S46 and S54). However, before using the FMM output to calculate ages for the FAY-NE1 samples, it is important that the factors contributing to the large scatter observed in the dose distributions are identified. Of the four samples inferred to have been deposited by aeolian processes, three (Tr4-10, Tr28-06 and Tr28-08) yield dose distributions containing an overwhelmingly dominant component (>80 %), while 66 % of grains in the fourth sample (Tr4-02) are contained within one component. In all cases, the majority of the remaining grains are contained in components with a lower De than that of the dominant component. We interpret these lower De components as resulting from post-depositional intrusion of overlying younger grains, probably by bioturbation. The large (25%) low De component observed in sample Tr4-02 is not seen in other samples. This component may be explained by spatially concentrated bioturbation, such as excavation and subsequent infilling of an animal burrow, though no evidence for burrowing was observed during excavation of the sample. Samples Tr4-02 and Tr28-06 both contained small components with high equivalent doses (126 ? 14 and 65.4 ? 7.5 Gy respectively). These components cannot be explained either by inadvertent sampling of underlying material, or by the intrusion of older grains from beneath, since both samples overlie Assemblage A sediments which are not greatly older. They may be caused by insitu decomposition of clasts within the sediment, or by a small proportion of grains being insufficiently exposed to sunlight prior to deposition, though neither explanation is considered likely due to the finite equivalent doses measured and aeolian nature of the host sediment. While no explanation can be offered for these high dose components, the stratigraphic consistency between age estimates (Table S8) for samples containing a high dose component (Tr4-02 and Tr28-06) and those which do not (Tr4-10 and Tr28-08) gives confidence that they do not compromise the dating of the FAY-NE1 site. Of the samples from within Assemblage C, the dose distribution for Tr19-09 shows limited scatter, while those for Tr19-03 and Tr19-04 show much larger scatter. Notably, samples Tr19-03 and Tr19-04 contain a large (38 and 22 % respectively), high-dose component which we attribute to the presence of grains which were not fully bleached prior to deposition. The presence of partially-bleached grains is not unexpected for sediments deposited by slope-wash processes. No such component is present in sample Tr19-09. The majority of grains in samples Tr19-04 (73 %) and Tr19-09 (94 %) are contained within a single component. The tight distribution of De values within this component, especially for sample Tr19-09, strongly suggests that it represents the true burial age of the sediment rather than being an artifact of using the FMM to fit a partially bleached dataset. While the component containing the largest proportion of grains in sample Tr19-03 is more scattered, comparable ages for this sample and Tr19-04 give some confidence that this component also represents the true burial age. All three Assemblage C samples also contain one or more low-dose components. The presence of low-dose components in equivalent dose distributions have been explained by small-scale variations in the beta dose rate (beta heterogeneity) experienced by individual grains (S61). Beta heterogeneity may occur due to spatial variability in the concentration of beta emitters coupled with the short penetration range of beta particles (c.3 mm, S25). Two categories of beta heterogeneity are recognised (S61). Firstly, heterogeneity caused by the presence of large non-radioactive clasts (coldspots) within a higher radioactivity sediment (e.g. limestone flakes within a sandy matrix). Secondly, heterogeneity due to high beta activity mineral grains (hotspots) scattered at low abundance within a low radioactivity sediment (e.g. feldspar grains with high 40K concentrations scattered within a quartz dominated sediment, S63). In the present study, beta heterogeneity due to coldspots is unlikely, since samples were taken from sand layers or sandy pockets within clast rich sediments. In both cases, samples were collected more than 3 mm away from large clasts, and the sandy pockets did not contain small limestone fragments which could act as coldspots. Where the effects of beta heterogeneity have been observed in single-grain datasets (S61), they were attributed to the shielding of grains by gypsum, opal or calcite, which are not present in the samples from FAY-NE1. Beta heterogeneity due to feldspar hotspots (S63) may affect samples from FAY-NE1 due to the relatively low beta and total dose rates at the site. However, numerical modelling (S63) indicates that low dose grains resulting from this effect are unlikely to have De values below 30-40 % of the De of the dominant population. The low dose components for FAY-NE1 samples range from 5-43 % of the De of the dominant population, from which we infer that these components are not due to hotspots within the sediment. In the absence of beta heterogeneity, we attribute the presence of low dose components in the Assemblage C samples to intrusion of younger grains from overlying sediments. C.3.f Environmental dose rate calculations For HF acid etched sand-sized quartz grains, the environmental dose rate consists of external beta, gamma and cosmic ray components. An alpha dose rate due to uranium and thorium within quartz grains has been included in several recent studies, but is here omitted (S64). Beta and gamma dose rates may be calculated from the water content and radioisotope concentration of the host sediments, while cosmic dose rates are calculated from the sample location (latitude, longitude and latitude) and burial depth. Owing to the differing penetration of beta (c.3 mm) and gamma (c.30 cm) radiation (S25), and the inhomogenous nature of the sampled sediments, two separate radioisotope concentration measurements were made for each sample. The beta dose rate was calculated based on radioisotope concentrations measured using ICP-MS (U and Th) and ICP-AES (K). These measurements were performed on dried, homogenised sub-samples of the sample used for dating. The gamma dose rate was calculated based on radioisotope concentrations measured using an EG&G Ortec Micronomad in-situ gamma-spectrometer, which had been calibrated against the Oxford concrete blocks (S65, S66). Dose rates were calculated using the conversion factors of (S67), assuming that the present day radioisotope concentrations had prevailed throughout the burial period. Secular equilibrium in the uranium and thorium decay series was assumed. Dose rates were corrected for the effects of HF etching (S68), grain size (S69) and an assumed water content of 5?5 %. The assumed water content value encompasses the full range of likely burial conditions. Assumed values were used in preference to present-day measured values due to the existence of significant climatic fluctuations over the period of interest (S70) and the fact that samples were collected from sections which had been exposed for considerable periods of time. Cosmic ray dose rates were calculated using site location (25?N, 56?E, 180 m elevation) and present day burial depths, assuming an overburden density of 1.85 g/cm3 (S71). We have also allowed for shielding from cosmic rays by the nearby rock face, using the cos2? zenith angular distribution of cosmic rays (S72). This correction is small (1-2 % decrease in the total dose rate) due to the horizontal distance of sample sites from the rock face (14-17 m) and the small proportion of cosmic dose originating from high zenith angles. Dose rates are presented in Table S7. It was not possible to obtain a field gamma-spectrometer reading for sample Tr19-04 and hence the gamma dose rate for this sample was calculated using ICP-MS/AES radioisotope concentrations. On theoretical grounds, it is expected that this approach will give an overestimation of the gamma dose rate, since the site contains abundant limestone clasts which are assumed to have lower radioisotope concentrations than the sandy sediment. For the FAY-NE1 samples, total dose rates determined using ICP-MS/AES radioisotope concentrations to calculate the gamma contribution are around 10 % higher than those calculated using field gamma-spectrometer data. There is considerable inter-sample variability in this discrepancy, precluding the calculation of a realistic gamma dose rate correction for Tr19-04. Consequently, we note that the total dose rate for sample Tr19-04 (Table S7) is likely to overestimate the true value by c.10 %, leading to a corresponding underestimate in the calculated age. Therefore, we regard the age for sample Tr19-04 as a minimum age estimate. C.3.g Age determination and analysis Ages were calculated using the equivalent dose component (Db) containing the largest proportion of acceptable grains for each sample (Table S6), divided by the total dose rate (Dr, Table S7). Ages are presented in Table S8. The datum for these ages is 2010. D. Paleoclimate and paleogeography of SE-Arabia during the Upper Pleistocene During the Pleistocene, the expansion, contraction and extinction of earlier hominin populations has been closely related to changes in climate and environment (S74). Evidence for earlier, pre-AMH, dispersals into Arabia comes from a number of Acheulean sites, which suggest that hominins may have spread into the peninsula post 800 ka (S6, S7). Evidence for wetter, interglacial conditions in the Arabian peninsula during the mid to late Pleistocene is well-documented (S70, S75, S76). During these wetter periods it is possible that several pulses of pre-AMH emergence occurred. Proxy evidence indicates that MIS 6 was not as cool or variable as other glacial periods, and although sea levels were up to 120m lower than the present (S77, S78), several periods of increased global temperature and increased precipitation appear to have occurred in the low latitudes (S75, S76). The end of MIS 6 is marked by a period of rapid global warming at ~130 ka, characterised by a return to full interglacial conditions, perhaps in as little as 70 years (S79). The onset of the Last Interglacial period around 130,000 years ago (MIS 5e) was characterised by an abrupt and drastic increase in rainfall over Arabia that lasted until approximately 120 ka (S75, S76). This evidence is supported by speleothem records from the Near East with significantly depleted ?18O values between 128 and 120 ka (S80). An extensive network of lakes developed across the Arabian Peninsula (S70), including the mega-lake at Mudawwara, Jordan which covered in excess of 2,000 km? (S81). Lake and speleothem records indicates that MIS 5e and 5a were wet across Arabia, punctuated by drier episodes during MIS 5d and 5b, as denoted by the accumulation of aeolian dunes in the Wahiba Sands (S82). It has been postulated that the onset of arid conditions at the MIS 5a/ 4 boundary may have coincided with the Toba eruption 74 ka (S19, S83), although this remains a matter of debate (S84). Toba Ash does, however, provide a key stratigraphic marker which has been detected in Arabian Sea sediments (S85). Evidence for aridification during MIS 4 is found in the Wahiba Sands, where dune deposition occurred between 71-57 ka (S82). Marine evidence suggests that stage 3 (60 - 25 ka) was complex and comprised a series of fluctuations in aridity as well as increased phases of monsoon intensity (S85). Arabian Sea ?15N records suggest increased monsoon strength between 60 and 30 ka, punctuated by a series of abrupt arid phases corresponding with Heinrich events H4-H6 (S86). This view is corroborated by speleothem records on the island of Socotra, where a speleothem record spanning 55 to 42 ka BP shows rapid changes in the Indian Ocean Monsoon with corresponding changes in rainfall (S87). The early MIS 3 records show strong links between the Indian Ocean and North Atlantic regions (S85, S86). Fluvial sands in the eastern Arabian shield region are OSL dated between 54 and 34 ka (S84), whilst interdunal sabkah sediments in the Liwa region of the UAE have been dated between 54 and 34 ka (S89). Both lines of evidence support the notion of wetness in Arabia during MIS 3 (S70). Marine core evidence suggests widespread aridity occurred from 33 ka with a pronounced event corresponding to the H3 event identified in the North Atlantic and the Arabian Sea (S85). The late MIS 3 and MIS 2 environment was characterised by arid conditions across Arabia, with the development of large mega-linear dunes across the Rub' al-Khali and Wahiba deserts (S82, S90, S91). Pluvial conditions existed during the early to midHolocene across Arabia (S70). At the onset of interglacial conditions at the MIS 6/5e boundary, global sea-levels were still depressed by up to -120 m (S77, S92, S93). At its southern end, in the region of the Bab al Mandab and Hanish Islands, the width of the Red Sea would have been reduced to less than 4 km (S94). Fig. 3 shows that a lag of ~5000 years occurred between the peak of the Indian Ocean Monsoon Index (S95) and the rise in sea-level to the peak interglacial high stand conditions during MIS 5e ~ 120 ka (S77). Thus full interglacial conditions, coupled with a sea-level low stand, would have been present in the southern Red Sea basin until eustatic adjustment to increased global temperatures had taken place (S77, S93). Likewise evidence from the Persian Gulf indicates sea-level high stands up to 6m higher than today during 5e (S96). During the rest of MIS 5 sea level only fell below -50 m in the southern Red Sea for any period of time between 90 - 86 ka. At -50m a total of five crossings, including two in excess of 10 km would have been present (S93). Between MIS 4 and 1 sea-levels were generally below -60 m in the southern Red Sea (Fig. 3, S93, S94). Evidence from East Africa shows that AMH were present during MIS 6 based on fossil evidence from sites including Herto, Ethiopia Omo-Kibish, Ethiopia (195?5) (S97), (160?2-154?7 ka) (S98), Singa, Sudan (133?2 ka) (S99), Mumba, Tanzania (130 ka) (S100) and lithics associated with dated sediments on the Red Sea coast of Eritrea (S101). We suggest that the rapid climatic amelioration that occurred across the Saharo-Arabian belt led to conditions suitable for the dispersal of AMH out of Africa. Based on the climatic and sea level evidence presented above, a short window of opportunity was present at the onset of MIS 5e, when Arabia experienced increased precipitation, the development of vegetation cover and low sea-level conditions. This would have provided suitable conditions for the expansion of AMH populations across the Red Sea into Arabia. The importance of lower sea levels on dispersal routes across the Red Sea has been considered in some detail (S93). The legacy of Quaternary climate changes is evident in the faunal and floral distribution across the subcontinent. Southern Arabia has strong floral and some faunal affinities with East Africa whilst south-eastern Arabia has a strong Asiatic association. Within Arabia there is a distinct division between these two biogeographical zones, caused by the Nejd Plateau. This region is characterized by flat topography, poor vegetation cover and an absence of surface water, rendering it a formidable barrier to floral and faunal migration (S102, S103). Only during MIS 5 were conditions favourable for AMH to expand their range across Arabia, since many of the interior terrestrial barriers, including the Nejd Plateau, were open. Today the Nejd poses a formidable barrier with little vegetation, no surface water and is a flat, almost featureless expanse in excess of 1,000 km wide. The opening up of the Nejd was important in connecting southern Arabia to south-eastern Arabia. By c.125 ka, AMH with an African connection had spread as far as Jebel Faya (Assemblage C) but would not have been able to spread directly across the Persian Gulf owing to sea-level highstands during 5e (S94). The FAY-NE1 evidence suggests that occupation at the site (Assemblages C and B) persisted into MIS 5a. The onset of arid conditions in MIS 5d or MIS 5b (S82) is likely to have disconnected the AMH population in SE Arabia from that of southern Arabia with the closure of the Nejd route. Owing to the lack of any diagnostic links in lithic typology with either African, Levantine or Zagros Assemblages, B and A must represent autochthonous developments within SE Arabia. We assume that the sterile sediments separating Assemblages B and A accumulated during MIS 4. Recurrent human presence at Faya during MIS 3 is shown by Assemblage A as late as 40-38 ka. The workshop-nature of the site during this period (see Section B) might indicate that it was only visited periodically for re-provisioning with flint implements by a population occupying niches along the adjacent coasts. The onset of hyper-arid conditions during late MIS 3 stopped these activities and may have led to a considerable decrease in regional population densities. This period corresponds with the drying up of lakes across the region and with the deposition of sterile sands at Faya between 39-34 ka. At Faya there is no evidence for human presence between MIS 3 and MIS 1. Figures Figure S1: View of northern Jebel Faya from the north-east. The Jebel comprises Neogene limestones with rich seams of chert. The FAY-NE1 rockshelter is behind the white vehicle. Figure S2: Jebel Faya rockshelter from above, looking north, showing eboulis blocks from roof collapse and the location of excavation trenches. Figure S3: Plan of the FAY-NE1 trenches (bold numbers) discussed in this paper, showing the location of OSL samples (filled circles). Grid in meters, relative to site datum. Figure S4: Detail from the schematic main east-west section (54-45 m). Artifacts found in primary contexts during the excavation of trenches 4, 27, 34 and 19 are plotted, alongside the location of Tr19 OSL samples. Figure S5: Assemblage C: Hand axe pre-form. Additional Assemblage C material is illustrated in Fig. 2. Figure S6: Assemblage B: 1, burin; 2, end scraper; 3, radial core; 4, end scraper; 5, orthogonal core; 6, side scraper; 7, parallel core; 8, denticulate. Figure S7: Assemblage A: 1, converging straight side scraper; 2, lateral denticulate; 3, inversely retouched end scraper; 4, multiple platform parallel core; 5, Kombewa core; 6, single platform parallel core; and 7, radial core. Figure S8: Typical OSL decay and growth curves for individual grains of sample Tr19-04: A) three OSL decay curves; B) a grain for which Ln/Tn does not intercept the growth curve; C) a grain where Ln/Tn intercepts the growth curve at the point where growth has ceased, precluding the calculation of a meaningful De; D and E) grains yielding De values close to the mode; F) a grain yielding a large De. Figures B-F: Ln/Tn (closed circle), Lx/Tx (open circles). Figure S9: Radial plots for the equivalent dose distributions obtained from single grains of quartz from Tr4 and Tr28 samples. Each population of grains identified by the finite mixture model is indicated by either a grey bar or a solid line. The grey bar represents the De (? 2 standard errors) of the dominant component, assuming no overdispersion. This equivalent dose has been used in age calculations. The black line(s) represent the equivalent dose(s) of minor components. Figure S10: Radial plots for the equivalent dose distributions obtained from single grains of quartz from Tr19. Sample Tr19-03 contains two large components, both of which are identified by grey bars. The lower component has been used for age calculations (see text). Tables Radiocarbon age Calibrated (cal BP, 1?) (years BP) (marine R=300 ? 100 years) Hd26089 Shell (Turbo sp.) 9,583 ? 66 10,405 - 9,711 Hd27511 Shell (Turbo sp.) 9,657 ? 50 10,380 - 10,078 Table S1: Radiocarbon ages from within the sediments containing early Neolithic artifacts. Lab-code Material Pooled mean (cal BP) 10,084 10,198 Sample Context Assemblage association Tr4-02 Sandy matrix Above A Tr4-10 Sterile sand layer Within A Tr19-03 Sandy matrix Within C Tr19-04 Sandy matrix Within C Tr19-09 Sandy matrix Within C Tr28-06 Sterile sand layer Above A Tr28-08 Sterile sand layer Within A Table S2: FAY-NE1 OSL samples analysed in this study samples. Sample names indicate trench and sample number, such that Tr4-02 indicates the second sample from trench 4. Sample Tr19-04 Tr19-09 Tr28-06 Tr28-08 PH1a PH2b Measured/given dose ratio 260?C/10s 220?C/10s 0.97?0.02 1.02?0.02 1.03?0.05 0.99?0.01 1.01?0.04 Table S3. Dose recovery results for five samples from FAY-NE1, using the preheating regime adopted for single-grain measurements. a PH1 is the preheat carried out prior to measurement of Ln or Lx. b PH2 is the preheat carried out prior to measurement of Tx. Tr4-10 Sample Tr4-02 Tr4-10 Tr19-03 Tr19-04 Tr19-09 Tr28-06 Tr28-08 Total number of grains measured 3600 2700 3300 3600 3600 4000 4800 Grains rejected for the following reasons Tn <3* ?BG 2267 1772 2003 2232 2792 2618 3420 Poor recycling 1060 737 1112b 1055b 612b 1003 1102 ratio Depletion by IR 16 27 12 29 30 119 55 0 Gy dose >5% 69 33 40 12 4 55 92 of Ln No Ln/Tn 26 6 41 96 41 40 26 intersection Poor DR curve 24 38 7 67 17 35 31 Grains stucka 5 0 4 2 4 2 10 Sum of rejected grains 3467 2613 3219 3493 3500 3872 4736 Acceptable individual De values 133 87 81 107 100 128 64 Table S4. The number of single grains which were measured, rejected after application of the criteria outlined in Section C.3.c, and accepted for inclusion in the calculation of the burial dose. a Grains remaining in single-grain sample holders from previous measurements were identified under a binocular microscope and rejected from further analysis. b For Tr19 samples, an additional repeated dose point (c.200 Gy) was used to calculate a second, high dose recycling ratio (S39). Grains were rejected where either or both recycling ratios differed from unity by greater than 10 %. Sample Tr4-02 Tr4-10 Tr19-03 Tr19-04 Overdispersion (%) 77?5 48?4 67?6 51?4 Table S5. Overdispersion values for FAY-NE1 single-grain datasets. Tr19-09 33?4 Tr28-06 58?4 Tr28-08 60?6 N Proportion (%) Proportion (%) Proportion (%) Proportion (%) ?d k=1 k=2 k=3 k= 4 (%) Tr4-02 20 133 1 25 8 66 Tr4-10 19 87 11 --89 Tr19-03 19 81 2 14 46 38 Tr19-04 20 107 1 3 22 73 Tr19-09 20 100 6 --94 Tr28-06 17 128 4 9 6 81 Tr28-08 20 64 12 --88 Table S6: The number of equivalent doses (N) included in the finite mixture model. Optimal BIC and llik values were obtained for the overdispersion (?d) and number of dose components (k) indicated. Components are ranked by De from lowest to highest. The percentage of grains consistent with each component is also shown, with the dominant component being indicated in bold. Sample Tr19-03 contains two large components, both of which are identified by bold text. The lower component (k=3) has been used for age calculations (see text). Sample Sample name Total dose rate, Dr Beta Gamma Cosmic (Gy/ka) Tr4-02 1.75 ? 0.2 5?5 0.65?0.07 0.21?0.02 0.14?0.00 1.00?0.07 Tr4-10 0.72 ? 0.2 5?5 0.55?0.06 0.25?0.02 0.17?0.01 0.96?0.06 Tr19-03 2.1 ? 0.2 5?5 0.46?0.05 0.16?0.01 0.14?0.00 0.77?0.05 Tr19-04 2.4 ? 0.2 5?5 0.59?0.06 0.33?0.03 0.14?0.00 1.05?0.07 Tr19-09 2.4 ? 0.2 5?5 0.73?0.07 0.21?0.02 0.14?0.00 1.08?0.07 Tr28-06 0.65 ? 0.2 5?5 0.67?0.07 0.26?0.02 0.17?0.01 1.10?0.08 Tr28-08 0.9 ? 0.2 5?5 0.54?0.06 0.21?0.02 0.16?0.01 0.91?0.06 Table S7. Sample depth, water content and dose rate for FAY-NE1. Depths are below present ground surface and hence cannot be used to infer stratigraphic relationships within or between trenches. No field gammaspectrometer reading could be obtained for sample Tr19-04, hence the gamma dose rate for this sample has been calculated using ICP-MS/AES radioisotope concentrations. Sample depth (m) Moisture content (%) Dose rate (Gy/ka) Proportion of accepted Age Db grains contributing to Db (Analysed / accepted) (%) (Gy) (ka) Tr4-02 3600 / 133 66 38.5?1.7 38.6?3.2 Tr4-10 2700 / 87 89 38.7?1.6 40.2?3.0 Tr19-03 3300 / 81 46 97.8?10.1 127?16 Tr19-04 3600 / 107 73 99.6?12.2 94.8?13.0 Tr19-09 3600 / 100 94 132?6 123?10 Tr28-06 4000 / 128 81 37.3?1.5 34.1?2.8 Tr28-08 4800 / 64 88 35.1?1.6 38.6?3.1 Table S8. Summary dating results and ages. Uncertainties in the age estimates are based on the propagation, in quadrature, of errors associated with individual errors for all measured quantities. In addition to uncertainties calculated from counting statistics, errors due to (1) beta source calibration (3 %, S29); (2) single-grain instrument reproducibility (1.5 %); (3) dose rate conversion factors (3 %) and attenuation factors (3 %) have been included (S73). Sample name Number of grains References S1. S2. S3. S4. S5. S6. S7. S8. S9. S10. S11. S12. S13. S14. S15. S16. S17. S18. S19. S20. S21. S22. S23. S24. S25. S26. S27. S28. A. E. Marks, in Evolution of Human Populations in Arabia: Paleoenvironments, Prehistory and Genetics, M. D. Petraglia, J. I. Rose, Eds. (Springer, Dordrecht, 2009), pp. 295-308. H-P. Uerpmann, D. T. Potts, M. Uerpmann, in Evolution of Human Populations in Arabia: Paleoenvironments, Prehistory and Genetics, M. D. Petraglia, J. I. Rose, Eds. (Springer, Dordrecht, 2009), pp. 205-214. S. McBrearty, World Archaeol. 19, 388-420 (1988). P. Van Peer, Curr. Anth. 39, 115-140 (1998). O. Bar-Yosef, Phil. Trans. R. Soc. Lond. B. 337, 193-200 (1992). M. D. Petraglia, N. Drake, A. Alsharekh, in Evolution of Human Populations in Arabia: Paleoenvironments, Prehistory and Genetics, M. D. Petraglia, J. I. Rose, Eds. (Springer, Dordrecht, 2009), pp. 103-116. M. D. Petraglia, J. World Prehist. 17, 141-179 (2003). R. G. Klein. Evol. Anth. 9, 17-36 (2000). P. Mellars, Proc. Nat. Acad. Sci. 103, 9381-9386 (2006). M. Mehlman, thesis, Univ. of Illinois, (1989). J. Yellen et al., Paleoanthropology 10, 25-100 (2005). J. D. Clark, J. World Prehist. 2, 235-305 (1988). J. D. Clark, The prehistoric cultures of the Horn of Africa. (Cambridge Univ. Press, Cambridge, 1954). C. Perl?s, L'Anthropologie 78, 529-52 (1974). D. Pleurdeau, Afr. Arch. Rev. 22, 177-197 (2005). L. Meignen, in The definition and interpretation of Levallois technology, H. Dibble, O. Bar-Yosef, Eds. (Prehistory Press, Madison, 1995), pp.361-380. R. Crassard, in Evolution of Human Populations in Arabia: Paleoenvironments, Prehistory and Genetics, M. D. Petraglia, J. I. Rose, Eds. (Springer, Dordrecht, 2009), pp. 151-168. F. Biglari, in Iran, Fragments from Paradise, (Instituto Nacional de Antropologia e Historia, Mexico City, 2007), pp.31-39. S. H. Ambrose, J. Hum. Evol. 34, 623-651 (1998). S.A. Brandt, Afr. Arch. Rev. 4, 41-82 (1986). N. Coinman, in More than meets the eye, A.N. Goring-Morris, A. Belfer-Cohen, Eds. (Oxbow, Oxford, 2003), pp.151-170. K. Monigal, in More than meets the eye, A.N. Goring-Morris, A. Belfer-Cohen, Eds. (Oxbow: Oxford, 2003), pp.118-133. M. Otte, J. Koz?owsk? L'Aurignanien du Zagros (ERAUL, Li?ge, 2007). D. J. Huntley, D. I. Godfrey-Smith, M. L. W. Thewalt, Nature 313, 105-107 (1985). M. J. Aitken, An Introduction to Optical Dating, (Oxford Univ. Press, Oxford, 1998). Z. Jacobs, R. G. Roberts, Evol. Anth. 16, 210-223 (2007). G. A. T. Duller, Luminescence dating: guidelines in using luminescence dating in archaeology, (English Heritage, Swindon, 2008). A. G. Wintle, Boreas 37, 369-370 (2008). S29. S30. S31. S32. S33. S34. S35. S36. S37. S38. S39. S40. S41. S42. S43. S44. S45. S46. S47. S48. S49. S50. S51. S52. S53. S54. S55. S56. S57. S58. S59. S60. S61. L. B?tter-Jensen, C. E. Andersen, G. A. T. Duller, A. S. Murray, Radiat. Meas. 37, 535-541 (2003). G. A. T. Duller, L. B?tter-Jensen, A. S. Murray, A. J. Truscott, Nucl. Instrum.Methods 155, 506-514 (1999). G. A. T. Duller, L. B?tter-Jensen, A. S. Murray, Radiat. Meas. 37, 543-550 (2003). S. J. Armitage, R. M. Bailey, Radiat. Meas. 39, 123-127 (2005). M. Ballarini, A. G. Wintle, J. Wallinga, Anc. TL 24, 1-7 (2006). A. S. Murray, A. G. Wintle, Radiat. Meas. 32, 57-73 (2000). S. J. Armitage, G. A. T. Duller, A. G. Wintle, Radiat. Meas. 32, 571-577 (2000). S. J. Armitage et al., Geomorph. 82, 295-308 (2006). Z. Jacobs et al., Science 322, 733-735 (2008). R. G. Roberts, R. F. Galbraith, J. H. Olley, H. Yoshida, G. M. Laslett, Archaeometry 41, 365-395 (1999). J. Wallinga, A. S. Murray, G. A. T. Duller, G.A.T. Radiat. Meas. 32, 691-695 (2000). A. S. Murray, A. G. Wintle, Radiat. Meas. 37, 377-381 (2003). K. J. Thomsen, A. S. Murray, L. B?tter-Jensen, Radiat. Meas. 39, 47-61 (2005). Z. Jacobs, G. A. T. Duller, A. G. Wintle, Radiat. Meas. 41, 264-277 (2006). G. A. T. Duller, Anc. TL 25, 15-24 (2007). G. A. T. Duller, L. B?tter-Jensen, A. S. Murray, Radiat. Meas. 32, 453-457 (2000). Z. Jacobs, G. A. T. Duller, A. G. Wintle, J. Hum. Evol. 44, 613-625 (2003). H. Rodnight, G. A. T. Duller, A. G. Wintle, S. Tooth, Quat. Geochron. 1, 109-120 (2006). K. J. Thomsen, A. S. Murray, L. B?tter-Jensen, J. Kinahan, Radiat. Meas. 42. 370379 (2007). G. A. T. Duller, Radiat. Meas. 37, 161-165 (2003). H. Yoshida, R. G. Roberts, J. M. Olley, G. M. Laslett, R. F. Galbraith, Radiat. Meas. 32, 439-446 (2000). R. M. Bailey, S. J. Armitage, S. Stokes, Radiat. Meas. 39, 347-359 (2005). Z. Jacobs, G. A. T. Duller, A. G. Wintle, C. S. Henshilwood, (2006b) J. Hum. Evol. 51, 255-273 (2006). H. Rodnight, Anc. TL 26, 3-9 (2008). R. F. Galbraith, R. G. Roberts, G. M. Laslett, H. Yoshida, J. M. Olley, Archaeometry 41, 339-364 (1999). R. F. Galbraith, G. M. Laslett, Nucl. Tracks Radiat. Meas. 21, 459-470 (1993). R. F. Galbraith, P. F. Green, P.F. Nucl. Tracks Radiat. Meas. 17, 197-206 (1990). R. Roberts et al., Nature 393, 358-362 (1998). R. G. Roberts, R. F. Galbraith, H. Yoshida, G. M. Laslett, J. M. Olley, Radiat. Meas. 32, 459-465 (2000). J. M. Olley et al., Sediment. Geol. 169, 175-189 (2004). J. M. Olley, T. Pietsch, R. G. Roberts, Geomorph. 60, 337-358 (2004). G. A. T. Duller, Quat. Geochron. 1, 296-304 (2006). Z. Jacobs, A. G. Wintle, R. G. Roberts, G. A. T. Duller, J. Arch. Sci. 35, 1808-1820 (2008). S62. Z. Jacobs, A. G. Wintle, G. A. T. Duller, R. G. Roberts, L. Wadley, J. Arch. Sci. 35, 1790-1807 (2008). S63. Y. S. Mayya, P. Morthekai, M. K. Murari, A. K. Singhvi, Radiat. Meas. 41, 10321039 (2006). S64. D. Vandenberghe, F. De Corte, J-P. Buylaert, J. Ku?era, P. Van den haute, Radiat. Meas. 43, 771-775 (2008). S65. E. J. Rhodes, J-L. Schwenninger, Anc. TL 25, 5-8 (2007). S66. M. J. Aitken, Thermoluminescence Dating, (Academic Press, London, 1985). S67. G. Adamiec, M. J. Aitken, Anc. TL 16, 37-49 (1998). S68. D. G. Hong, thesis. University of Edinburgh, United Kingdom (1988). S69. V. Mejdahl, Archaeometry 21, 61-72 (1979). S70. A. G. Parker, in Evolution of Human Populations in Arabia: Paleoenvironments, Prehistory and Genetics, M. D. Petraglia, J. I. Rose, Eds. (Springer, Dordrecht, 2009), pp. 38-49. S71. J. R. Prescott, J. T. Hutton, Nucl. Tracks Radiat. Meas. 14, 223-227 (1988). S72. M. A. Smith, J. R. Prescott, M. J. Head, Quat. Sci. Rev. 16, 299-320 (1997). S73. A. S. Murray, J. M. Olley, Geochronometria 21, 1-16 (2002). S74. R.A. Foley, J. Hum. Evol. 26,277-289 (1994). S75. S. J. Burns, D. Fleitmann, A. Matter, U. Neff, A. Mangini, Geol. 29, 623-626 (2001). S 76. D. Fleitmann, A. Matter, C.R. Geosci. 341, 633-642 (2009). S77. M. Siddall et al., Nature 423, 853-858 (2003). S78. K Lambeck, Earth.Plan. Sci. Lett. 142, 43-57 (1996). S79. M. J. Kelly et al., Palaeogeog. Palaeoclim. Palaeoecol. 326, 20-38 (2006). S80. M. Bar-Matthews, A. Ayalon, M. Gilmour, A. Matthews, C. J. Hawkesworth, Geochim. Cosmochim. Acta 67, 3181-3199 (2003). S81. N. Petit-Maire et al., Glob. Plan. Change 72, 368-373 (2010). S82. F. Preusser, C. R. Geosci. 341, 621-632 (2009). S83. M. R. Rampino, S.H. Ambrose. Geol. Soc. Am. Spec. Pub. 345, 71-82 (2000). S84. F. J. Gawthorne-Hardy, W.E.H. Harcourt-Smith, J. Hum. Evol. 45, 227-230 (2003). S85. H. Schulz, U. von Rad, H. Erlenkeuser, Nature 393, 54-57 (1998). S86. S87. M. A. Altabet, M. J. Higginson, D. W. Murray, Nature 415, 159-162 (2002). S. J. Burns, D. Fleitmann, A. Matter, J. Kramers, A. A. Al-Subbary, Science 301, 1365-1367 (2003). S. J. McLaren, F. Al-Juaidi, M. D. Bateman, A. C. Millington, J. Quat. Sci. 24, 198207 (2009). S. Stokes, H. Bray, A. Goudie, W.W. Wood, in Developments in Water Science 50, A. S. Alsharhan and W. W. Wood, Eds. (Elsevier, Amsterdam), pp.371-378, A.S. Goudie et al., Sedimentology 47, 1011-1021 (2000). A.G. Parker, A.S. Goudie, Geomorphology, 101, 458-470 (2008). D. G. Martinson et al., Quat. Res. 27, 1-29 (1987). G. N. Bailey, in Evolution of Human Populations in Arabia: Paleoenvironments, Prehistory and Genetics, M. D. Petraglia, J. I. Rose, Eds. (Springer, Dordrecht, 2009), pp. 15-37. S88. S89. S90. S91. S92. S93. S94. S95. S96. S97. S98. S99. S100. S101. S102. S103. C. A. Fernandes, E. J. Rohling, M. Siddall, J. Biogeog. 33, 961-966 (2006). D. C. Leuschner, F. Sirocko, Palaeogeog. Palaeoclim. Palaeoecol. 197, 83-95 (2003). A. H. Williams, G.M. Walkden, Geol. Soc. Lond. Spec. Pub. 195, 371-386 (2002). I. McDougall, F. H. Brown, J. G. Fleagle, Nature 433, 733-736 (2005). T. D. White et al., Nature 423, 742-747 (2003). F. McDermott et al., J. Hum. Evol. 31, 507-516 (1996). G. Br?uer, M.J. Mehlman, Am. J. Phys. Anth. 75, 69-76 (1988). R. C. Walter et al., Nature 405, 65-69 (2000). D. E.Wildman et al., Mol. Phyl. and Evol. 32, 287-296 (2004). A.G. Parker, J.Rose. PSAS, 38, 25-42 (2008).