A taxonomic and taphonomic study of Pleistocene fossil deposits from the western Nefud Desert, Saudi Arabia

Abstract Over the past decade, a growing interest has developed on the archaeology, palaeontology, and palaeoenvironments of the Arabian Peninsula. It is now clear that hominins repeatedly dispersed into Arabia, notably during pluvial interglacial periods when much of the peninsula was characterised by a semiarid grassland environment. During the intervening glacial phases, however, grasslands were replaced with arid and hyperarid deserts. These millennial-scale climatic fluctuations have subjected bones and fossils to a dramatic suite of environmental conditions, affecting their fossilisation and preservation. Yet, as relatively few palaeontological assemblages have been reported from the Pleistocene of Arabia, our understanding of the preservational pathways that skeletal elements can take in these types of environments is lacking. Here, we report the first widespread taxonomic and taphonomic assessment of Arabian fossil deposits. Novel fossil fauna are described and overall the fauna are consistent with a well-watered semiarid grassland environment. Likewise, the taphonomic results suggest that bones were deposited under more humid conditions than present in the region today. However, fossils often exhibit significant attrition, obscuring and fragmenting most finds. These are likely tied to wind abrasion, insolation, and salt weathering following fossilisation and exhumation, processes particularly prevalent in desert environments.

Considering the importance of the Arabian Peninsula in hominin prehistory, and the regular occurrence of spatially associated lithic artefacts and fossils at relict lake and river sites within the Arabian interior (e.g., Delagnes et al., 2012;Scerri et al., 2015;Breeze et al., 2017;Groucutt et al., 2018;Roberts et al., 2018;Stewart et al., 2019a), it is crucial that we develop an understanding of the preservational pathways that skeletal elements in these environments might take. Fortunately, the need for taphonomic frameworks with which to study large vertebrate fossil assemblages has long been recognised and a wealth of actualistic and naturalistic landscape studies within tropical settings have been conducted and applied to questions surrounding hominin and animal landscape use, interactions, and behaviour (e.g., Behrensmeyer, 1978;Behrensmeyer and Dechant Boaz, 1980;Blumenchine, 1986Blumenchine, , 1988Blumenchine, , 1989Domínguez-Rodrigo, 2001;Faith and Behrensmeyer, 2006;Faith, 2007). Considerably less attention, however, has been devoted to understanding the taphonomic processes operating in desert environments (e.g., Andrews and Whybrow, 2005;Denys et al., 2007). Deserts experience rapid sediment deflation and depositional hiatuses (Parton et al., 2018) that prolong deposition and facilitate the exhumation of fossils. Once on the surface, fossils in deserts are exposed to extreme temperature fluctuations, frost, strong winds, sandstorms, and rare but intense downpour events, all of which may serve to significantly modify fossil appearance and assemblage composition (Denys et al., 2007). The Nefud Desert is no exception. Extreme diurnal temperature variations cause rocks to fracture and exfoliate, winter temperatures reach below freezing, and strong winds occasionally incite severe sandstorms (Edgell, 2006). How these conditions might affect the preservation of fossils has not been systematically investigated.
Here we report a taxonomic and taphonomic assessment of ten Arabian Pleistocene fossil deposits (Fig. 1). Fossils include both surface and excavated specimens collected from the western Nefud Desert, northern Saudi Arabia. This study adds new taxa to the currently depauperate Arabian fossil record. It summarises the common taphonomic processes that dictated bone assemblage accumulation in Pleistocene Arabian palaeolake settings and reports on those that continue to modify fossil assemblages today.

PREVIOUS INVESTIGATIONS
The site of Khall Amayshan-1 (KAM-1, or "Thomas loc-1") has previously been investigated by various research teams (Thomas et al., 1998;Rosenberg et al., 2013;Scerri et al., 2015). The palaeolake sequence is elevated up to 2.5 m above the basin floor and comprises interdigitated marls, silts, and sands, indicating lake-level fluctuations throughout its existence (Rosenberg et al., 2013;Scerri et al., 2015). Diatoms are dominated by freshwater species and indicate a shallowing of the lake moving up the sequence (Rosenberg et al., 2013). Thomas and colleagues (1998) recovered fossils from the surface of the palaeolake and initially attributed the site to the early Pleistocene based on their interpretation of a mostly extinct faunal assemblage that included Pelorovis sp. cf. P. oldowayensis and tentatively attributed Hexaprotodon (Thomas et al., 1998). However, subsequent investigations of the site using optically stimulated luminescence (OSL) dating found the deposits directly beneath the fossils to range between 117 ± 8 ka and 99 ± 7 ka (Rosenberg et al., 2013). Rosenberg et al. (2013), in their stratigraphic log of the site (their fig. 12, p. 117), indicated that fossils were present in the uppermost marl unit (∼30 cm) and in the underlying calcareous sands (∼50-60 cm) but provided no details on the remains themselves. Scerri et al. (2015) also noted fossils scattered on the surface of the palaeolake deposit and reported 106 Middle Palaeolithic lithic artefacts. Artefacts were made of both local (quartzite) and nonlocal (chert) material and likely reflect several discrete flaking episodes, indicating the repeated use of the lake by late Pleistocene hominins.
A second site just north of KAM-1, called Khall Amayshan-4 (KAM-4), comprises six distinct palaeolake deposits-northwest (NW) lake, northeast (NE) lake, south (S) lake, southwest (SW) lake, southeast (SE) lake, and central lake (Fig. 2). Each palaeolake deposit was associated with a discrete lithic artefact assemblage ranging from Acheulean to Middle Palaeolithic, suggesting that lake formation at the site was not contemporaneous . Abundant lithics, including some refits, indicate that some knapping took place on site .
An additional site located a few kilometres northeast of KAM-4, called Site 16.3, was also investigated by Rosenberg et al. (2013;Fig. 3). The deposit consists of a ∼170-cm-thick laminated white diatomite underlain by white sands dated by OSL to 99 ± 7 ka. Rosenberg et al. (2011Rosenberg et al. ( , 2013 found that the aeolian sediments underlying palaeolake deposits typically predated lake formation by only a few thousand years, suggesting lake onset at Site 16.3 lies close to 99 ka. Diatom analysis demonstrated an abundance of freshwater species and a gradual change toward a dominance of planktonic and facultatively planktonic taxa, indicating a transition to deeper waters moving up the sequence (Rosenberg et al., 2013).

Surveys
The 2013-2016 surveys focused on palaeohydrological features (i.e., relict lakes, wetlands, and rivers) owing to their importance as palaeoenvironmental proxies and high potential to yield archaeological and palaeontological material, as demonstrated by previous research (e.g., Petraglia et al., 2012;Rosenberg et al., 2013;Scerri et al., 2015;Stimpson et al., 2015Stimpson et al., , 2016Stewart et al., 2019a). Palaeolakes and wetlands were identified via remote sensing (see Breeze et al., 2015Breeze et al., , 2016Breeze et al., , 2017 and were investigated on both the fringe of the present-day dune field (via four-wheel-drive vehicles) 2 M. Stewart et al. and deeper into the desert interior (via helicopter, provided the Saudi Geological Survey) (Fig. 1). A total of 40 sites were investigated, most of which yielded significant environmental archives, many with lithic artefacts, and some with fossils (for further details, see Breeze et al., 2017). Ten of the fossil-bearing deposits are reported in the present study. All sites reported here comprise inverted relief outcrops of ancient lake deposits situated in interdunal depressions within the western Nefud Desert, attesting to the significant sediment erosion and deflation of palaeolake sediments in the region. While direct chronological information is only available for KAM-1 and Site 16.3, it is likely that all the deposits date to the Pleistocene. Indeed, all dated palaeolake deposits in the western Nefud Desert can be tied to the humid periods of the middle and late Pleistocene. At each location, pedestrian surveys were conducted across the extent of all exposed palaeolake surfaces, as well as part way up the sand dunes flanking these (for additional details, see Breeze et al., 2017). Given that each site comprised only a small area(s) of exposed palaeolake sediments isolated within interdunal depressions, site surveys were relatively quick and straightforward. Surveys were conducted by "fanning" out across the landscape. This assisted in locating all areas of exposed palaeolake sediments. All fossils, regardless of size, were plotted using a total station or differential global positioning systems (DGPS), given a unique identifier number, and collected. Given that most fossils are surface finds recovered from restricted areas atop palaeolake surfaces with which they can be parsimoniously associated, we treat each site as a single, distinct fossil assemblage but note that time-averaging and mixing may have occurred. Shallow excavations (using trowels and brushes) were conducted at two sites (Site 16.3, WNEF16_1) to recover dense concentrations of fossils that were found eroding out of the palaeolake surfaces. The discovery of fossils eroding out of the fringe of the KAM-4 NW lake deposit prompted the excavation of four 2 × 2 m trenches (Fig. 2). Excavations focused on the palaeolake sediments, which extended to a depth of ∼120 cm before reaching the underlying sterile dune sands. All excavated sediment was sieved through a 2-mm mesh, and all fossils were collected.

Systematic palaeontology
Each fossil specimen was identified to the lowest taxonomic level possible; this process was facilitated by osteological collections housed at the Australian National University (ANU) and University of New South Wales (UNSW), as well as by indirect comparisons (photographs, measurements) with material housed at the Smithsonian Institution National Museum of Natural History (NMNH), USA. Fossil dimensions (length, width) were measured using digital callipers and additional comparative morphometric data were collected following von den Driesch (1976).

Taphonomic analysis
Taphonomic analysis was conducted following established methods and fossils were investigated by eye, geological hand lens (10-20×), and light binocular microscope (up to 40×) (for a detailed discussion of the methods, see Stewart et al., 2019a). Specimen counts were reported in standard quantitative units: number of recovered specimens (NRSP), number of identified specimens (NISP), minimum number of elements (MNE), and minimum number of individuals (MNI). MNE was calculated as the minimum number of skeletal units needed to account for all specimens of a given skeletal unit without taking into consideration the age or side (for bilaterally paired elements) of the animal (Bunn and Kroll, 1986). MNI was calculated similarly but by taking into consideration age and side. Fossils were assigned an animal size class based on a simplified version of the size classes listed in Bunn (1982): small-sized animals (size classes I-II; <100 kg; e.g., gazelle), medium-sized animals (size class III; 100-340 kg; e.g., wildebeest), and large-sized animals (size classes IV-VI; >340 kg; e.g., elephant).
Each fossil was assigned a weathering stage ranging from zero (unweathered) to five (extensively weathered) following Behrensmeyer (1978). Fossils were also assessed for root etching, manganese staining, and sediment abrasion, although their precise frequencies were not always quantified. Root etching was defined as irregular U-shaped grooves, often occurring in dense concentrations, and are the result of the dissolution of bone tissue by plant root chemicals (Fernández-Jalvo and . Manganese staining was defined as the blackening of bone, often appearing patchy and spatially confined, and is typically the product of deposition in water or water-lain sediments but may also arise from dry bacterial growth (Fernández-Jalvo and . Sediment abrasion was defined as the rounding, pitting, and polishing of fossils and the cause of abrasion (e.g., windblown sediment, water-sediment abrasion, trampling) was distinguished based on criteria outlined in the literature (e.g., Olsen and Shipman, 1988;Fernández-Jalvo and Andrews, 2003;Denys et al., 2007). Carnivore tooth marks were identified as pits, scores, furrows, and punctures following Binford (1981), and tooth pit dimensions were recorded following Domínguez-Rodrigo and Piqueras (2003). Long-bone circumference completeness (%) was recorded as type 1 (less than half), type 2 (more than half but incomplete), or type 3 (complete) following Bunn (1982). Longbone breakage was recorded as green (fresh) or dry based on the fracture angle, outline, and edge characteristics described in Villa and Mahieu (1991). When suitable, density-mediated attrition was investigated by Spearman's rank-order correlation coefficient between MNE and wildebeest bone mineral density (Lam et al., 1999). To investigate preferential fluvial transport of bones, fossils were placed into categories based on Voorhies transport groups (Voorhies, 1969). Additionally, the tooth to vertebrate ratio (T/V) was also calculated as a measure of hydraulic winnowing, with values between 0.44-1.5 and 3.12-3.48 indicating limited and strong hydraulic winnowing, respectively (following Behrensmeyer, 1975). Statistical analyses were carried out in PAST (Hammer et al., 2001), and graphics were generated using RStudio (RStudio Team, 2015).
Description: Two small tortoise plastron fragments, between 4 mm and 6 mm in thickness, and three carapace fragments. A single, poorly preserved vertebrae with maximum length and width of 27.7 mm and 21.4 mm, respectively, is also likely that of a tortoise. Given the fragmentary nature of these specimens, little more can be said regarding their taxonomy.
Remarks: Various small bird bones were recovered, but all are fragmented and provide no further diagnostic characteristics.
Description and remarks: Thick eggshell fragments with a mostly smooth surface and pores tightly concentrated in shallow circular to semicircular pits are referred to ostrich eggshell (Fig. 4C). Shell thickness measured ∼2.0 mm and pit diameter measured between 0.5 mm and 0.75 mm, consistent with late Pleistocene ostrich eggshell recovered from sites in Asia (Shani et al., 1989;Blinkhorn et al., 2015) and Africa (Texier et al., 2013). Description and remarks: Teeth are rooted, brachydont, heavily worn, and exhibit a simple occlusal pattern (Fig. 4B). The m3 has a triangular occlusal outline that is narrower to the posterior, whereas the m2 is square and with two buccolingually arranged lophs. The preserved tooth-row length is 4.9 mm, and the complete tooth row length (m2-m3) is estimated to be 7.3 mm. Dipodidae can be discounted based on either their more complex occlusal pattern and/or small size, and in the case of Paradipus by the presence of enamel on the molar buccal and lingual surfaces. Overall, the molar shape and cusp disposition is typical of that of murids. Muridae is a diverse and widespread family of rodents and species of Apodemus, Arvicanthus, Acomys, Rattus, Mus, Nesokia, among others, can be found in Arabia today (Harrison and Bates, 1991), while earlier murid taxa (e.g., Parapeomys, Dendromus) are known from Miocene fossil deposits in eastern Arabia (Thomas et al., 1982;Whybrow and Hill, 1999;Bibi et al., 2013). Nesokia can be discounted based on their more hypsodont teeth with highly developed lophs (see below), whereas extant species of Gerbillinae, Acomys, Apodemus, and Mus can mostly be discounted based on their small size (see Harrison and Bates, 1991). The fossil specimen does, however, fall within the size range of some murid taxa that inhabit Arabia today, such as Rattus and Arvicanthus. However, given that the teeth are heavily worn and do not preserve the loph arrangement, coupled with the poor fossil record for the region, we assign the KAM-4 specimen to the family Muridae.
Description and remarks: Left maxilla with large, fully hypsodont molars and strongly developed lophs is clearly reminiscent of Nesokia (Fig. 4A); however, it cannot be assigned to species on the basis of the material preserved. M1 and M2 have a combined length of 6 mm. Individual cusps are still visible, suggesting the tooth is in early stages of wear.
Description and remarks: Tooth enamel fragments were recovered and are consistent with elephantids, but the fragments are small and poorly preserved and little more can be said regarding their taxonomy.  (Fig. 4D). It is far too large to be one of the wild asses, and its length is consistent with the large African Equus greyvi, some species of extinct Eurasian stenonoid horses (i.e., Equus altidens, Equus suessenbornensis), and wild horse (Equus ferus ferus) (Supplementary Table S1). A principal component analysis (PCA) using five metric variables (Supplementary Table S1) found the KAM-4 specimen to be most similar to E. altidens and E. grevyi (Fig. 5). However, given the poor preservation of the specimen and the fact that the PCA analysis did not discriminate well among taxa on either axis, we refer the KAM-4 specimen to genus only.

6
M. Stewart et al.  Description and remarks: A metacarpal recovered from KAM-1 (KAM-1/F34) is referred to a hippopotamid (Fig. 6B). The specimen is bleached white and exhibits cortical exfoliation probably due to wind erosion. The proximal articular surface is triangular in outline and dips to the palmar, while the lateral side extends substantially farther to the proximal than does the medial side, consistent with the third metacarpal of the right forefoot. Its size fits well within the range of modern Hippopotamus amphibius (Supplementary  Table S2) and is far too large to be considered Hexaprotodon liberiensis. Thomas et al. (1998) also reported numerous hippopotamid remains from KAM-1. Only a single element was described-a fifth metacarpal (JM 301) "morphologically similar to that of Hippopotamus" but "much more slender" and tentatively referred to Hexaprotodon (Thomas et al., 1998, p. 148). The authors were probably somewhat influenced by their early Pleistocene interpretation of the site and belief that Hexaprotodon occupied the region up until the Holocene (Thomas et al., 1998). Thomas et al. (1998), however, were incorrect. Recent OSL dating found that the KAM-1 lake deposit dated to the late Pleistocene (Rosenberg et al., 2013), while Hip. amphibius is the only known hippopotamid to have occupied Africa and southwest Asia during this time (Stewart et al., 2019b). The material described here, and probably the material reported by Thomas et al. (1998), can be referred to Hip. amphibius.
Hippopotamid remains are relatively abundant at Site 16.3 (n = 23) and consist primarily of fragmented appendicular and axial remains. A well-preserved and complete right astragalus (16.3/51) is too large and squamate to be that of a Hexaprotodon sp. and is consistent in overall shape and size to Hip. amphibius (Fig. 6A (Hooijer, 1950). A smaller and more obliquely positioned distal metapodial (16.3/33) may represent a second or fifth metapodial and, if correct, is too large to be considered Hex. liberiensis (Supplementary  Table S4). Two robust proximal phalanges (16.3/35, 82) can be attributed to either the second or third digits, one of which can be referred to as a juvenile on the basis of its unfused (and absent) proximal epiphysis (Fig. 6C). A distal phalanx (16.3/334A) with a greatest breadth measuring 27 mm and a poorly preserved intermediate phalanx (16.3/ 334B) are also referred to Hippopotamus. Given the late Pleistocene age of the lake with which the fossil remains are associated (Rosenberg et al., 2013) and similarities between the hippopotamid fossils and Hip. amphibius, we attribute the Site 16.3 remains to the extant Hip. amphibius accordingly.
Description and remarks: A right astragalus (KAM4/010) recovered from KAM-4 NW lake is similar in overall size and morphology to Hip. amphibius (Fig. 6E). The specimen is, however, poorly preserved and distorted, making a specific assignment difficult, and we refer this specimen to Hippopotamus sp. accordingly.
Description and remarks: Two specimens recovered from KAM-4 NE lake are probably a hippopotamid but are too incomplete to be confidently attributeda large globular femur head (Fig. 6F) with a diameter of 77.6 mm and a distal fibula.  Description and remarks: Numerous fossils represent a large buffalo, and while most of the specimens are fragmented, some are complete enough to provide taxonomic insight. A distal phalanx (16.3/unnumbered; Fig. 4G) with a curved plantar surface and poorly developed processes extensorius and tuberculum flexorium is consistent with Syncerus and Pelorovis, but not Bos (Peters, 1986;Thomas et al., 1998). Its length falls outside the size range reported for the extant Syncerus caffer (Supplementary Table S5), and in this respect it more closely resembles the larger extinct Syncerus antiquus (placed within Syncerus rather than Pelorovis following Gentry [2010]) and Pelorovis. Indeed, its size closely matches the Pelorovis sp. cf. P. oldowayensis distal phalanx reported from KAM-1 (Thomas et al., 1998; but see below). Two calcanea (16.3/291, 436) are consistent with those of Bos and S. antiquus in size and in exhibiting a short sustentaculum tali and well-developed processes coracoideus, features that are absent in S. caffer (Gentry, 1967; Peters,   Table S5). We therefore refer the Site 16.3 large bovid material to Syncerus and note that the late Pleistocene age of the site makes attribution to S. antiquus most likely.
A large bovid distal humerus (KAM-1/F38) recovered from KAM-1 is provisionally referred to S. antiquus, as its size exceeds those of S. caffer and Bos (Supplementary Table S5). Thomas et al. (1998) also reported remains of a large buffalo from KAM-1, which were provisionally attributed to P. oldowayensis. However, as discussed elsewhere (Stewart et al., 2019b), more recent dating of the KAM-1 palaeolake to the late Pleistocene is suggestive of the younger species S. antiquus. The specimen described here, and probably those reported by Thomas et al. (1998), can be provisionally referred to the late Pleistocene S. antiquus.
Description and remarks: A single complete M2 or M3 (WNEF16_1; Fig. 4E) differs from Bubalus in its less complex occlusal outline, while the infundibulum of the distal loop exhibits a marked internal fold as seen in Syncerus and Pelorovis, and sometimes in Bos, albeit typically less pronounced (Martínez-Navarro et al., 2007). The small and simple basal pillar is consistent with Pelorovis and S. antiquus, whereas those of other large Bovinae are typically more complex and medially pronounced (Gentry, 1967). That said, rare examples of S. caffer with relatively short basal pillars were encountered when examining comparative material. Its size (occlusal length = 31.6 mm, posterior lobe width = 21.3) is comparable to S. caffer and S. antiquus. The former is unknown from north of the Sahara, while the latter populated southwest Asia during the Pleistocene (Stewart et al., 2019b). Therefore, we provisionally attribute the WNEF16_1 specimen to Syncerus.
Material: Site 16.3right horn core (16.3/416). Description and remarks: A small right horn corn (Fig. 4H) with backward curvature, slight torsion, some mediolateral compression, and flattened lateral side is referred to gazelle (Gentry and Gentry, 1978). The transverse and anteroposterior diameters measured 21.9 mm and 28.1 mm, respectively. The size and compression index (77.9 mm) are consistent with various species of gazelle, including Gazella dorcas and Nanger soemmeringii (Gentry, 1964;Gentry and Gentry, 1978). Material: WNEF16_25horn core (WNEF16_25). Description: A single fragmented but well-preserved and near-complete horn core (Fig. 4J). The horn core exhibits strong mediolateral compression and is without torsion, keels, or ridges, matching those of the living Hippotragus spp. The preserved length along the anterior of the horn core is 370 mm, but its original length may have been substantially longer. Given the fragmentary nature of the specimen and lack of morphometric data, only a generic assignment is possible.
Description and remarks: Small, straight semicircular horn cores without torsion attributable to the genus Oryx were recovered from numerous sites. The basal horn core dimensions of the KAM-4 NW lake specimen (KAM-4/014) are larger than comparative material of the Arabian endemic Oryx leucoryx, smaller than Oryx gazella and Oryx beisa, and within the range of the desert-adapted Oryx dammah (Supplementary Table S6). The KAM-4 NW lake specimen differs from the latter in having more widely diverging horn cores, but in this regard is comparable to all other living species. It differs from the fossil specimens from the nearby middle Pleistocene site of Ti's al Ghadah in being less robust and having more divergent horn cores (Thomas et al., 1998;Stimpson et al., 2016). Stimpson et al. (2016) suggested that the large and robust Oryx fossils from Ti's al Ghadah might represent a now-extinct large-bodied Pleistocene form of O. leucoryx. Given the intermediate size of the specimens described here, it is possible that the KAM-4 NW lake specimen represents a transitional form between the large-bodied middle Pleistocene Oryx and the smaller, extant Arabian O. leucoryx.

Taphonomic analysis
Various taphonomic considerations and their role in shaping the Arabian Pleistocene fossil assemblages are discussed in the following sections. The sites with limited fossil material (i.e., WNEF16_20, WNEF16_22, and WNEF16_25) are excluded from the discussions unless otherwise specified. Full results of the quantitative and qualitative taxonomic and taphonomic analysis are presented in Tables 1-3 (raw data in Supplementary Tables  S7 and S8) and complete skeletal inventory in Supplementary Table S9.

10
M. Stewart et al.  Thomas et al. (1998). Taxa listed in bold recovered during excavation.
Khall Amayshan-1 (KAM-1) Fossils were recovered from within and atop the uppermost marl unit at KAM-1 (see Rosenberg et al. [2013] and Scerri et al. [2015] for detailed descriptions of the entire sequence). It should be noted that Thomas et al. (1998) also collected fossils at this site that are not included in this analysis, and, therefore, the fossils studied here may not be an accurate reflection of the original taphocoenosis. To the best of our knowledge, no other fossils have been collected from the other sites analysed here. Seventy-six fossils were recovered at KAM-1, representing four taxa and matching those reported by Thomas et al. (1998). Small vertebrate remains are absent, and small fossils in general are underrepresented at the site (Fig. 7). Fossils range in colour from light brown and grey to bleached white (Fig. 8E), and several specimens exhibit manganese staining. No fossils were unweathered, and most exhibited stage 2 and 3 weathering (n = 31, 81%). For two-thirds of the fossils it was impossible to determine weathering stage due to significant abrasion and cortical exfoliation, and this is the case for most assemblages analysed here. Wind abrasion appears to have been particularly pronounced in the KAM-1 fossil assemblage, and all specimens exhibit some degree of rounding, pitting, and polish, with some fossils having been reduced to near pebbles (Fig. 8F).
Limb-bone elements dominate the assemblage (n = 25, 76% of NISP) and long-bone midshafts make up most of these (n = 15, 45% of NISP). Most midshafts have fracture angle, outline, and edge characteristics consistent with dry fracturing (n = 14, 74%), and most retain their complete circumference (n = 12, 67%). A handful of specimens exhibit plastic deformation (i.e., crushing), suggesting that the site once lay under a significant weight of sediment, and rare instances of plastic deformation were also observed in the KAM-4 and Site 16.3 assemblages (e.g., Fig. 8I). Easily transported elements (Voorhies groups I and II) are best represented, and the tooth to vertebrae ratio (T/V = 2.5) falls between the values reported by Behrensmeyer and Dechant Boaz (1980) for minimal and strong hydraulic sorting.

Khall Amayshan-4 (KAM-4) NW lake
Renewed investigations at KAM-4 revealed previously unrecognised palaeolake deposits (Fig. 2), and fossils were recovered from the surfaces of three of these: KAM-4 NW lake, KAM-4 NE lake, and KAM-4 S lake. The KAM-4 NW  lake fossils were found concentrated on a sandy area between two of the excavated trenches (Trenches 1 and 3) that is interpreted as an ancient sandspit that led into the lake. Twenty-two fossils were discovered during excavations and represent the first time that fossils have been recovered from within stratigraphy at the site. The KAM-4 NW lake assemblage is one of the largest and most taxonomically diverse sites reported here, with a total of 596 fossils representing eight taxa. Remains of small vertebrates-which comprise rodents, birds, tortoises, and small bovids-were recovered from atop the deposit and during the excavation of palaeolake sediments and were relatively more abundant in the latter. Medium-sized animals dominate the assemblage, and it seems likely that most of this material is attributable to one or several medium-sized bovids (e.g., Oryx). Large-sized animals are very rare, but recovered materials indicate the presence of Hippopotamus and an elephantid.
Fossils range in colour from grey and light brown to maroon, the latter being more characteristic of the excavated assemblage ( Fig. 4A and B). Manganese staining and root etching were occasionally observed, and two bone fragments recovered during excavation are blackened and appear burnt, although additional analyses are needed to confirm this. Likewise, the KAM-4 NW lake assemblage appears significantly impacted by wind abrasion, with all fossils exhibiting rounding, pitting, and polish, and in some instances significant exposure of cancellous bone (e.g., Fig. 8C). Interspecimen differences in abrasion, more characteristic of wind abrasion, were also observed (e.g., Fig. 8D). The fossils for which weathering could be reliably determined range between weathering stages 0 and 4, with most falling into the latter two stages (n = 153, 74%), and a Spearman's test found a significant positive correlation between specimen size (length) and weathering stage (Spearman's rho, r s = 0.166, P < 0.018). A small number of fossils, mostly appendicular elements, show evidence of carnivore gnawing (n = 8, 1.3%).
the assemblage has undergone minimal density-mediated attrition. This is also borne out by the lack of correlation between medium-sized bovid long-bone survivorship (MNE) and wildebeest bone mineral density (Spearman's rho, r s = −0.192, P = 0.369; Supplementary Table S11). This should, however, be interpreted with caution, given the heavily fragmented nature of the limb-bone portion of this assemblage (see Marean et al., 2004;Clark, 2019). There is a clear bias toward easily transported elements (Voorhies groups I and II) and hydraulic winnowing is also supported by the low tooth to vertebrae ratio (T/V = 0.03). Very small animal remains (rodents, birds, and tortoise)which are relatively abundant at the site-also support the preferential transport of easy-to-transport remains. However, the bulk of these findings (n = 13, 72.2%) were recovered from the excavated portion of the assemblage, suggesting that once exposed, microfaunal remains are transported or destroyed. Alternatively, microfaunal remains may have been simply missed during pedestrian surveys, and more intensive surveys should be undertaken in the future to test for this.

Khall Amayshan-4 (KAM-4) NE lake
The NE lake fossil assemblage is the smallest and least diverse of the KAM-4 fossil deposits, with only 17 fossils representing two taxa. Small vertebrates are absent, mediumsized animals make up about two-thirds of the assemblage, while large animals make up the remaining third and indicate the possible presence of Hippopotamus.

Khall Amayshan-4 (KAM-4) S lake
Sixty-seven fossils were recovered from the KAM-4 S lake representing at least two taxa. Small vertebrates are represented by a few small bovid remains as well as a single piece of ostrich eggshell, which, given that ostriches inhabited the Arabian Peninsula until historic times (Potts, 2001), may be of more recent origin. Medium-sized mammals make up the remainder of the fossil material, and it appears that much of this material is attributable to one or several medium-sized bovids (e.g., Oryx).
Fossils range from dark brown to maroon, exhibit rounding and pitting characteristic of wind abrasion, and occasional root etching and manganese staining (e.g., Fig. 8G). Fossils appear less weathered than at the other  Table S10). Most midshaft fragments are represented by pieces retaining less than half their circumference (n = 4, 80%), and most of these have fracture angle, shape, and edge characteristics consistent with dry fracturing (n = 6, 75%).

Site 16.3
Follow-up investigations at Site 16.3 showed additional sandy marl sediments situated above the upper diatomite unit that were not recognised by Rosenberg et al. (2013), perhaps because they were only preserved in some sections of the site. These consisted of silty sediments capped by a sandstone-like sandy marl and are interpreted as representing the final desiccation phase of the lake, meaning that the sequence records an entire dry-wet-dry cycle. The recovery of fossils from areas where the sandstone had eroded suggests that bones were deposited as the lake was drying. Most fossils were found on the surface of the palaeolake and concentrated in the southern half of the deposit, while at the centre of the palaeolake a single Syncerus skeleton was found eroding out of the surface of the palaeolake sediment ( Fig. 9A and B). The Site 16.3 fossil assemblage is the largest assemblage investigated here, with 615 fossils recovered representing six taxa, although the Site 16.3 lake is comparatively large and a non-insignificant number of these fossils can be attributed to a single large bovid. Small vertebrates are again rare, and although medium-sized mammals are most abundant, only a single horn core tentatively attributed to Oryx could be more precisely identified. Large-sized mammals are abundant and include Hippopotamus amphibius (NISP = 23, MNI = 2) and Syncerus sp. cf. S. antiquus (NISP = 53, MNI = 1). A dense concentration of Syncerus and unidentifiable bone fragments (n = 90) was recovered from an area of ∼3 m 2 at the centre of the palaeolake exposure and, given their propinquity and similar appearance and preservation, are interpreted as a single Syncerus skeleton.
Fossils range from pale yellow to brown, and most specimens exhibit pitting, rounding, and polish consistent with wind abrasion (e.g., Fig. 8H). Some fossils retain sandstone cemented to their surface, consistent with the hypothesis that the fossils eroded from the uppermost sandstone unit preserved in some sections of the site. Manganese staining (e.g., Fig. 6D) and root etching were occasionally observed, with some instances of the latter being shallow and differing in colour from the surrounding bone (e.g., Fig. 8H), indicating that etching occurred more recently and probably following fossilisation (Fernández-Jalvo and . Fossils show weathering stages 0-4, with stages 2 and 3 being most common (n = 172, 63%) and, as at KAM-4 NW lake, there is a positive correlation between specimen size (length) and weathering (Spearman's rho, r s = 0.227, P < 0.001).
Thirty-nine elements (6.3%) show evidence for carnivore gnawing, including 28 measurable tooth pits with major and minor axes ranging from 1.6 mm to 9.7 mm and 1.5 mm to 7.0 mm, respectively (Supplementary  Table S10). Two tooth pits on epiphyseal and diaphyseal fragments measured 7.7 × 7.0 mm and 7.5 × 6.7 mm, respectively, being most consistent with gnawing by a large carnivore such hyena or lion (cf. Delaney-Rivera et al., 2009). Appendicular elements are abundant (n = 64, 43% of NISP), as too are carpals/tarsals (n = 21, 14% of NISP), and fragile vertebrae (n = 21, 14% of NISP) and ribs (n = 17, 11% of NISP). Most long bones retain less than half their original circumference (n = 17, 65%) and exhibit fracture patterns consistent with dry fracturing (n = 16, 62%). Abundant easily transported elements (Voorhies groups I and II) and the low tooth to vertebrae ratio (T/V = 0.0) suggest that the assemblage may have undergone some hydraulic winnowing.

WNEF16_1
Similar to the individual Syncerus skeleton recovered at Site 16.3, the WNEF16_1 assemblage comprises a single large buffalo skeleton-provisionally attributed to Synceruseroding out of the palaeolake surface in a dense 1-m 2 concentration. While the bulk of the assemblage consists of unidentifiable bone fragments, most appear to be from a large animal, are similar in preservation and appearance, and are interpreted as representing a single individual. Fossils are sandy in colour and exhibit occasional superficial root etching. Interestingly, the WNEF16_1 assemblage exhibits the full spectrum of weathering stages, with most falling into stages 2 and 3 (n = 146, 83%), and demonstrates, rather importantly, that significant variations in weathering can occur across very small spatial and temporal scales. Again, there is a positive correlation between specimen length and weathering (Spearman's rho, r s = 0.311, P < 0.001). The assemblage is very heavily fragmented, with almost the entire assemblage comprising small midshaft (n = 91, 43% of NRSP) and unidentifiable (n = 111, 53% of NRSP) bone fragments. All long bones have been fragmented such that they retain less than half their circumference (n = 91, 100%), and most display fracture angle, outline, and edge characteristic consistent with dry fracturing (n = 42, 67%).

Species representation and palaeoenvironmental implications
This study represents the first widespread taphonomic study of fossil faunal assemblages in the Arabian Peninsula and an important initial step in elucidating the processes governing fossil preservation and site formation in this unique region. Overall, the fauna identified here are comparable to other Pleistocene sites in the western Nefud Desert and include three previously unrecognised taxa in the Arabian Pleistocene fossil record-Nesokia sp., Muridae sp., and Hippotragus sp. The fauna from sites KAM-1 and Site 16.3, notably Hippopotamus, Equus, and Syncerus, is consistent with palaeoenvironmental data that demonstrate that at times during Marine Isotope Stage 5 (MIS 5; ca. 130-74 ka) the western Nefud Desert was significantly more humid, with expansive grasslands and interdunal lakes (e.g., Rosenberg et al., 2013;Jennings et al., 2015;Breeze et al., 2016;Groucutt et al., 2018). The presence of Hippopotamus and elephant at KAM-4 NW lake suggests a similar environmental scenario. While less can be reliably said about the other sites, the small sample sizes and lack of chronological data, the presence of taxa such as Hippotragus sp. (as at WNEF16_25), and the accompanying palaeolake sediments and taphonomic evidence (see next section of this Discussion) all suggest that these assemblages were deposited under conditions more humid than characterize the region today.
Overall, small vertebrates are drastically underrepresented in the Arabian fossil record. Even today, the hyperarid Nefud Desert is host to an array of small birds, reptiles, rodents, lagomorphs, and carnivores (see Harrison and Bates, 1991), and greater precipitation and grassland availability during interglaciations (i.e., an increase in primary productivity) would have only increased small animal species richness and abundance (the "energy-richness hypothesis"; Hawkins et al., 2003). The scarcity of small animal remains may reflect a sampling bias against smaller and more difficult to spot fossils during pedestrian surveys and is, in part, supported by the more abundant and diverse small animal assemblages in the excavated portion of the KAM-4 NW lake assemblage, as well as at the nearby sites of Al Wusta  and Ti's al Ghadah (Stimpson et al., 2016;Stewart et al., 2019a). Alternatively, this bias might be the result of taphonomic processes that preferentially destroy smaller bones, and other researchers have noted that small animals have a reduced chance of being preserved in the fossil record (Behrensmeyer and Dechant Boaz, 1980;Haynes, 1988;Andrews, 1990). Moreover, while microfauna (small mammals, birds, and reptiles) may be accumulated by large carnivores (e.g., Kruuk, 1976;Williams et al., 2018), they are more typically accumulated by smaller mammals and raptors  and, as such, have vastly different taphonomic histories. The small sample size in the present study precludes any detailed analysis, but future studies may elucidate the accumulation processes of microfauna in lakeside environments in Arabia.
Medium-sized bovids dominate the Arabian Pleistocene fossil record, and to date, remains of Oryx, Hippotragus, and Alcelaphinae have been identified (McClure, 1984;Stimpson et al., 2016;Groucutt et al., 2018;Stewart et al., 2019a). Oryx is particularly well represented in the Arabian Pleistocene fossil record, and analysis of the morphometry and morphology of some of these remains suggest that they may represent a now-extinct species, as has been alluded to for the remains recovered from the middle Pleistocene site of Ti's al Ghadah (Thomas et al., 1998;Stimpson et al., 2016). Large-sized animals include Syncerus, Hippopotamus amphibius, and elephant, and these taxa have been identified at other nearby sites (Stimpson et al., 2016;Groucutt et al., 2018), as well as farther south (McClure, 1984). Hippopotamus is relatively common in the Arabian fossil record, perhaps reflecting the research focus on palaeolake deposits.

Taphonomic analysis
The taphonomic study of the fossils also provided some key insights into the environmental and depositional histories of the assemblages. The dark brown/maroon colouration of some of the assemblages (as at KAM-4 NW lake; Fig. 4A) is indicative of deposition in oxygenated and biologically active soils, while manganese staining commonly reflects deposition in water or water-lain sediments and is consistent with the lacustrine setting in which the fossils were recovered (e.g., Fernández-Jalvo and Andrews, 2016; Fig. 8G). Root etching was occasionally observed and, in some instances, was deep and similar in appearance to the cortical bone, suggesting that etching occurred before fossilisation. In contrast, some root marks (as at WNEF16_1 and Site 16.3) were more surficial and differed in colour from the surrounding fossil bone, indicating etching occurred more recently (Fernández-Jalvo and Andrews, 2016; Fig. 8H). The latter is interesting in that it suggests that fossils remained near the surface during subsequent humid phases, probably facilitated by slow sedimentation rates and deflation during more arid intervals.
The varied weathering pattern observed at large and taxonomically diverse sites such as KAM-4 NW lake and Site 16.3 probably reflects attritional bone accumulations in which carcasses were continually added to the landscape. Lakes are attractive features in semiarid settings and provide a constant input of bones variably affected by weathering processes through time (Behrensmeyer, 1978;Haynes, 1988). Interestingly, the single Syncerus skeleton at WNEF16_1 exhibits the full spectrum of weathering stages, demonstrating that significant variations in weathering can occur across very small spatial and temporal scales. During our surveys, we also observed a camel carcass that, in a single cranium, was partially fleshed below the surface of the sand but exhibited stage 2 weathering above ( Fig. 9C and D). Andrews and Whybrow (2005) also observed this, albeit not to the extent reported here, when monitoring a single camel carcass in a desert environment over a 15-year period; the authors attributed this to shifting sands covering and protecting small bones such as ribs and vertebrae. Likewise, smaller fossils are less weathered in some of the sites analysed here, although this is may be due to other factors preferentially protecting smaller bones (e.g., vegetation, lake water). While poor preservation and extensive wind abrasion make commenting on the duration of exposure problematic, we note that the scarcity of unweathered bone suggests that burial was typically slow or that bones were regularly re-exposed before fossilisation.
Wind abrasion is perhaps the most significant process affecting the Arabian surface fossil deposits and is known to be particularly prevalent in desert regions where there is abundant sandy substrate and little in terms of vegetation to stabilise sediments (Fernández-Jalvo and Andrews, 2016). The KAM-1 and KAM-4 fossil assemblages are most significantly impacted by wind abrasion, with all fossils showing some degree of rounding, pitting, and/or polish consistent with abrasion by windblown sand. While the extent of wind abrasion probably largely reflects time since exhumation, the position in the landscape relative to the surrounding dunes may also play a role, with dunes acting as either wind funnels or protective barriers. It is also possible that wind has spatially redistributed fossils in the landscape. Indeed, observations in a similar desert environment (Fuego-Patagonia) found strong winds significantly redistributed lithic artefacts in the landscape, many of which exceeded the size and weight of fossils reported here (Borrazzo, 2016). Such a process could explain the contrast between the densely concentrated Syncerus skeleton and the sparsely concentrated loose surface fossils at Site 16.3 (Figs. 3 and 9B).
Carnivores probably also played a role in distributing bones across the landscape, and despite the paucity of carnivore remains in the present study, their presence is inferred from tooth-marked bone (e.g., Fig. 8B). Large tooth pits, tooth notches, and green-fractured bones attest to the role of large bone crushing carnivores, such as hyena, in the accumulation of at least some of these assemblages (KAM-4, Site 16.3). That said, tooth-marked specimens are typically very rare and/or assemblages too small to provide more detailed insights into the role of carnivores as bone accumulators. The individual Syncerus skeleton from Site 16.3 does, however, appear to have been primarily accumulated by large carnivores, as suggested by gnawed ribs, epiphyses, and midshaft fragments. The scarcity of carnivore tooth marks here may be due to the various aforementioned attritional processes and greater evidence for carnivore processing at the well-preserved site of Ti's al Ghadah (Stewart et al., 2019a) lends some credence to this hypothesis.
Overall, skeletal part representation is poor, and unidentifiable fragments dominate across all assemblages (making up two-thirds of all material analysed here), followed by midshaft fragments, and small compact bones. The predominance of dense elements suggests that these assemblages have undergone significant density-mediated attrition, although the primary taphonomic process controlling for this remains difficult to assess. The scarcity of dense dental remains in the Arabian fossil deposits is at odds with this, but Behrensmeyer (1978) noted teeth in the Amboseli National Park, Kenya, rapidly disintegrated as a result of surface desiccation (see also Conard et al., 2008). Such a process may explain the scarcity of dental remains in the Nefud Desert fossil assemblages and is perhaps best exemplified at WNEF16_1, where, except for a single complete molar, dental remains are represented by heavily fragmented and fragile enamel 18 M. Stewart et al.
and dentine splinters. The KAM-4 NW lake and Site 16.3 assemblages appear to have been less influenced by densitymediated attrition, as evidenced by abundant ribs, vertebrae, and small fragile animal remains. Overall, the results suggest that while density-mediated attrition was probably significant, it may not fully explain the taxonomic and skeletal part representation across all sites. The clear bias towards easily transported elements and low tooth-to-vertebrae ratios suggest that some of the assemblages may have been influenced by hydraulic winnowing, perhaps as a result of lake-level fluctuations. Indeed, previous studies at KAM-1 found that the lake underwent frequent lake-level fluctuations (Rosenberg et al., 2013) and preliminary analysis of the KAM-4 NW lake sediments have identified fossil-bearing units (3a, 3b, and 4b; Fig. 2) with abundant allogenic sands, low relative CaCO 3 , and "rip up" marls (i.e., eroded and reworked marl fragments) indicative of a dynamic lake system. However, as highlighted earlier, the low number of teeth at these sites may also be explained by the preferential destruction of dental remains in lacustrine settings, and the values reported here are similar to those for nontransported bone assemblages from lake margin habitats in the Amboseli National Park, Kenya (Behrensmeyer and Dechant Boaz, 1980). While changes in local landscape hydrology are likely to have altered bone accumulations in some form, their specific influence is not currently discernible.
Bone breakage is ubiquitous, and most long bones retain less than half of their original circumference and exhibit fracture patterns consistent with dry fracturing. The KAM-1 assemblage is unique in having more complete long bones, which may reflect the more resilient bones of the larger animals preserved at the site (Hippopotamus, Syncerus). Bone fragmentation may have also been promoted by the physiochemical stresses associated with repeated temperature and moisture fluctuations, which are known to accelerate bone degradation and weathering (Behrensmeyer, 1978;Haynes, 1988;Pokines et al., 2018), and it has been argued that the poor preservation of fossil assemblages within the South African interior is the result of bone deposition within river, lake, and spring margins (Hutson, 2018). Dense, highly mineralised fossils, such as fossil teeth, have been found to be disproportionately affected by drastic temperature fluctuations, leading to their underrepresentation in some fossil assemblages (Conard et al., 2008).
More recently, fragmentation may have been exacerbated by insolation and salt weathering of fossils, processes particularly pronounced in desert environments and thought to be the source of much of the fragmentary geological material that fills the Nefud Desert today (Edgell, 2006). The former involves the differential expansion and contraction of rock constituents brought on by extreme diurnal temperature variations (Gutiérrez, 2005), whereas the latter involves the precipitation and expansion of salt crystals within rock cracks (Cooke and Smalley, 1968;Cooke, 1981). Palaeolakes in the Nefud Desert contain abundant carbonates and sulphates see also Fig. 4K), and salt crystals were identified in the WNEF16_25 horn core that was found fragmented but retaining its shape in the landscape (Fig. 4J). Bauer (1988, p. 45) noted that animal teeth from caves in northeast Saudi Arabia were "split and crumbling" due to the formation of gypsum crystals within the lamellar bone. Mehlman (1987) went as far to suggest that salt weathering may be the most significant factor driving the breakdown of fossil material at Lake Eyasi in East Africa.
Finally, there are several instances of plastic deformation, and similar observations were made nearby at the site of Al Wusta . This could be the result of significant lake sediment buildup or dune migration over the sites. The latter is supported by the discovery of buried or partially buried palaeolakes beneath the present-day sand dunes (Schulz and Whitney, 1986;Breeze et al., 2017), while the varying positions of lakes at KAM-4 also suggest that the dune topography shifted substantially between phases of lake formation. Taken together, processes such as salt weathering, insolation weathering, and plastic deformation appear to have significantly modified the Arabian fossil assemblage appearance and composition and are likely to continue to do so into the future.

CONCLUSIONS
The fauna identified in this study strongly support the idea that at times during the Pleistocene, the western Nefud Desert boasted large grasslands and permanent river and lake systems. Notable is the presence of Hippopotamus, Syncerus, and elephants at several noncontemporaneous sites in the Nefud Desert (Thomas et al., 1998;Stimpson et al., 2016;Groucutt et al., 2018) and elsewhere in Arabia (McClure, 1984), demonstrating that pluvial phases were periodically wet and vegetated enough to allow for the dispersal of large water-dependent taxa into the Arabian interior. Climatic amelioration also facilitated the dispersal of hominins  and large carnivores (Stimpson et al., 2016;Stewart et al., 2019aStewart et al., , 2019b into Arabia, with the presence of the latter inferred here from tooth-marked bone. While the role that carnivores played in the accumulation and modification of the Arabian fossil deposits remains difficult to assess, it seems likely it was greater than the current observations suggest. The same seems likely for hominins, given the extensive Arabian archaeological record  and likely evidence for carcass processing by hominins at the well-preserved site of Ti's al Ghadah (Roberts et al., 2018;Stewart et al., 2019a).
The sedimentological and taphonomic data also indicate that bones were deposited in conditions more humid than characterize the region today. Most striking about the Arabian fossil assemblages is their significant attrition, obscuring and fragmenting most finds. Fragmentation of bone may have been further exacerbated by the physiochemical stresses associated with lacustrine environments, principally repeated wetting and drying events. Perhaps even more significant in the assemblages studied here are post-exhumation processes such as wind abrasion, insolation, and salt weathering. Studies in hyperarid environments such as Arabia and the Sahara are further confounded by rapid sediment deflation and depositional hiatuses that promote the mixing of temporally discrete deposits. Indeed, the lack of evidence of butchery in assemblages with tightly associated fossils and lithic artefacts (as at WNEF16_1) suggests that there may not be an immediate temporal correlation between spatially associated objects. While more work is needed to investigate the impact of desertic processes on fossils, the fossils examined here clearly demonstrate that these processes are highly destructive and have drastically altered fossil appearance and assemblage composition.