Palaeolithic artefact deposits at Wadi Dabsa , Saudi Arabia ; a multi-scalar geoarchaeological approach to building an interpretative framework

Surface artifacts dominate the archaeological record of arid landscapes, particularly the Saharo‐Arabian belt, a pivotal region in dispersals out of Africa. Discarded by hominins, these artifacts are key to understanding past landscape use and dispersals, yet behavioral interpretation of present‐day artifact distributions cannot be carried out without understanding how geomorphological processes have controlled, and continue to control, artifact preservation, exposure and visibility at multiple scales. We employ a geoarchaeological approach to unraveling the formation of a surface assemblage of 2,970 Palaeolithic and later lithic artifacts at Wadi Dabsa, Saudi Arabia, the richest locality recorded to date in the southwestern Red Sea coastal region. Wadi Dabsa basin, within the volcanic Harrat Al Birk, contains extensive tufa deposits formed during wetter conditions. We employ regional landscape mapping and automatic classification of surface conditions using satellite imagery, field observations, local landform mapping, archaeological survey, excavation, and sedimentological analyses to develop a multiscalar model of landscape evolution and geomorphological controls acting on artifact distributions in the basin. The main artifact assemblage is identified as a palimpsest of activity, actively forming on a deflating surface, a model with significant implications for future study and interpretation of this, and other, surface artifact assemblages.


INTRODUCTION
Surface accumulations of lithic artefacts are key archaeological archives for understanding patterns of mobility and landscape use by past hunter-gatherer populations (see, for example Holdaway, Douglass, & Fanning, 2012;Holdaway & Fanning, 2014;Davies, Holdaway, & Fanning, 2016, In Press). While surface assemblages may lack the tight chronological constraints provided by sealed archaeological remains, the patterning of artefacts across a landscape is the legacy of a suite of activities that resulted in the discard of artefacts. This patterning therefore has the potential to retain signatures of past hunter-gatherer landscape use recorded on a spatial scale beyond that possible from single site excavations (Foley & Lahr, 2015;. Examination of surface artefact distributions within their landscape context, and in particular consideration of locations where large numbers of artefacts are observed, opens the possibility of using these artefacts and assemblages to understand how past populations interacted with their landscapes. Understanding the patterning of surface assemblages may inform us about the ways in which landscape features such as water or raw material sources and topography may have influenced mobility and activity within a landscape, and thus the mechanisms and conditions of dispersals (Finlayson, 2013;Devès, Sturdy, Godet, King, & Bailey, 2014;Holdaway, King, Douglass, & Fanning, 2015;Winder et al., 2015;Breeze et al., 2016;).
Due to its proximity to the Indian Ocean Monsoon and the orographic rainfall generated by the Arabian Escarpment (Rosenberg et al., 2012), the south-eastern Red Sea littoral region was a potential refuge during periods of aridity across the Saharo-Arabian belt (Bailey, 2009;Parker, 2009;. It contains a rich archaeological record covering potentially 1.8 million years (Zarins et al. 1980(Zarins et al. , 1982Inglis et al., 2014;Bailey et al., 2015;Foulds et al., 2017). The discovery of a dense accumulation of almost 3,000 Early Stone Age (ESA) and Middle Stone Age (MSA) artefacts associated with extensive tufa deposits at Wadi Dabsa in the volcanic Harrat al Birk, Asir Province Foulds et al., 2017) represents the richest Palaeolithic locality recorded to date in southwestern Saudi Arabia (Sinclair et al., In Press), offering a rare chance to examine ESA and MSA activity at different scales within a landscape that may have proved persistently attractive to hominins over an extended period of time. Research questions at Wadi Dabsa focus on examining the archaeological assemblage in its landscape context. Does the location of the artefacts mark a particular concentration of hominin activity, perhaps attracted by the intersection of factors such as accessible raw material, access to water and prey animals, and good views across the basin? Under what environmental and landscape conditions did this activity take place? Did the landscape, and therefore the opportunities it afforded, change during the period of hominin occupation in the region? How does our understanding of hominin dispersals change in response to the answers to these questions?
Interpretation of landscape use and hominin behaviour from surface artefact distributions across a modern landscape is not a straightforward matter of mapping artefact densities. Firstly, landscapes are dynamic in space and time, and landscape features and environmental conditions that may have attracted past populations, such as lakes or sources of raw material, may have disappeared and reappeared multiple times; we therefore need to reconstruct the history of these features to understand when and where these conditions existed.
Secondly, as well as shaping the resources available to past populations, the geomorphological evolution of the landscape shapes, at all spatial scales, where artefacts are preserved, exposed and rendered visible to survey, processes which continue to impact the record today (Fanning & Holdaway, 2004;Fanning, Holdaway, & Rhodes, 2007). To develop interpretations of the surface archaeological record, we must therefore first understand its geomorphic context, and to what extent artefact patterning, at all scales, may be a product of geomorphological processes rather than behavioural ones (Holdaway & Fanning, 2008;Knight & Zerboni, 2018). Thirdly, the artefacts which are observed cannot be interpreted as an ethnographic snapshot of the past (Stern, 1994); nearly all archaeological deposits are palimpsests of repeated activities, and thus represent varying time envelopes (Bailey, 2007;Stern, 2008) which may, in the case of Palaeolithic surface artefact assemblages, encompass tens or even hundreds of thousands of years.
This paper considers the dynamic landscape and geomorphic context of the Wadi Dabsa assemblage, the processes shaping its formation, and their implications for the future analysis and interpretation of this important assemblage. At the basin-scale, it presents a model of landscape and environmental evolution to be tested and constrained by future chronometric dating. It constructs a framework for understanding the geomorphological processes shaping the distribution of surface artefacts across the basin, and the implications of this framework for interpretation of the observed artefact distribution at the basin scale. It examines the processes controlling the dense accumulation of surface artefacts observed at one specific location within the Wadi Dabsa basin, and the extent to which the formation history of this assemblage can link the artefacts to the landscape stratigraphy, and to constrain the time depth of the assemblage, laying the foundations for future detailed analysis of one of the richest localities recorded to date in southwestern Saudi Arabia.

STUDY AREA: THE HARRAT AL BIRK
The Harrat Al Birk, SW Saudi Arabia, is one of several harrat (volcanic fields) across the Arabian Shield (Coleman, Gregory, & Brown 1983). Spanning the area between the present-day Red Sea coastline and the foothills of the Arabian Escarpment, between N18.76-17.79º latitude, it is a major (over 1,800 km 2 ) and distinct geomorphological zone that interrupts the flat Tihamat Asir coastal plain bordering the Red Sea (Figure 1; Vincent, 2008).
The Harrat al Birk comprises extensive basaltic flows, volcanic deposits, and cinder cones formed by repeated eruptions. The dates for eruptions are poorly-constrained but estimates using K-Ar whole rock dating range from 0.2 to 2.6Ma (Coleman et al., 1983;Dabbagh et al., 1984). The most recently-measured K-Ar ages were obtained from a basalt flow south of the town of Al Birk, 20 km south of Wadi Dabsa, placing it at 1.37 ± 0.02 and 1.25 ± 0 0.02 Ma . Given the wide variation in surface erosion of the lava flows, the Harrat was likely formed through an extended period of punctuated volcanism, related to intra-plate volcanism similar to other Arabian harrat (Bosworth & Stockli, 2016). The timing of this volcanism places it within the known timeframe of the earliest dated hominin dispersals out of Africa, ~1.8 Ma (Ferring et al. 2011), meaning that it is possible that the landscape of the Harrat al Birk, and the resources it provided, underwent intermittent, significant transformation during the period in which hominin populations inhabited the region, changes which need to be understood before interpreting past hominin occupation of the region.

Figure 1: Study region location in southwestern Saudi Arabia. The Harrat al Birk is a major landscape feature interrupting the low-lying Tihamat Asir coastal plain that is bounded by the Arabian escarpment and the present-day coastline. Elevation data: ASTER GDEM 30m.
The Harrat is drained by deeply-incised wadis (up to 30-40m deep), with the incision in places exposing the underlying schistic bedrock. The wadis have occasionally been filled with later lava flows (e.g. Wadi Najla; Inglis et al., 2015). After rainfall (primarily during the winter months), water pools at points within the wadi channels. Tufa and other carbonate deposits precipitated by flowing and ponding of carbonate-rich water are common in the wadis, indicating past periods of increased water flow (Dabbagh et al., 1984). The steep topography of the wadis, and the roughness of the basaltic surface within the harrah, may have allowed the tracking and trapping of prey by hominin populations, as well as facilitating movement from the coastline into the interior (Kübler, King, Devès, Inglis, & Bailey, In press).
A narrow (0.5-1km wide) coastal plain consisting of sabkha and alluvial deposits abuts the western edge of the harrah. Marine deposits, in the form of beachrock and fossil coral, have been recorded at elevations of between 3 and 7.5 m asl in numerous locations along this coastline (Zarins et al., 1981;Dabbagh et al., 1984;Bailey et al., 2015). Whilst yet undated, it is likely that these past sea level indicators mark an MIS5e high sea stand (Inglis, Bosworth, Bailey, & Rasul, In Press), indicating that during periods of high sea level, populations in the Harrat al Birk had access to marine resources. During periods of low sea level, however, including most of the last 70,000 years, the Red Sea coastline would have been further west, leaving the western edge of the Harrat al Birk up to 100 km from the sea during the Last Glacial Maximum (Lambeck et al., 2011;Bailey et al., 2015).

Wadi Dabsa
Wadi Dabsa drains the western side of the Harrat Al Birk, entering the Red Sea near Dabsa town. Two main tributaries begin in the basaltic flows to the east, joining just upstream of Dabsa, ~4 km from the present-day coastline ( Figure 2). The northern tributary runs NE-SW for approximately 12km, deeply incising basaltic flows before opening out to a wide channel where it joins the southern tributary. The southern tributary, the focus of investigations presented here, runs ENE-W for ~7 km through the basaltic flows before a change in topography is marked by a widening of the channel prior just upstream of its junction with the northern tributary. Two kilometres upstream of the confluence, the southern tributary flows through a wide basin 800 m x 1400 m, the floor of which is almost completely covered by tufa carbonate deposits. The main channel, running along the contact between the tufa and adjacent basalt, incises the tufa to a depth of over 2 m. A smaller but deeper (> 4 m) channel that incises both tufa and basalt flows from NE to SW across the centre of the tufa exposure, dividing it into what we have termed the western and eastern sub-basins. Tufa barrages are also present within the southern tributary wadi channel, as well as in the incised fan-shaped area where the wadi becomes unconstrained downstream of the western sub-basin ( Figure 2). These tufa deposits have been incised to depths of over 6m in places.
The Wadi Dabsa basin may have been particularly attractive to past populations given: (i) the presence of ponded and/or flowing water, recorded in the tufa deposits, including pool-(paludal-) settings and fluvial barrages with morphologies indicative of both the presence of vegetation (reeds and mosses), and also inorganic surface drapes across the main basin, with the addition of a fluvial cascade to the west of the incised channel that emerges into the fan feature; (ii) extensive sources of raw material for artefact manufacture, in the form of the basalt flows; (iii) complex topography that allows the tracking and trapping of prey (King & Bailey, 2006;Devès et al., 2014); and (iv) and the potentially favourable edaphic properties of soils developed on volcanic deposits (Kubler, Owenga, Reynolds, Rucina, & King, 2015;Kübler et al., In Press).  (Foulds et al., 2017). Using a 50 x 40 m grid (designated Locality 106, or L0106), 903 artefacts were collected from the surface of sediments overlying the tufa, yet the relationship of the artefacts to the underlying tufa remained unclear: some artefacts were observed underneath large loose tufa slabs suggesting that they may have weathered in situ from the tufa. Focussed investigation in 2017 sought to further record the extent, character and geomorphological setting of the L0106 artefacts in their wider landscape.

METHODS
Working from the kilometre-to metre-scale, an interdisciplinary approach to the Wadi Dabsa basin was employed to understand the basin's evolution, the stratigraphic relationship of the observed archaeology to this changing landscape, and the active and past geomorphological processes within the landscape that would have controlled the preservation, exposure and visibility of Palaeolithic artefacts.

Regional Landscape Mapping
Landform units at the region scale were mapped using satellite imagery to develop a broad-scale understanding of the landscape and build a working landscape stratigraphy. Units were defined through the visual examination of Pléiades 1 satellite imagery which has 4 multispectral bands (dated 2/29/2016) with reference to the ASTER 30 m Global Digital Elevation Model (GDEM30m) as well as true colour satellite imagery accessed through Google Earth and mapped in QGIS 2.0. Units were defined by their geological and geomorphological origins and the processes that had most recently shaped them, e.g. basalt flows, basalt terraces, tufa.
Observations were groundtruthed in 2017, and the data imported into ArcGIS 10.4.1 to allow visualisation. When dates become available, this working landscape stratigraphy will be timeconstrained using chronometric dating.

Surface Classification Using Satellite Imagery
The area of the basin underlain by tufa, defined in the regional mapping (Section 4.1), formed the main focus of investigation given the L0106 artefact accumulation was recorded on this unit. As observed on remote sensing imagery and in the field, the land surface on the tufa appeared to be quite variable, from bare tufa to fluvial and aeolian sediments, with very different implications for the preservation, exposure and visibility of artefacts. Understanding the variability of surface condition across the basin would therefore allow a first-level assessment of artefact visibility, and the specific geomorphological setting of the dense artefact accumulation.
Detailed remote mapping of surface condition was conducted using unsupervised, automatic classification of images from Google Earth (RGB) into 10 classes in ENVI 5.2 (output aggregated to 5 pixels and smoothed to 3 pixels). The resulting classes were assigned interpretations based on visual examination of the satellite imagery, and then groundtruthed through visits to eight locations within the basin.

Investigations at the L0106 and L0130 Grids
In 2015, a 40 x 50 m grid was established across the area of dense artefacts at L0106 and divided into 10 x 10 m squares, with each square further subdivided into four 5 x 5 m quadrants designated w, x, y, and z ( Figure 3; Inglis et al., 2015). In 2017 this grid was extended NW by 20 m, and a second 50 x 60 m grid (L0130), offset from L0106 by 5 m to the northwest to avoid a road, was established and similarly divided into 10 x 10 m squares and 5 x 5 m quadrants. The position of additional artefacts, located within the basin but beyond the artefact collection girds, was recorded using either a Trimble Geo7X without an external antenna or a Garmin 62s handheld GPS, with each piece assigned a unique waypoint number. Surface condition within each of the defined landform units is variable and affects the visibility of artefacts. The main surface condition types were described within each quadrant across the grid where artefacts were collected or recorded, as well as the percentage of the quadrant that comprised bare tufa surface.
Artefact collection and recording: The artefacts within the grids were sampled using two complementary strategies. All surface artefacts visible within grid L0106 were collected and bagged by quadrant for post-excavation analysis. Rows 1-4 were collected in 2015, and Rows 5-6 in 2017, providing a 100% sample of the surface artefacts in L0106 (3,000 m 2 ). In L0130 a restricted sampling and recording-only strategy was employed to: a) determine the extent of the artefacts across the tufa surface; and b) document their typologies for comparison to those observed in L0106; whilst c) leaving a significant sample of the artefacts in their geomorphological context for future potential study. Quadrants y and x of alternate squares in Rows 3-6, and all squares in Rows 1-2 were surveyed (total sample: 33%, 1,000 m 2 ) and the artefacts photographed and basic attributes recorded before being left in the field. Artefacts collected from the locality prior to the establishment of the grid, or lying outside the grid were collected for analysis after recording of their locations using a handheld Garmin 62s GPS.
Test pits: To investigate the depth and nature of the sediments overlying the tufa, two small test pits were dug in Quadrants 6Az/By (Test Pit 1) and 5Ax (Test Pit 2) of L0106, in the 'Crest' landform unit (Section 4.3.1) where artefacts were most abundant ( Figure 3). Each pit was begun as a 0.5 x 0.5 m square, before extension to 0.5 x 1 m. After the first 0.5 x 0.5 m sondage, during which all lithics were bagged together, the extensions were excavated stratigraphically, with contexts defined based on lithological changes and lithics bagged according to context.
Bulk sedimentology: Basic sedimentological analyses of the <2mm fraction were undertaken to characterise the sediments exposed within the test pits and aid their interpretation. Percent organics (%LOI organics) and percent carbonate (%LOI carbonates) for the test pit samples were determined by loss on ignition, heating the dry samples sequentially to 550 °C and 950 °C respectively and noting the mass loss. Measurement of the low ( lf) and high ( hf), frequency mass magnetic susceptibility of the test pit sediments was carried out using a Bartington Instruments MS2, 294 consisting of a Magnetic Susceptibility Meter MS2 and MS2B 295 Dual Frequency Sensor. Percent frequency dependent susceptibility ( fd) was calculated for each of the samples.
Particle size analysis was undertaken using a Malvern Mastersizer 2000 following dispersal of the sediment in water, with subsamples measured both with and without dissolution of carbonates using 37% HCl. The latter set allowed comparison of the non-carbonate material in the sediment with the residue derived from a loose block of tufa collected from the grid surface after dissolution with 37% HCl to further understand the origin of the sediment unit overlying the tufa (see below).
Data storage and display: All data from the artefact collection grids was collated in shapefiles in ArcGIS, to allow the visualisation of the data across the grids in both two and three dimensions. The data will be deposited with the Archaeology Data Service, for open access in perpetuity.

Regional Landform Mapping and Landscape Stratigraphy
Visual mapping of landform units from the satellite imagery was aided by the general lack of vegetation in the region. The regional landscape is dominated by basaltic lava flows, some of which can be associated with specific cinder cones (e.g., flow F1 appears to have originated from cinder cone C1 as shown in Figure 4), and their stratigraphic relationships to earlier and later flows are easily defined. Other flows have less obvious origins, or less clear boundaries. The tufa deposits (covering around 1 km 2 ) are constrained to topographic lows, primarily in the main basin and incised wadi channels.
A working landscape stratigraphy, with identified relationships marked by arrows, is presented in Figure 4. A sequence of flow emplacement, from older to younger flows, is proposed, extending from the southwest, where the weathered condition indicates an older age of the volcanic deposits, towards the northeast, where well-defined flows, likely from more recent eruptions, emanate from the C1, C2 and C3 cones. Tufa comprises the major sedimentary unit within the basin, and within the fan-shaped deposit downstream (to the west), with smaller patches of tufa in the wadi channels that link these areas (Figure 4). A 220 m-long incised section of the wadi marks the southwestern boundary of the two main sub-basins, which to the east separates tufa and alluvium surface cover, and further east marks the contact with the adjacent basalt. The surface of tufa in the eastern sub-basin slopes ENE-WSW, from ~124 to ~89 m asl (ASTER GDEM 30m), and the main drainage is consequently E-W; N-S orientated, low, eroded barrages to the west of the L0106/L0130 artefact grids in the eastern sub-basin also suggest that, in this area in the past,  Whilst it appeared from the satellite imagery that nearly all the basalt flows underlie, and were therefore emplaced before, the tufa, the lobate form of the southwestern arm of F4 suggests that it may have been emplaced after tufa deposition. The stratigraphically youngest flow in contact with the tufa, F4, appears to have followed the course of an existing drainage channel that emptied into the basin (Figure 4). Given the impact the emplacement of F4 would have had on the basin's hydrology, potentially blocking a source of water flow into the basin, understanding the relationship between the tufa and this basalt flow was of high priority for understanding the attractiveness of the area to hominins. At all locations in the field where the contact between tufa and basalt could be observed (difficult in areas where unconsolidated aeolian and alluvial sand had collected), the tufa overlay and thus post-dated the basalt. At the southwestern lobe of F4, a small patch of tufa overlay the base of the flow, indicating that here too, a period of tufa deposition had post-dated emplacement of F4. It therefore appears from the observations to date that tufa deposition occurred within a basin whose morphology was defined and occasionally modified by the morphology of the basalt flows. Given the complexity of the tufa formation across the basin, however, the potential for interleaved or coeval deposition of lava flows and tufa cannot be dismissed.

Surface classification and groundtruthing
Unsupervised landscape classification, using Google Earth imagery (RGB), of the Wadi Dabsa basin and its immediate surroundings into ten land surface classes included three classes related to the main areas of sedimentation conditions on the surface of the tufa (Classes Eight to Ten), and seven classes corresponding to differences in the surfaces of the surrounding basalt flows, as well as areas where basalt is exposed in wadis crossing the basin (Classes One to Seven, Figure 6). From satellite observations, these three main tufa surface classifications were interpreted as corresponding to: bare surfaces composed of tufa (Class Nine, Dark Red in

The L0106/0130 Survey Grids
Artefact  In addition to the broad landform units, the percentage of the surface of each quadrant covered by bare, in situ tufa was recorded.

Stratigraphy and Sediments
The two test pits excavated within the Crest landform unit (Quadrants 6Az/6By and 5Ax) both contain the same stratigraphy: a buff-coloured silty fine sand layer of variable thickness (~5-10 cm) overlies, with a clear boundary, grey, silty fine sand (~20-25cm thick), which has an abrupt but irregular lower boundary with the underlying tufa unit (Figure 8). Both units contain occasional, irregularly shaped subangular weathered tufa material, poorly sorted and ranging in size from gravel to boulders, as well as rare basalt boulders, and basalt artefacts.

Figure 8: Images of Test Pit 1 prior to extension showing a) the section (scale is 10 cm), and b) plan views (scale is 25 cm). The former shows the buff-coloured sediment overlying the grey, while in the latter, an artefact amongst tufa clasts within the grey unit is indicated. Photos: H. Robson.
Laser particle size analysis of the <2mm sediment fraction of the test pit sediments ( Figure 9) indicates a broad similarity in the particle size distributions of the layers exposed in the test pit stratigraphy, with moderately to poorly-sorted particles between 250-0.89 µm in size. Removal of carbonates has a negligible effect on the PSD of the grey layer, although in the case of the buff layer, it appears to remove the small sand peak and increasing the percentage volume of material in the size range 100-1 µm. The PSD of the tufa residue after carbonate dissolution comprises sediments within the same range as that of the sediment layers, but with two differences: it contains a better sorted coarse silt/fine sand peak centred around 30-50 µm, as well as a small, non-carbonate coarse sand.  The buff layer has a higher lf than the grey layer (Table 2), indicating a higher content of ferromagnetic minerals in the former. Given the low fd%, it appears that there are very little, if any, superparamagnetic minerals within the two sediments (Dearing et al., 1996). The low lf of the grey layer indicates a general lack of ferromagnetic minerals; the value of the texturally similar buff layer, by contrast, is higher. The grey layer is higher in organic content than the buff (2.97% and 6.34% respectively), although the carbonate content of the two samples are broadly similar -8.21% in the buff layer, and 7.28% in the grey layer.
The bulk sedimentological observations support an interpretation of the buff layer as an active weathering layer of the grey sediment: the PSDs are largely similar between the two layers, save for the small coarse peak in the buff silt (which is not present in the non-carbonate PSD, indicating that it may consist of fine carbonate sand), that either was never present, or has been removed from, the underlying grey layer. The difference in lf between the layers may be the result of wetting and drying of the sediment, oxidising iron present in the surface sediment and thereby increasing the proportion of ferromagnetic minerals in the buff layer, also contributing to its redder colour relative to the grey sediment beneath. This grey-to-red colour change was also observed in Sample <823> when the sample was burnt at high temperature during the LOI analyses.
The mechanisms controlling the organic content are less clear but may reflect the limited amount of organic material preserved in the grey layer breaking down through oxidation and surface weathering in the buff layer. Given the small number of samples, it is unclear how far a difference of a single percentage point in %LOI carbonates indicates a meaningful difference in composition between the samples yet may reflect the variation in small amounts of carbonate sand identified through the PSDs.
These data offer some insights into the relationship between the tufa and the overlying sediment. The abrupt tufa/sediment boundary suggests deposition of transported sediment over a tufa surface, rather than the sediment being a product of the weathering of the tufa. Fragments of tufa observed within the grey and buff sediments may have weathered sub-aerially, rather than being remnants of a partially weathered tufa unit. The PSDs of the grey and buff sediments show that they are both composed of poorly sorted clayey silts, perhaps indicating deposition through shallow overland flow; the greater degree of sorting in similarly-sized material in the tufa residue may be the result of faster flowing water during tufa formation.

Excavated Artefacts
A total of 61 lithic artefacts were recovered from the test pit excavations, 13 from the buff layer (2 MSA, 11 non-diagnostic) and 27 from the grey (4 MSA, 1 LSA and 22 nondiagnostic). The initial sondage of each pit did not distinguish sediment layers, therefore the 21 artefacts excavated in this sondage (1 ESA; 2 MSA; 18 non-diagnostic) cannot be assigned to either layer. When separated by layer, the artefacts exhibit distinct differences in surface characteristics: those from the grey layer all possess a greyish patina and sharp edges, while those from the buff layer exhibit a range of weathering states from sharp/patinated to dark brown polished surfaces with subangular to subrounded edges similar to the outcropping basalt ( Figure 11).
These observations suggest that at least some of the artefacts recovered from the buff layer originated in a similar sedimentary context to the grey layer. The presence of partially polished artefacts in the buff layer support the hypothesis that this layer is an actively deflating and weathering zone of the grey sediments, as their grey patina is being polished away through surface abrasion during exposure. Whether the polished/brown artefacts from the buff layer ever had the grey patina in the first place, however, cannot be confirmed from these observations.

Collection and Recording of Surface Artefacts from L0106/130
The assemblage collected from 120 quadrants in L0106 during the 2015 and 2017 seasons numbers 2,431 artefacts (Table 3)   The assemblage from Grid L0106 contains artefacts from all stages of lithic reduction including tested clasts, cortical and semi-cortical flakes, cores and core-preparation materials, retouched pieces, many shatter fragments, and some probable hammerstones attesting to the flaking of basalt at this location ( Figure 10). There are more than 280 retouched pieces that include a range of classic Acheulean retouched forms (bifaces, cleavers, large cutting tools) from the ESA as well as a series of prepared cores and flakes dating to the MSA, indicating a range of activities that might have been undertaken at the site and, in certain cases, the preparation of armatures for use elsewhere. Finally, there are a small number of true blades and bladelets deriving from prismatic blade cores, of MSA age or possibly younger.
The assemblage recorded but not collected from L0130 is smaller in number than that recorded in L0106, and almost certainly missing the less visible pieces, due to the more rapid survey of the quadrants. Despite this, the range of technological pieces, as well as retouched pieces, suggests that the assemblage in this grid closely mirrors that from L0106 both in chronological periods and range of technological activity.

Artefact Condition
The (Rows 1-4), and from which all artefacts had been collected. Given that the areas of grey patina on these pieces are the same as that observed on the stratified material from the test pits, it is likely that these artefacts were originally contained within the artefact-bearing unit, but had become uncovered between 2015 and 2017, and their grey patinas partially polished away.
In Quadrant 5Jw, a basalt handaxe partially encased in tufa was found on the edge of a small rill that drains the NW corner of L0130 ( Figure 12). While the find was a loose clast, and therefore removed from the geomorphological context in which it developed the tufa coating, it suggests that tufa deposition may have occurred after manufacture of at least some of the artefacts. Artefacts retrieved from the test pits, in contrast, stratigraphically overlie and therefore post-date the deposition of tufa at this location.

Spatial Distribution of Surface Artefacts Within L0106/L0130
The mean number of artefacts per quadrant collected from the surface of L0106 is 20.33 (0.81 artefacts/m 2 ), although a standard deviation of 23.50 indicates that there is considerable variability in artefact number across the grid, with the high numbers of recorded artefacts in 5Cw and 5Cy (132 and 145 respectively) skewing the mean (Figure 13a). There is an increase in artefact counts per quadrant towards the northeast edge of the grid (Figure 13a), on the flattest part of the Crest landform unit, and the landform units with the highest visibility surfaces (i.e., Crest and Upper Slope) have generally higher artefact counts per quadrant while the lowest artefact counts are found on the landform unit with the lowest visibility surface (Drainage). Testing of these models in the L0130 grid is not possible given the different survey technique, as well as the fact that the surveyed squares did not intersect the Crest landform unit in this grid. Beyond this broad scale patterning, there seems to be little relationship between number of artefacts and the percentage of bare tufa within a quadrant, although at the subquadrant scale, artefacts were more often observed on areas of sediment which had become trapped between tufa outcrops, rather than on the bare tufa itself. Mapping artefacts according to identified chronotypology (ESA, MSA, or nondiagnostic) was carried out to attempt to separate phases of activity on the site (Figure 13b; the small number of LSA artefacts precluded meaningful mapping). The ESA artefacts are distributed largely evenly across the grid in low densities, whilst the patterning of the MSA artefacts across the grid is broadly consistent with that shown by the total assemblage. The ESA artefacts, however, comprise just 94 of the total of 2,431 artefacts collected from L0106, and average less than a single artefact per quadrant. The majority of the assemblage (1,605 artefacts, 66%) cannot be typologically assigned to a chronological period.
To statistically test the apparent relationships between artefact counts and landforms, the expected and observed artefact counts of quadrants completely within one landform (a total of 62 quadrants) were tested using chi squared tests (Table 4), with the null hypothesis that artefact counts were unrelated to the landforms. In all cases (except for the LSA artefacts whose low number precluded testing), the p value was well below the significance value of 0.05, indicating that the null hypothesis can be rejected, and that the observed relationships between artefact counts and landforms mapped above are significant.

Geomorphology and Artefacts at L0106/L0130
Investigation of the geomorphological context of artefacts recorded in Grids L0106/0130 suggests the manner in which the surface assemblage has developed across space and over time and continues to accumulate in the present. This has important implications for any interpretations that might wish to propose specific differentiation of hominin behaviours based on the local spatial patterning of the assemblage.
A comparison between the state of patination of the artefacts within sediments exposed in the test pits, and the artefacts collected on the surface, indicate that some of the artefacts recovered from the present day surface were previously contained within the grey sediment unit, and thus deposited following cessation of tufa deposition at this location and have been  (Fanning et al., 2007;Fanning, Holdaway, Rhodes, & Bryant, 2009). While the exposure of additional artefacts during this survey hiatus may go some way to accounting for the higher artefact counts observed in the 2017 squares (Rows 5 and 6), new artefacts on the 2015 rows were relatively few and could not alone have accounted for the observed increase between Rows 4 and 5.
Furthermore, patterns in artefact count by landform in Rows 5 and 6 mirror those in Rows 1-4.
Variations in artefact counts across landform types reflects the geomorphological processes acting on the distribution of the artefact assemblage ( Figure 14): high artefact counts on the Crest landform appear to be linked to the predominantly low-energy deflation of the fine sediment of the grey, artefact bearing unit, that exposed artefacts to survey whilst still leaving large clasts and artefacts broadly in place, creating a high visibility 'pavement' of clasts and artefacts. The Upper Slope has equally good artefact visibility, but with a steeper topography that is more prone to erosion and downslope moment of material, perhaps reflected in the somewhat lower artefact counts in these quadrants. Finally, the Lower Slope and Drainage landforms mark a shift to a predominantly depositional environment that decreases the visibility of artefacts, evidenced by lower artefact counts. While the spatial distribution of the surface artefacts are largely consistent with geomorphological processes acting on the artefact-bearing unit overlying the tufa as described above, it should not be assumed that all the surface artefacts originated from this same unit.
Grey-patinated artefacts mark an origin within the grey unit, and exposure of artefacts to wind erosion removes the patination. After some unknown time at the surface, it would be difficult to visually distinguish previously patinated (i.e., buried) artefacts from those that had never been patinated. Furthermore, the discovery of a tufa-encased handaxe suggests evidence for human activity and artefact deposition before, or contemporaneous with, tufa deposition. This is the inverse relationship to that of the artefacts contained or originating within the sediments overlying the tufa and strengthens the chronotypological assessment of the assemblage as possessing deep time depth, with periods of ESA and MSA activity at the location now deflated into a single surface assemblage, or palimpsest. Comparing distributions of the ESA and MSA artefacts identifiable to a chronotypological cultural period is problematic given the major variation in sample size between the ESA and MSA artefact assemblages, but the MSA artefact distribution (i.e. highest counts in the Crest landform), coupled with the recovery of diagnostic MSA artefacts from the test pits appears closely consistent with an origin of the MSA artefacts in the grey sediment. unit according to the above model.
In summary, the spatial patterning at L0106/0130 can mostly be explained by geomorphological controls acting on the preservation, exposure and visibility of an (MSA) artefact-bearing unit that is preserved primarily in the Crest unit. The null hypothesis that geomorphology, rather than behaviour, controls patterning at the grid scale cannot be rejected, precluding interpretation of spatial patterning of artefacts at this scale purely in behavioural terms. This is not to say, however, that the assemblage can yield no behavioural data. Refitting of flakes and cores from the site would further understanding of the integrity of the artefact assemblage, essential for assessing the constraints and potential for further lithic analyses at the locality (E. Foley, Spry, & Stern, 2017). Already one refitted prepared core and flake has been identified, and a more extensive refitting of prepared flakes and cores is planned (Sinclair et al., In Press). A major challenge for this future analysis of the L0106/0130 material, therefore, will be determining whether separate periods of activity can be distinguished within the surface assemblage using evidence beyond spatial location alone.

Surface Variability and Artefact Distribution in the Wadi Dabsa Basin
The observed artefact distribution within the L0106/0130 grid appears to have been heavily influenced by the geomorphological conditions that control the preservation, exposure and visibility of artefacts. Therefore, therefore variability in surface conditions across the basin are likely similarly determining factors in artefact spatial patterning at the basin scale, a critical issue when considering whether L0106/130 marks a particular, persistently attractive location in the landscape.
At L0106/0130, the quadrants with the highest artefact counts were found on subdued topography covered by fine-grained, artefact-bearing sediments allowing the formation of lag deposits through winnowing of fine material by wind and water, (as well as sporadic highenergy surface sheet flow during rainstorms) by exposing artefacts and other clasts, but with insufficient energy to move them into areas of net sedimentation where they could be buried.
Such deposits, sometimes referred to as deflationary desert pavements (Laity, 2011), are familiar to those working in Palaeolithic archaeology in the arid zone (e.g. Hill, 2000;Jones et al., 2011;Knight & Zerboni, 2018). Similar surface conditions have been observed elsewhere in the Wadi Dabsa basin during exploratory transects, but often with only rare artefacts on their surfaces. Whilst this indicates that the large numbers of artefacts at L0106/0130 may indeed result from a concentration of hominin behaviour, this model would need systematic testing through further refinement of the sediment mapping in the basin.
One avenue for refining this mapping is the application of automatic classification described above. The analysis described in Section 4.2 identified differences in surface sedimentation, including across the artefact collection grids where the landforms mapped in the field broadly correspond to the distinctions made by the automatic classification. These broad classes, however, encompassed significant variations that, at the grid scale, were shown to affect artefact visibility: the Crest and the Upper Slope landforms were contained within Class Nine, whilst Class Ten included both the Lower Slope and Drainage landforms. This lack of resolution may result from the parameters inputted into the classification -of the ten defined classes, seven were automatically assigned to the variability on surfaces other than tufa -and future supervised classification of the tufa surface using high resolution multi-spectral satellite imagery and the groundtruthing observations above, combined with LiDAR DEM, may allow the finer resolution of surface condition. Given that the landform units control artefact patterning in the grid on the tens of metres scale, there may, however, be a limit to what can be defined through remote sensing, and what surface conditions must be logged in the field, once more demonstrating the need for assessment of geomorphological conditions at multiple, nested scales to fully understand the distribution of artefacts across the basin.

Broad-Scale Landscape Stratigraphy, Evolution and Hominin Activity
The landscape at Wadi Dabsa provided a major intersection of features often deemed attractive to hominin population, including abundant raw material sources, water that would have been essential to both hominins and the prey they hunted, and topographic complexity that would have allowed populations to observe, trap, and hunt animals within the basin. Yet, whilst some of these elements would have existed since the initiation of volcanism at this location, the presence of water is likely to have been episodic, responding to environmental changes; the topography that shaped its flow may also have varied through the emplacement of basaltic lava flows. To consider the Wadi Dabsa basin from the perspective of past homininlandscape interactions, we need to determine when these favourable conditions existed, and how and when they changed over time within the timeframe of potential occupation of the region (potentially the last 1.8 Ma).
The geomorphological framework developed above provides a relative stratigraphy of landscape development which will be tested and constrained by further analyses and chronometric dating of the sediments within which the archaeological material is situated.
Firstly, the broad morphology of the basin, defined by basaltic flows, was likely established prior to tufa deposition -all field observations indicate tufa deposition post-dated basalt flow emplacement. It cannot be assumed, however, that no earlier phases of tufa deposition occurred in the basin floor that have been subsequently buried by later tufa deposition or overlapped by basalt flows. Indeed, the >2 m depth of tufa exposed in a wadi the southern edge of the basin illustrates the potential depth and complexity of the tufa deposits. More detailed mapping and periods of tufa deposition.
From the tufa forms, settings and macro-morphologies observed, it is clear that the morphology of the basin encouraged ponding of water and establishment of marshy areas that may have attracted prey and hominins to the basin. Water flowing from constrained channels into the wide basin would have lost energy, as well as potentially becoming dammed at the pinch point at the outflow channel at the western edge of the basin. Carbonate precipitation from calcium-carbonate rich water is encouraged by the presence of organic material: vegetation acts as a substrate for accumulation (Pedley, 1990) and algae aid removal of carbon dioxide from solution to encourage carbonate precipitation (Ford & Pedley, 1996), whilst flowing water encourages turbulent carbon dioxide degassing, again encouraging carbonate precipitation (Ford & Pedley, 1996). Drivers behind periods of tufa formation in this now semiarid to arid basin were climatic, corresponding to periods of increased rainfall and reduced evapotranspiration as documented elsewhere in Saudi Arabia (Kabesh and Abdel-Motelib, 2014), western Namibia (Brook, 1999), north western Spain (Pena, Sancho & Lozano, 2000), Croatia (Horvatincic, Calic, & Geyh, 2000), Israel (Livnat and Kronfeld, 1985); the Grand Canyon USA, (Szabo, 1990), Egypt (Smith, Giegengack, & Schwarcz, 2004;, and Namibia (Stone, Viles, Thomas & Van Calsteren, 2010). From existing speleothem chronologies on the Arabian Peninsula it is likely that this occurred in (but may not have been limited to) interglacial periods (Parker, 2009;Preusser, 2009;Fleitmann et al., 2011;). Disentanglement of the drivers of water flow in the basin will be aided by further analysis of tufa morphology and fabric from macro-morphology to the detailed description of micro-morphology of thin sections using microscopy, in order to understand the depositional environments. In addition, stable isotopic analysis of the carbonate will to provide insights into the moisture sources and temperatures during deposition (Andrews, Pedley, & Dennis, 2000;Gandin & Capezzuoli, 2008).
At some point, the environment that had promoted large-scale tufa deposition ceased to exist within the basin, and incision and weathering of the tufa began. Incision may have been driven by the changing base-level of fluvial erosion, linked to falls in sea level, as well as changing precipitation patterns resulting in more flashy episodes of high energy flow that would have eroded the tufa (cf in the Naukluft Mountains, Namibia, Viles, Taylor, Nicoll, & Neumann, 2007), or a combination of the both.
Situating the artefacts observed at L0160/0130 within the landscape stratigraphy is possible through the relationships between the artefacts and tufa observed on the grids, and this stratigraphy will be chronologically constrained by the ongoing programme of 238 U-230 Th and 39 Ar/ 40 Ar dating. The tufa-encased handaxe suggests that hominins were present within the basin, discarding artefacts before, or contemporary with, at least one phase of tufa formation.
By contrast, the artefacts overlying tufa in the grids indicate activity also took place after tufa deposition ceased. Chronometric dating of the tufa, and the basalt outcrops from which the artefacts may have been manufactured, will constrain this activity within the wider chronology of the occupation of the Arabian Peninsula by providing maximum and minimum ages for the tufa-encased handaxe, and a maximum age for the deposition of the artefacts in the grey unit.
Hominin activity cannot, at present, be associated directly with a specific environment.
It is likely that activity would have taken place during periods in which water was flowing at Wadi Dabsa, but it is also possible that activity could have taken place both before and after tufa deposition in different environmental conditions, perhaps during periods of reduced, but still present, flow. The complexity of the tufa deposits and degrees of weathering observed in the basin means that it is very possible that there were multiple phases of tufa deposition in different areas of the basin, controlled by climatic fluctuations as well as topographic changes such as the damming and rerouting of watercourses by vegetation and development of tufa barrages, all of which need to be the focus of future survey and analysis.

CONCLUSIONS
The geoarchaeological investigations at Wadi Dabsa reported here have demonstrated the fundamental relationship between the visibility of artefacts to present-day survey and their geomorphological setting, relationships which are integral to the interpretation of this locality and all surface records with regards to hominin landscape use.
The 2,970 artefacts observed at L0106/0130 make this the largest Palaeolithic assemblage recorded to date in southwestern Saudi Arabia. The geomorphological observations recorded above indicate that this assemblage is not the product of a single period of activity at the site, but rather a palimpsest of artefacts deposited before and after, and potentially during, phases of tufa deposition at the locality. This extended time depth is also indicated by the techno-typological characteristics of the artefacts within the assemblage. As with all surface assemblages, this time depth must be considered when framing its analysis and interpretation. Patterning of the artefacts at the grid scale provides a powerful demonstration of the interaction between geomorphological processes and the degree to which artefacts are preserved, exposed and visible at a location, and constrains the research questions that can be applied at each scale. Whilst no similarly dense accumulations of artefacts have been observed elsewhere during the exploratory survey of the basin, only the detailed mapping of the tufa surface, and systematic survey of similar geomorphological settings, can determine the potential for this assemblage to mark a particular, persistent focus of hominin activity within the basin.
Investigation of the landscape at the basin scale has allowed the construction of a wider landscape context for the archaeology that will aid future interpretations of activity in the basin.
Whilst surface assemblages are often seen to lack chronological control, the stratigraphic relationships between artefacts and tufa, both underlying and encasing them, observed at Wadi Dabsa allow us to start to link these artefacts within a relative stratigraphy that can be constrained by absolute dating of the tufa and basalt flows. This will allow consideration of the timing of the evolution of the landscape, any responses to regional and global environmental change, and the timing and conditions of hominin activity in the basin over time, and how it may have changed with the landscape.
The geomorphological framework presented in this paper provides the fundamental framework for the understanding of this important locality and its artefact assemblage, through laying out the dynamic processes behind the evolution of the Wadi Dabsa landscape, considering the changing opportunities within the basin for past hominin groups, as well as the ways in which geomorphological approaches have shaped, and will continue to, shape the distribution of the observed artefacts, and, by extension, our interpretations of the Palaeolithic occupation of the Red Sea coastal region.

ACKNOWLEDGEMENTS
We thank HRH Prince Sultan bin Salman bin Abdul Aziz, President of the Saudi