Introduction

The transition from handheld bifacial large cutting tools to Levallois prepared core technology is central in defining the onset of the Middle Stone Age (MSA) in Africa and the Middle Palaeolithic (MP) in Eurasia (e.g. Tryon et al., 2005; Adler et al., 2014; Moncel et al., 2020; Zaidner & Weinstein-Evron, 2020). While Levallois encompasses many distinct methods of core preparation for producing an end-product of predetermined shape, one has received particular attention in recent debates; Nubian Levallois technology has been the focus of numerous studies examining human adaptation strategies, dispersals, and demographic interactions (e.g. Goder-Goldberger et al., 2016; Olszewski et al., 2010; Rose et al., 2011; Van Peer, 1998). Nubian Levallois cores (henceforth NLCs) are a type of preferential core for producing elongated pointed end-products (Usik et al., 2013; Van Peer, 1992), characterised by a pointed morphology but differing from other Levallois production methods by the use of both proximal (preferential) and distal (preparation) platforms (Hallinan et al., 2022a). This creates a distinctive median distal ridge on the flaking surface, formed by preparatory removals from the distal end (Type 1), laterals (Type 2), or a combination of both (Type 1/2) (Guichard & Guichard, 1965; Usik et al., 2013).

Unlike the vast spatiotemporal distribution of other Levallois methods across Africa, Europe and southwest Asia (Dibble & Bar-Yosef, 1995; Lycett & von Cramon-Taubadel, 2013), the narrower occurrence of NLCs in space and time has raised questions about the significance of this method (Groucutt, 2020; Hallinan et al., 2022a). Sites featuring NLCs have been documented in currently arid environments across Northeastern Africa (Hallinan & Marks, 2023; Van Peer, 1998), the Levant (Hussain et al., 2015; Goder-Goldberger et al., 2016; Oron et al., 2024) and parts of Arabia (Crassard, 2008; Crassard & Hilbert, 2013; Rose et al., 2011), as well as from Southern Africa (Hallinan & Shaw, 2015, 2020; Will et al., 2015). However, the depositional context of most of these sites is as surface finds, which consequently lack chronological control. Based on the limited absolute ages available, NLCs are broadly associated with MIS 5 (Oron et al., 2024; Rose et al., 2011; Smith et al., 2007). While NLCs have been used as a proxy for understanding modern human dispersal and interactions within and outside of Africa at this time (Goder-Goldberger et al., 2016; Rose et al., 2011; Van Peer, 1998), it is important to note that this scenario cannot be universally generalised, with much later independent invention proposed in Southern Africa (Hallinan & Shaw, 2020; Will et al., 2015). Thus far, the only securely established association between Nubian technology and human remains is at the Nile Valley site of Taramsa, where anatomically modern human remains date to around 60 ka (Van Peer et al., 2010; Vermeersch et al., 1998). However, its exclusive association with Homo sapiens has been challenged and refuted (Blinkhorn et al., 2021a; Hallinan et al., 2022b).

Most broad discussion of the occurrence of NLCs in certain assemblages has been framed in terms of a widespread shared ‘Nubian Complex’, spanning Northern Africa during MIS 5, and potentially even earlier and later (Van Peer, 1998, 2016; Van Peer & Vermeersch, 2000). However, in spite of recent attention, there are currently no studies that quantitatively examine inter-regional patterning in the distribution of NLCs. Instead, the focus has largely been on lithic assemblages at a local-scale, confirming the attribution of NLCs based on their distinctive technological and morphological attributes (e.g. Hallinan & Shaw, 2020; Usik et al., 2013), metric characteristics (e.g. Groucutt & Rose, 2023; Hilbert et al., 2016), and chaîne opératoire (e.g. Chiotti et al., 2009; Oron et al., 2024).

Behaviour, Environment and Lithic Variability

Lithic artefacts are observed in their discarded state; therefore, they can provide certain insights into human technological and economic behaviour (Dibble et al., 2017). On the most fundamental level, core size can indicate both the minimum dimensions required by knappers to produce implements to meet their functional demands, and (in the context of a whole assemblage) the intensity of raw material reduction from an initial nodule to its discarded state (Douglass et al., 2018; Lombao et al., 2023). Core shape, in terms of length-width-thickness proportions, will also be affected by a combination of technological choices, functional needs and economic constraints (Brantingham & Kuhn, 2001; Groucutt & Rose, 2023; Rezek et al., 2011; Van Peer, 1992).

Both human behaviour—from population-level cultural patterns to the individual knapper’s decisions—and environmental factors play important, often interconnected, roles as drivers of lithic variability. To begin to understand how these factors are expressed in lithic assemblages, it is essential to situate them in their spatial context, in terms of resource availability (i.e. water, food and stone), environmental opportunities and climatic stress. For example, raw material type and availability will affect strategies related to stone acquisition, lithic maintenance and eventual discard (e.g. Bamforth, 1986; Henry, 1989; Hilbert et al., 2016; Marks et al., 1991). Landscape physiography, in terms of altitude and topography, has a major influence in structuring the distribution of food and water resources, with a resultant impact on hunter-gatherer group mobility (Binford, 1980, 2001; Henry, 1994). Water sources are especially important in governing the distribution of human settlement (Hallinan & Parkington, 2017; Kelly, 1983; Yellen, 1977) and technological strategy (Maloney, 2021). Lastly, the climatic context of temperature and precipitation, as well as the seasonal and longer-term stability of these conditions, is crucial for examining patterning at broader scales (Binford, 2001). Specifically, research has highlighted how dramatic shifts in climate can trigger cultural change and innovation (Bicho & Cascalheira, 2018; Ziegler et al., 2013) or open biogeographic opportunities for demographic dispersals (Carotenuto et al., 2016; Drake et al., 2011; Foley, 2018).

Numerous studies have considered the impact of these environmental factors on aspects of lithic variability at various spatial and temporal scales (e.g. Blinkhorn & Grove, 2021; Blinkhorn et al., 2021b; Mackay et al., 2014; Scerri et al., 2014; Wilkins et al., 2017). However, this approach has not yet been systematically and quantitatively applied to questions surrounding the distribution and variability of Nubian Levallois technology at an inter-regional scale. While the prevalence of surface sites without chronological control presents challenges for how Nubian technology is framed temporally in regional human dispersal narratives, the spatial context of NLCs, in conjunction with metric attributes, can provide new insights into human adaptation associated with this technology.

This study undertakes a multi-regional comparison of NLCs spanning Africa and southwest Asia. Using both published and new data, we investigate Nubian technology in terms of NLC metric variability and spatial distribution, addressing the following questions: What are the spatial and environmental characteristics of sites with NLCs? Are there regional or environmental trends, and how does this compare with contemporaneous sites without NLCs? To what extent do NLCs exhibit metric variability within and between regions? Did certain environmental conditions influence core dimensions? Lastly, what implications do these patterns have for understanding the role of NLCs in modern human evolution?

Study Region: Physiography, Environment and Climate

Our study focused on five regions where NLCs have been identified at multiple sites. The Saharo-Arabian belt may be considered a single biogeographic zone (Holt et al., 2013), but it is comprised of distinct regional physiographies and environments that responded differently to climatic fluctuations through time. During MIS 5, periodic wet episodes are argued to have made the arid zone more habitable, opening up opportunities for dispersal across the Sahara, as well as expansion through the Levantine corridor into the Arabian Peninsula (Abbas et al., 2023; Groucutt et al., 2015). An alternative model has proposed a southern expansion route out of Africa from the Horn of Africa, across the Red Sea (Armitage et al., 2011; Erlandson & Braje, 2015; Mellars et al., 2013). During glacial periods of lowered sea-levels, the exposure of the coastal shelf would have reduced the distance across the Bab al’Mandeb strait to as little as 4 km (Lambeck et al., 2011; Rohling et al., 2013). While the distribution of NLCs across these adjacent regions has been variably cited in support of both expansion routes (Rose & Marks, 2014; Rose et al., 2011; Vermeersch, 2001), testing dispersal scenarios is beyond the scope of our present study.

Northern Africa

Northern Africa is of central importance not only to the research history of NLCs, but also in the development of early modern humans and expansions within and out of Africa (Bergmann et al., 2022; Beyin et al., 2019; Garcea, 2012; Nicholson et al., 2022). Spanning more than 5000 km from the Atlantic Ocean in the west to the Red Sea in the east, the region experiences major climatic contrasts between its northern Mediterranean coastline and the arid Sahara. The corresponding north–south rainfall gradient today sees higher winter rainfall of up to 1200 mm in this coastal zone, contrasting with less than 120 mm in the desert zone (Karger et al., 2017). In the northwest, the Sahara and Mediterranean are separated by the Atlas mountains, which reach elevations between 500 and 4000 m. Additional inland mountains such as the Libyan Messak and Tadrart Acacus break up the numerous sand seas. The Nile Valley is an important physiographic feature, regarded not only as a corridor linking Eastern with Northern Africa (Beyin et al., 2019; Van Peer, 1998), but also a key refugium during periods of climatic deterioration (Leplongeon et al., 2020, 2021).

‘Green Sahara’ models have proposed that a network of river channels across the desert to the coast provided alternative, perhaps more favourable, dispersal routes for humans during MIS 5e (Drake et al., 2011; Larrasoaña et al., 2013; Osborne et al., 2008). Compared with the broadly arid conditions of MIS 6, the migration of the west African monsoon during MIS 5 caused periods of increased humidity, which persisted until mid-MIS 4 (Drake & Breeze, 2016; Drake et al., 2013). Increased moisture availability is evidenced by speleothem growth (Henselowsky et al., 2023), accelerated groundwater movements (Osmond & Dabous, 2004), and lacustrine carbonate formation (Armitage et al., 2007), linked to the formation of numerous Saharan megalakes (Drake et al., 2011). Vegetation reconstructions have proposed the expansion of wooded grassland across areas that are today desert (Larrasoaña et al., 2013), while faunal remains from Aterian (MIS 5–4) sites indicate a range of Savanna-adapted species (Drake & Breeze, 2016).

Southern Levant

The southern Levant is located at a critical intersection between Africa and southwest Asia, forming a key biogeographic corridor for species expansions out of Africa (Abbas et al., 2023; Carotenuto et al., 2016; Hershkovitz et al., 2018; Mirazón Lahr & Foley, 1994; Tchernov, 1992). The Mediterranean woodland zone in the north receives more than 400 mm of rainfall today, spanning from the coastal plain inland to the upland Judean Mountains, including the Carmel mountain range, with elevations of 400–1000 m asl. The intermediate Irano-Turanian zone is a xeric steppe environment consisting of grass and shrubland, receiving 150–350 mm rainfall per annum (Soto-Berelov et al., 2012). The arid zone in the south forms the northern extent of the broader Saharo-Arabian desert belt that today receives less than 200 mm of rainfall—as little as 20 mm per annum in the southern Negev and southern Jordan (Karger et al., 2017). The Negev Highlands and Jordanian Plateau reach elevations of 850 m and 1600 asl respectively. The region is climatically under the influence of the Mediterranean from the north and Sahro-Arabian desert from the south.

Late Pleistocene climatic records identify a major shift in the interglacial stage MIS 5 from the cool and arid conditions of MIS 6. Hydrological systems were active in the present-day desert regions (Avni et al., 2017, 2021), with peak rainfall in MIS 5e indicated by leaf wax biomarkers in the Dead Sea area (Tierney et al., 2022). The interglacial peaks MIS 5e, 5c and 5a were punctuated by cooler conditions during MIS 5d and b (Chen & Litt, 2018; Mischke et al., 2021). However, active palaeolakes and wetlands in the Azraq Basin, Wadi Gharangal and Mudawwara would have provided important refugia during periods of broader regional aridity (Ahmad & Davies, 2021; Al-Saqarat et al., 2021; Cordova et al., 2013; Petit-Maire et al., 2010).

Arabian Peninsula

The Arabian Peninsula is fringed by the Red Sea and the Gulf of Aqaba to the west, the Gulf of Aden and the Arabian Sea to the south, and the Gulf of Arabia to the east. A vast interior escarpment region extends from northern Saudi Arabia to Dhofar in southern Oman, bordered by the ‘Asir-Hijaz and Yemeni highlands (up to 3000 m asl) to the southwest and Rub al’Khali (Empty Quarter) desert to the northeast. It is generally defined by arid to hyper-arid environments with present-day average annual precipitation of 89 mm per annum (Karger et al., 2017). However, southern Oman is uniquely impacted by the Indian Ocean monsoonal system, with summer rainfall creating seasonally lush tropical environments and feeding interior-draining river systems that recharge artesian springs (Hildebrandt & Eltahir, 2006). The periodic migration of the Inter Tropical Convergence Zone driving increased rainfall northwards suggests these conditions may have expanded inland at certain times.

There are no perennial watercourses in the Arabian Peninsula today, but a network of incised wadi systems indicates past fluvial activity (Breeze et al., 2015). Palaeolakes such as Jubbah in the north and Mundafan in the south attest to ‘green’ periods throughout MIS 5 (Crassard et al., 2013; Petraglia et al., 2012; Rosenberg et al., 2011, 2013) offering windows for human dispersals (Drake et al., 2013). Speleothem growth during MIS 5, particularly MIS 5e, also documents wetter climatic conditions across the peninsula (Fleitmann et al., 2011).

Eastern Africa

Eastern Africa played a central geographic role in human evolution and dispersal to southwest Asia and beyond, facilitated by two routes: via the Nile Valley to the Levant, or through the Bab al’ Mandeb strait to the Arabia (Beyin, 2006; Clark, 1988). The Horn of Africa, bounded by the Red Sea and Indian Ocean, experiences a comparable climate to that of the southernmost part of the Arabian Peninsula due to the effect of the African and Indian Ocean monsoonal systems. Today the region is characterised by a mosaic of tropical and temperate vegetation zones, ranging from forest and woodland in the highlands, which receive as much as 2500 mm rainfall, to savannah and grassland lowland plains (Kalisa et al., 2019).

Eastern African lake records document periods of humidity in early MIS 5 (Grove et al., 2015; Schaebitz et al., 2021) but see a shift to a drier environment from MIS 5c until mid-MIS 3 (Beverly et al., 2017). Based on climatic modelling at a regional scale, it has been proposed that MSA occupation in MIS 5 and 3 was both more common and more widespread than in glacial periods, spanning a wide diversity of environments (Timbrell et al., 2022). During MIS 5, warmer and drier locations at lower altitudes were generally preferred, often making use of ecotonal habitats between shrubland and woodland. In contrast, MIS 3 MSA sites occur in much more varied contexts, seeing expansions into previously uninhabited environments (Timbrell et al., 2022). This difference is argued to reflect human habitat choice rather than simply the available environmental niches.

Southern Africa

Southern Africa has a rich MSA archaeological record that has received particular attention for evidence of complex ‘modern human’ behaviours from around 100 ka at numerous, mostly coastal, cave sites (Wadley, 2015). The region has recently entered the ‘Nubian debate’ with NLCs identified at several sites in the interior Tankwa Karoo area that are suggested to date to MIS 3 (Hallinan & Shaw, 2015, 2020; Will et al., 2015). Southern Africa is currently characterised by nine terrestrial biomes (Rutherford et al., 2006), reflecting diverse geological and topographic landscapes structured by the Cape Fold Belt following the western and southern coasts, and the interior Great Escarpment plateau, with its eastern edge formed by the Drakensberg mountains. The Drakensberg and Lesotho Highlands reach elevations of 2000–3500 m with rainfall of over 1000 mm. This contrasts with the arid interior where rainfall ranges from 300 to less than 100 mm. The climate is influenced by Atlantic and Indian Oceanic systems which cause highly seasonal rainfall that would have fluctuated in strength and distribution throughout the Quaternary (Chase & Meadows, 2007).

MIS 3 is no longer regarded as a period of widespread aridity (Mitchell, 2008), instead witnessing variable and fluctuating conditions across the different biomes. While the coastal montane regions may have seen reduced occupation of caves under drier conditions, pollen evidence from the interior indicates relatively cooler and wetter conditions (Brook et al., 2010; Scott et al., 2022). New indications of palaeolakes in the central interior during MIS 3 suggest a resource-rich landscape that may have drawn human settlement away from the coastlines in this period (Carr et al., 2023).

Datasets and Methods

Archaeological Data

In a thorough review of the literature, Hallinan (2024) compiled a database identifying 154 published sites where NLCs are described or have been retroactively identified. Of these sites, 31 are queried or rejected as not fully demonstrating the established technological and morphological criteria for NLCs (Hallinan et al., 2022a; Usik et al., 2013). For our analysis, a further level of screening removed sites that had insufficient spatial, chronological or technological information, resulting in a sample of 84 sites with NLCs present (Fig. 1, Supplementary Information 1). While very few sites with NLCs have been radiometrically dated (Table 1), they have been broadly associated with MIS 5 in Northern Africa, Arabia and the Levant, MIS 3 in Southern Africa, and both MIS 5 and MIS 3 in Eastern Africa.

Fig. 1
figure 1

Map showing the distribution of sites with Nubian cores included in the study

Full size image
Table 1 Chronometric ages for sites with Nubian Levallois cores (accepted as present in Hallinan, 2024 and this paper). Sites in italic have relevant ages but are not included in this study
Full size table

A sub-sample of 14 assemblages with NLCs for which full metric data were available was subjected to further analysis to address regional and environmental variability at a finer behavioural scale (Table 2). These comprised measurements of core length, width and thickness, from which indices of elongation (length/width, ICE) and flattening (width/thickness, ICF) were calculated. Measurements were derived from published sources (K’one, Nahal Paran 9, Wadi Sabra, Wadi Sana, Wadi Wa’Shah), our own data (Wadi D/E, Wadi Bairiya, Tweefontein), or supplied directly by relevant authors (Dimona, H2, TH.69, TH.377, TH.383, Uitspankraal 7). Inter-observer variability of artefact measurements has been tested for unretouched flakes (Pargeter et al., 2023), with length, width and thickness measures showing strong inter-analyst agreement. Equivalent tests have not yet been carried out on cores; however, given the distinctive morphology of NLCs that aids their orientation on a consistent axis, we find reasonable grounds to assume that inter-observer variation on these metrics will not introduce substantial bias to our results.

Table 2 Sites forming the Nubian core metric sample
Full size table

Comparative MSA/MP sites with Levallois technology but without NLCs were identified for the appropriate time periods in these regions, drawing on existing syntheses (e.g. Blinkhorn & Grove, 2021; Dibble et al., 2013; Mitchell, 2008; Prévost & Zaidner, 2020) supplemented by a review of the primary published literature (Fig. 2, Supplementary Information 1). In general, only sites with chronometric information were included, but for the Nile Valley where Nubian technology plays a central role, more comprehensive survey data provided important observations of sites with an absence of NLCs (Marks, 1968a, 1968b; Vermeersch et al., 2000). This generated a sample of 81 sites attributed to MIS 5 or 3 where NLCs are absent, or where their claimed presence has not been securely demonstrated (e.g. Bir Tarfawi (Van Peer & Vermeersch, 2000), Skhul (Groucutt et al., 2019), ‘Ain Difla (Demidenko & Usik, 1993; Groucutt, 2020), Umm al Sha’al (Crassard et al., 2019), Jebel Katefeh (Groucutt et al., 2015), Aduma (Yellen et al., 2005) and Mochena Borego (Brandt et al., 2017); see discussion in Hallinan (2024)). A number of sites from Sudan were not included in the analysis due to concerns related both to the identification of NLCs and associated very old or very young ages, some of which may be problematic (Masojć, 2018; Masojć et al., 2017; Osypińska & Osypiński, 2015; Osypiński & Osypińska, 2016; Osypiński et al., 2021; Van Peer et al., 2003).

Fig. 2
figure 2

Map showing the distribution of contemporaneous MSA and MP sites without Nubian cores included in the study

Full size image

While in some regions (Northern Africa, Arabia) these selection criteria introduce a numeric bias towards sites with NLCs present, this nevertheless should not undermine the statistically robust identification of overall environmental patterns. However, for this reason, certain spatial tests (e.g. spatial autocorrelation, cluster analysis) were not deemed appropriate. Furthermore, regional patterning in technological and cultural variability of assemblages with NLCs will not be addressed here, but rather will be the subject of other papers in this volume.

Spatial and Environmental Data

Site locations were either obtained from co-ordinates given in the primary literature or from the ROCEEH ROAD database (Kandel et al., 2023) (Supplementary Information 2). Site altitude, topographic context and various bioclimatic variables were extracted from the modelled palaeoclimatic dataset for the last 120 ka (Beyer et al., 2020), available in the R package pastclim (Leonardi et al., 2023). Data were available at the spatial resolution of 0.5° grid cells (55 km2); these parameters have been applied to other regional analyses (e.g. Blinkhorn & Grove, 2018), with a 50-km hunter-gatherer home range observed in the ethnographic data (Binford, 2001). Since absolute chronological context is currently scarce for sites with NLCs (Table 1), broad time-slices for MIS 5 and MIS 3 were extracted for the appropriate regions to capture climatic variation throughout these periods. Bioclimatic data for each site were taken at 2000-year time-slices; for MIS 5 in 21 slices for the period 120–80 ka, and for MIS 3 in 16 slices for the period 60–40 ka. Mean annual temperature (bio01), mean annual precipitation (bio12) and net primary productivity (NPP; gC/m2 per year) provided numeric bioclimatic variables, with biome type recorded as a categorical variable. Net primary productivity—reflecting the accumulation of biomass—and biome data were derived from the Biome4 global vegetation model (Beyer et al., 2020; Kaplan et al., 2003). The bioclimatic values for each site were plotted to assess variability within each time period (Supplementary Information 4: Figs. S4.1–S4.12). Median values were considered appropriate for temperature, rainfall and NPP, and the modal biome was selected.

The topographic variables of elevation and rugosity provided additional information about landscape, derived from the ETOPO1 dataset (Amante & Eakins, 2009; Leonardi et al., 2023). Studies have found that areas of high topographic complexity are biodiversity hotspots (Badgley et al., 2017) that may have been particularly favourable habitats for past humans (Bailey et al., 2011; Holdaway et al., 2015; Winder et al., 2015). Terrain roughness has been used elsewhere as an indicator of the energy expenditure required to move across a landscape, with the underlying assumption that more rugged terrain has greater energetic demands (Blinkhorn & Grove, 2021; Timbrell et al., 2022); here, cost-path analysis is not carried out as it is not suited to the broad spatial scope of the study.

Statistical Analysis

Data visualisation and statistical analysis on both the metric and environmental data were carried out in R (R Core Team, 2024) using various packages (ggplot2: Wickham, 2016; dunn.test (Dinno, 2017); car: Fox & Weisenberg, 2019; tidyverse: Wickham et al., 2019; broom: Robinson et al., 2023). All site data and R script used for spatial and environmental analysis are provided (Supplementary Information 2; Hallinan & Samawi, 2024). Lithic metric data are provided as a summary rather than raw data since they were partly supplied by other authors (Supplementary Information 3).

Logistic regression models were used to test for the influence of different environmental factors on the presence or absence of NLCs at a site. In these models, the binary presence-absence response variable was regressed against the environmental predictor variables, and the resulting model was used to determine the associated probabilities for each variable’s impact on the presence of NLCs at a site. The models were tested for multicollinearity using a variance inflation factor, and correlation matrices were used to explore covariance relationships among the environmental variables. This was performed separately for sites dating to MIS 5 (Northern Africa, the Levant, Arabia and Eastern Africa) and MIS 3 (Eastern Africa and Southern Africa) to remove temporal bias. It is also important to note that the frequency of sites with NLCs is much lower in Eastern and Southern Africa than across the Saharan-Arabian belt. To investigate regional patterns in more detail, pairwise t-tests compared the presence of NLCs against each environmental variable separately to minimise the influence of covariance between factors.

For the sub-sample of 14 sites for which metric data were available, core length, elongation and flatness indices were compared. For length, width and thickness, the coefficient of variation (CV) was used as an indicator of within-assemblage variability, calculated by dividing the standard deviation by the mean and expressed as a percentage (Eerkens & Bettinger, 2001). Elsewhere, CV values under 20% are interpreted as low variance (i.e. very high standardisation), and 40–50% shows high variance that likely indicates random variation (Groucutt & Rose, 2023). CV was only assessed for assemblages with more than ten NLCs, and only using single-dimensional rather than relative (ICE/ICF) variables (see discussion in Muller & Clarkson, 2023).

Significance in the metric data was tested using the non-parametric Kruskal–Wallis test on group medians. This tested the null hypothesis that there is no difference between median NLC sizes between samples from Northern Africa, Arabia and the Levant (the regions with the most samples, dating to MIS 5). Where the alternative hypothesis was supported, this was followed by post-hoc Dunn’s tests using pairwise comparisons of rank sums to identify which samples differed, using the Bonferroni method to limit the likelihood of Type 1 errors (false positives) (Hochberg, 1988).

Linear regression models were used to test for a relationship between environmental variables for each of these sites and core length, elongation and flattening. Correlation matrices were then used to explore pairwise associations among the environmental variables and core metrics to further assess these relationships. A significance threshold of 0.05 was used for all statistical tests.

Results

Environmental Characteristics of Sites with Nubian Cores

The environmental patterns for sites with NLCs were compared with other MSA/MP sites in each region for the relevant time periods (Fig. 3). Considering first the landscape setting of sites with NLCs, the lowest elevation sites were in the Nile Valley, with 26 sites below 250 m. The highest elevation sites (over 1000 m) were in Eastern Africa (Gademotta, Midhishi 2, K’one, Porc Epic), on the Southern African interior plateau (Orangia 1), and at the eastern end of the Northern African Atlas Mountains (Oued Djouf el Djemel) (Fig. 3A). Thirty-one sites were at elevations between 500 and 1000 m, with clusters located in Southern and Central Arabia, the Negev highlands, and the central Sahara. The remaining 21 sites were at mid-elevations (250–500 m), found in all regions. Rugosity does not correspond directly with elevation, with a contrast seen between rugged mountain-zone sites and those on elevated plateaux. The highest rugosity areas (200–460) were the Negev, Ethiopian highlands and the Tankwa Karoo basin in Southern Africa (Fig. 3B). The lowest rugosity (< 50) was in the Nile Valley at Wadi Halfa, Northern Arabia and the central Sahara. The sites with the highest elevations and highest rugosity also had the highest biomass, with NPP over 1000 at Gademotta and Rusinga Island (Fig. 3C). This was followed by 250–300 at Mediterranean-zone sites in Northern Africa, and Dhofar in Southern Arabia, lying at higher elevations but with low rugosity. The majority of sites had NPP of less than 100 (n = 61), in the Nile Valley, Sahara, Negev and Northern Arabia.

Fig. 3
figure 3

Environmental context of sites where Nubian cores are present and absent, grouped by region: A boxplot of site elevation; B boxplot of site topographic roughness; C boxplot of site net primary productivity; D histogram of vegetation biomes (see Fig. 4 for biome type names); E boxplot of site mean annual temperature; F boxplot of site mean annual precipitation

Full size image

In terms of vegetation biomes, 87% of sites were in areas classified as desert (n = 71) or barren land (n = 2), characterised by low annual rainfall, where plants generally consist of woody shrubs, grasses and succulents (Figs. 3D and 4). Of these, 61 sites were in areas with less than 100 mm rainfall, spread across Northern Africa, the Negev, and Northern and Central Arabia (Figs. 3E and 5). Sites on the Nejd plateau in Dhofar in southern Oman received between 170 and 220 mm, and the Tankwa Karoo site of Tweefontein, in Southern Africa, received 180 mm. Non-desert biomes occurred in areas with higher rainfall, with Eastern African tropical and temperate sites receiving 710–1590 mm. One of these sites, Rusinga Island in Lake Victoria, was in tropical forest—among the most productive biomes on Earth, with Gademotta situated in temperate coniferous forest, and K’one and Porc Epic in tropical xeric shrubland. In the 250–500-mm rainfall bracket, two Northern African Mediterranean sites (Bir Chaacha, Oued Djouf el Djemel) and Dimona South were in xeric shrubland, and Midhishi 2 in the Horn of Africa was in open coniferous woodland. Sites in Southern Africa’s interior are also located in xeric shrubland contexts; those near the Cederberg Mountains (Uitspankraal, Tankwa Karoo) receive elevated rainfall, while Orangia 1 is in a low rainfall area (160 mm).

Fig. 4
figure 4

Map of modelled vegetation biomes with sites where Nubian cores are present and absent for MIS 5 (Northern Africa, Levant, Arabia, Eastern Africa) and MIS 3 (Southern Africa, Eastern Africa). Biome data shown at time-slices of 120 ka (MIS 5) and 50 ka (MIS 3)

Full size image
Fig. 5
figure 5

Map of modelled mean annual precipitation with sites where Nubian cores are present and absent for MIS 5 (Northern Africa, Levant, Arabia, Eastern Africa) and MIS 3 (Southern Africa, Eastern Africa). Precipitation data shown at time-slices of 120 ka (MIS 5) and 50 ka (MIS 3)

Full size image

In the Köppen-Geiger climate classification system, 18 °C is regarded as the mean annual temperature threshold between hot and cold deserts. The Southern African MIS 3 sites all fall within the Karoo region, between 16 and 14 °C, and most of the Levantine Negev region sites range between 17 and 18 °C. The majority of desert sites are in the 19–25 °C range. The hottest desert sites, reaching temperatures over 24 °C, are in the Sahara, at Arouakim in the west, and Wadi Halfa area of the Nile in the east (Fig. 3F). Other non-desert sites experienced a range of temperatures, with cooler climates in the Mediterranean montane sites of Oued Djouf el Djemel and Bir Chaacha, and the humid Eastern African sites ranging between 17 and 22 °C.

Influence of Environmental Variables on the Presence of Nubian Cores

While broad trends show that the majority of sites with NLCs occur in hot, arid desert contexts, there are some notable exceptions in tropical environments of Eastern Africa and the Karoo ‘cold’ desert in Southern Africa. Sites in these regions also mostly date to MIS 3, rather than MIS 5 as in the Saharo-Arabian belt. Logistic regression models were used to determine the influence of environmental variables on NLC presence for each period. The model for MIS 5 sites suggested that rainfall (p-value, 0.000), rugosity (p-value, 0.013) and, to a lesser extent, NPP (p-value, 0.006) were significant predictors of the presence of NLCs at a site (Table 3). Additional tests for covariance between the environmental factors indicated a strong correlation between rainfall and NPP (Supplementary Information 4), therefore the regression was repeated with each variable removed. The models returned had lower residual deviance and higher AIC values, indicating the first model for all factors was the best fit of the data (Null deviance, 178.22; Residual deviance, 129.81; AIC, 141.81). Positive coefficients for rugosity and NPP suggest that NLCs are more likely to occur in areas that are topographically complex, and have relatively high biomass. The inverse is indicated for rainfall, suggesting that NLCs are more likely to occur in areas with low rainfall.

Table 3 Logistic regression coefficients for Nubian core presence at MIS 5 sites. P-value < 0.05 shown in bold
Full size table

The regression model for MIS 3 sites suggested that only temperature was a significant predictor of the presence of NLCs (p-value, 0.035), with a positive coefficient indicating they are more likely to occur at warmer temperatures (Table 4). The model is a good fit (Null deviance, 36.56; Residual deviance, 26.35; AIC, 38.35), with only minor changes of model fit when rainfall or NPP were excluded (Supplementary Information 4). However, overall the MIS 3 sites included only four sites each in Eastern and Southern Africa where NLCs were present (7 and 18 sites without, respectively), which may make the model less reliable than the larger sample sizes in MIS 5.

Table 4 Logistic regression coefficients for Nubian core presence at MIS 3 sites. P-value < 0.05 shown in bold
Full size table

Specific environmental relationships were also assessed for each region separately. In Northern Africa and the Levant, rainfall and NPP show a very strong correlation with the presence of NLCs (Table 5). These environmental factors reflect a distinction between coastal, Mediterranean biomes and the interior desert, where there is a clear geographic pattern that NLCs do not occur at Mediterranean sites. Additionally, temperature is significant in Northern Africa for the same geographic reasons, with cooler temperatures experienced near the Mediterranean coasts. In Arabia, NPP and rugosity show significant relationships, reflecting the rugged and more ecologically productive Dhofar and Hadramawt regions in the southwest where NLCs occur, in contrast with the flatter desert regions where they do not. Eastern Africa and Southern Africa show no correlations between environmental factors and the presence of NLCs, potentially reflecting the very complex landscapes and diverse environments in each region.

Table 5 Summary of pairwise (t-tests) relationships between Nubian cores and environmental variables, by region. P-value < 0.05 shown in bold
Full size table

Nubian Core Metric Variability

For a sub-sample of 14 sites, 513 NLCs were available for metric analysis, showing distinct patterns between the regions (Fig. 6). Overall NLC length ranged between 22 mm (K’one) and 173 mm (TH.383), with a mean value of 77 mm (Table 6, Supporting Information 3). Relatively large cores occur in the Levant (mean, 98 mm), Arabia (mean, 89 mm), and Northern Africa (mean, 75 mm). In contrast, small NLCs are found in Eastern Africa (mean, 40 mm) and Southern Africa (mean, 48 mm).

Fig. 6
figure 6

Box plots of Nubian core metrics (length, elongation and flattening) by site

Full size image
Table 6 Summary of Nubian core metric data. Length, width and thickness values rounded to the nearest mm. CV not calculated for samples smaller than 10. Summary statistics provided in Supplementary Information 3
Full size table

The CV gives an indication of the degree of within-assemblage metric variability, often framed in terms of standardisation (Table 6). The sites showing the highest standardisation in NLC length (i.e. the lowest CV) are in Arabia, with values under 20, and as low as 15 at the Dhofar site TH.377. In relative terms, higher variability is seen in Northern, Eastern and Southern Africa with values over 20. However, even the highest value of 30 at K’one still shows moderately high standardisation. Sample sizes at most Levant sites were too small to show meaningful variance, although the value of 20 at Dimona South shows similar standardisation to the Nile Valley sites. CV for width and thickness were more varied: for width, the lowest CV values were observed at Dimona South, TH.69 and Wadi Bairiya. K’one again shows the highest value, over 30, and other sites are between 20 and 25. For thickness, standardisation is relatively low with all values over 20, reaching a maximum of 48 at the Dhofar site TH.377. It is important to note that the CV measure for cores only provides information on core size at discard and not the end-product that was the target of Levallois production. Few other studies have applied CV to cores; however, Dhofar NLCs have been described as highly standardised elsewhere (Groucutt & Rose, 2023; Usik et al., 2013).

Nubian Levallois knappers across the regions discarded cores with varied shapes, assessed here in terms of elongation (ICE) and flattening (ICF) indices (Fig. 7). In terms of elongation, NLCs from Dhofar in Southern Arabia are the most elongated, from sites TH.377, TH.383, and TH.69. Less elongated, proportionally wider, cores occur in almost all other regions, with a mean ICE value of 1.2 observed at the newly reported site Wadi D/E (Northern Africa), Nahal Paran 9 (Levant), and Uitspankraal 7 (Southern Africa). In terms of flatness, the Dhofar site TH.69 shows the lowest flattening value (i.e. cores are narrow and thick), while higher values, indicating flatter cores, occur at K’one, followed by Wadi D/E and Nahal Paran 9.

Fig. 7
figure 7

Bar plots of Nubian core metrics (length, elongation and flattening) for wider regional Northern African (red), Arabian (orange), and Eastern African (blue) samples. Sites in the primary sample used in earlier analyses are outlined in black

Full size image

Kruskal–Wallis and post hoc Dunn’s tests were applied to assess whether there were regional similarities in core shapes between Northern Africa, Arabia and the Levant (MIS 5 assemblages) (Table 7). The null hypothesis of no difference was rejected, indicating significant differences for ICE (χ2, 26.20; df, 2; p-value, 0.001) and ICF (χ2, 54.43; df, 2; p-value, 0.001). For ICE, cores from Arabia are significantly more elongated than assemblages in Northern Africa (p-value, 0) or the Levant (p-value, 0.002), but cores from the Levant and Northern Africa showed no significant difference from each other (p-value, 1). For ICF, cores from Northern Africa were significantly flatter than those in Arabia and the Levant (p-values, 0), whereas the Levant and Arabia showed no regional differences.

Table 7 Dunn’s test results for comparisons between MIS 5 assemblages. P-value < 0.05 shown in bold
Full size table

Metric Variability in Environmental Context

When the core metric results are considered in environmental context, the linear regression model indicates that for mean core length, environmental factors explain 57% of the metric variance, but the overall model is weak, given the F-statistic of 2.10 (Table 8). However, within the variables, rainfall is significant at the 0.05 level (p-value, 0.037), and NPP is marginally significant at the 0.1 level (p-value, 0.066), suggesting a relationship with core size (Table 9). The negative coefficient for size and rainfall indicates that core size is expected to increase as rainfall decreases; the smallest cores are from K’one which receives the highest amount of rainfall (799 mm per annum), and also has the highest NPP (425). However, the largest cores are from Dhofar which receives only moderately low rainfall (221 mm), indicating this is not a simple relationship.

Table 8 Linear regression model fit for Nubian core metrics
Full size table
Table 9 Linear regression coefficients for Nubian core length. P-value < 0.05 shown in bold, P-value < 0.1 in italics
Full size table

For both elongation and flattening, the low F-statistics of 1.13 and 1.01 indicate there is no clear evidence of a relationship between these metrics and the environmental variables in these models, and no individual variables show significant coefficients (Table 8, Supplementary Information 4). Elongation and flattening indices are a simple indicator of core shape, but ongoing analysis using 3D geometric morphometrics is investigating variability in NLC shape at an inter-regional scale. The absence of a relationship between core shape and environmental factors presents the alternative hypothesis that shape may be influenced by behavioural factors (e.g. raw material, reduction method), that will be tested in future work.

Correlation matrices allowed pairwise comparisons of metrics and the interaction between each environmental variable (Table 10). Core length shows a weak positive correlation with temperature (0.23) and rugosity (0.19). Conversely, length shows a negative correlation with rainfall (− 0.45), NPP (− 0.01) and elevation (− 0.22). The statistical significance of the correlation between length and rainfall has already been observed in the linear model, suggesting that smaller cores occur in areas of higher rainfall. K’one is located in a tropical xeric shrubland biome with high NPP, and is the only site in Eastern Africa for which metric data are available. Most other sites occur in desert environments, with the exception of Dimona South and Uitspankraal 7 which fall in xeric shrubland with NPP of less than 150. Strong positive correlations are also indicated between rainfall and elevation (0.76), and NPP (0.72), suggesting that sites, such as K’one, at higher elevations experience wetter and more ecologically productive conditions than sites at drier, lower elevations. However, the situation of most Arabian and Negev sites on elevated desert plateaux at 500–700 m, which yield the largest cores, again indicates this is not a simple relationship.

Table 10 Correlation matrix summary for Nubian core metrics and environmental variables
Full size table

While it has already been demonstrated that core elongation and flattening do not show a strong relationship with any environmental variables, correlation matrices indicate the following patterns. A weak positive correlation suggests that more elongated cores may occur in areas with higher temperatures (0.28) and in areas with higher ecological productivity (0.36). The inverse is suggested for terrain roughness (− 0.24) suggesting that elongated cores occur in landscapes that are less topographically variable. Core flattening shows a moderate positive association between flatter cores and wetter environments (0.37), and higher elevations (0.27), but this pattern may again be driven by K’one which has the flattest cores.

Discussion

Environmental Characteristics of MIS 5 Sites with Nubian Cores

Overall, our results show that for MIS 5 sites, rainfall, rugosity and NPP are strong predictors of the presence of NLCs. In Northern Africa and the Levant, NLCs occurred at sites where rainfall and NPP were low, reflecting an association with the interior desert zone, in contrast to the coastal, more temperate Mediterranean biome sites which did not have NLCs. While the global climate models used for our analysis indicate low MIS 5 rainfall (Beyer et al., 2020), in some regions, increased precipitation and hydrological activity have been modelled, suggesting a more complex picture of local water availability (Larrassoaña et al., 2013; Breeze et al., 2015; Henselowsky et al., 2022). Furthermore, absolute precipitation in a region is not necessarily the sole determinant of water in that area. For example, the River Nile in Egypt and Sudan is contingent on flow from its tributaries, the White Nile, Blue Nile and Atbara, which in turn depend on tropical Eastern African rainfall. Evidence for Nile flood events during MIS 5 is therefore indicative of enhanced precipitation to the south, rather than in the Nile Valley itself (Williams et al., 2015).

Sites with NLCs are particularly common in the Nile Valley which is the historical focus of the study of this technology (e.g. Guichard & Guichard, 1965; Hallinan & Marks, 2023; Van Peer, 1998). While study bias may account for the density of reported sites in this area, the sporadic but persistent presence of NLCs across the Saharan region, and their conspicuous absence from well-dated rock shelters of the Northern African coastline, suggest some underlying behavioural and ecological patterns of Nubian technology. This distribution has been interpreted as a pan-Northern-African Nubian technocomplex that subsumes various regional industries (Van Peer, 1998; Van Peer & Vermeersch, 2000), but there is considerable variation in both the number of NLCs and their assemblage context across this vast area. Currently, there is insufficient chronological, spatial and assemblage-level resolution to evaluate this further.

The Mediterranean coast vs. interior desert distribution is mirrored in the Levant, with NLCs only identified in the Negev and Araba valleys. Elsewhere this has been characterised as representing two distinct technological packages: one featuring centripetal Levallois cores in the high-rainfall Mediterranean zone, and one where centripetal cores occur alongside NLCs in the desert (Barzilai et al., 2022; Goder-Goldberger et al., 2016; Richter et al., 2012). The MIS 5 central Negev site of Nahal Aqev is an exception, interpreted to represent a southern expansion of groups from the woodland ecozone, exploiting the environment supported by an active spring (Barzilai et al., 2022). The ecological boundary in the occurrence of NLCs in the Levant is further supported by intensive surveys along a 60 km stretch of the Jordan Valley, south of the Sea of Galilee, which recorded abundant Middle Palaeolithic activity yet found no NLCs (Jagher et al., 2023). This falls within the Irano-Turanian (scrubland) vegetation zone, with relatively high rainfall (250–350 mm) (Soto-Berelov et al., 2012), therefore the arid Negev zone pattern is so far upheld.

In Arabia, NLCs were associated with high rugosity and NPP, influenced by the Dhofar sites on the southern Nejd plateau. It is the only region where rainfall is not a major influencing factor, possibly due to limited variation across the generally arid Arabian Peninsula. However, water availability is expected to have played a major role, with NLCs occurring only in areas where palaeohydrological modelling indicates the presence of major drainages (Breeze et al., 2015). The Dhofar landscape features deeply incised wadi channels, with seasonal ground-water-fed springs that were likely to have been a key attraction to human groups (Rose et al., 2011). The Central and Northern Arabian areas where NLCs are also found have a similar complex landscape structure of fluvial channels, plateaux and inselbergs; however, NLCs occur here in much lower numbers than Dhofar, as well as alongside other Levallois core types (Crassard & Hilbert, 2013; Hilbert et al., 2017). Compared with Dhofar, temperatures were higher, and rainfall, rugosity and NPP were lower, suggesting that this environment demanded different adaptations. Arabian sites where NLCs do not occur are associated with palaeolakes (Groucutt et al., 2015; Petraglia et al., 2012), in landscapes with comparatively little topographic differentiation. Again, this presents a different resource structure demanding alternative behavioural strategies that may account for the absence of NLCs. This may be further supported by the absence of NLCs at dated MIS 5 palaeolake oases in the Egyptian desert, in contrast with their abundance in the Nile Valley and adjacent plateaux (Nicoll, 2018; Wendorf et al., 1993).

Environmental Characteristics of MIS 3 Sites with Nubian Cores

For MIS 3 sites, a positive correlation was observed with temperature, indicating that overall NLCs occurred in warmer climates. However, no specific associations with any environmental variables were observed in each region considered separately. MIS 3 sites in Eastern and Southern Africa cover a broader range of ecological diversity (nine biome types) than those occupied in MIS 5 Northern Africa, the Levant and Arabia (five biomes) which cover a spatially much larger area.

In Eastern Africa, very high rainfall is associated with the highlands, with four sites with NLCs located over 1000 m elevation (Gademotta, Midhishi, K’one and Porc Epic). Timbrell et al. (2022) observe that while precipitation seems to be a stronger determinant of Eastern African MSA occupation in MIS 5, in MIS 3 temperature is more significant. For the MIS 3 logistic regression model in our study (Table 4), we also determine that temperature (as opposed to precipitation in MIS 5) is significant in determining NLC presence, finding an association with warmer temperatures. However, this contrasts with the overall MIS 3 settlement preference for environments with cooler temperatures identified by Timbrell et al. (2022), which they argue reflects the use of mountain refugia.

In Southern Africa as well, MIS 3 occupation has been proposed elsewhere to be associated with mountain refugia (Mitchell, 2008). Unlike Eastern Africa, where elevation and rainfall show a generally strong positive correlation, there is a more varied pattern in topography and precipitation in Southern Africa. The lowest rainfall sites, both overall and of those with NLCs, are Tweefontein and Orangia 1 which received less than 180 mm, yet Orangia is at an elevation of 1200 m and Tweefontein at 400 m. Instead, the biome may serve as a clearer indicator in Southern Africa, whereby wooded and Mediterranean biomes do not have NLCs but desert and temperate xerophytic shrubland do. Broadly this corresponds to the Karoo region inland of the Cape Fold Mountain Belt, which today comprises succulent shrubland but at certain times, such as MIS 3, supported lakes and grassier vegetation than present (Carr et al., 2023). Research in this region is underdeveloped due to the paucity of rock shelter sites, but surface sites show a rich MSA record—including the only occurrences of NLCs in Southern Africa—that hint that the interior may have been an important counterpart to coastal settlement, under certain climatic conditions (Hallinan, 2021; Sampson, 1968, 1985).

Nubian Core Metric Variability

Strong patterns in core metrics were observed in our data between regions, with NLCs from Northern Africa, the Levant and Arabia discarded with larger dimensions than cores in Eastern and Southern Africa. To examine these patterns further, published metrics from additional sites in these regions were compared with our raw metric results. This included sites from the Nile Valley (Abydos and Nazlet Khater), Northern and Central Arabia (Al-Jawf (DAJ) and Al-Kharj (AK) regions), and Gademotta in Eastern Africa (Table 11; Fig. 7). Broadly, these additional samples followed the same regional trends (although note some small sample sizes): cores in Arabia and Northern Africa were larger (largest mean, 100 mm at AK-43), and Eastern Africa were smaller (mean, 42 mm at Gademotta).

Table 11 Metric data from published literature (mean values, in mm). Measurements for asterisked (*) sites were extracted using Web Plot Digitzer (version 4) from published plots
Full size table

Additional samples from the Nile Valley show similar values to our new sites from the low desert near Thebes (Wadi D/E and Wadi Bairiya). Abydos NLCs have an average length of around 70 mm and Nazlet Khater 60 mm, with the Thebes NLCs only slightly larger, around 70–80 mm (Table 11). Elongation and flattening values are not available for the Abydos sites, but all other Nile Valley cores have similar proportions, even though the Nazlet Khater cores are smaller overall. Nazlet Khater 1 is interpreted as a chert quarrying site exploiting in situ Nile gravels (Vermeersch et al., 2002), as part of a radiating (logistical, sensu Binford, 1980) mobility strategy (Van Peer, 2001). In contrast, the Abydos high desert sites are interpreted as showing a highly mobile circulating (residential, sensu Binford, 1980) strategy, with most sites not located directly on (although not far from) sources of chert (Olszewski et al., 2010). Therefore, at least some metric variation observed in Nile Valley NLCs may be explained in terms of locally available raw materials and different patterns of site use.

NLCs from additional samples in Northern (Al-Jawf) and Central (Al-Kharj) Arabia are smaller than the Dhofar assemblages TH.377 and TH.383 but are otherwise variable, with mean lengths between 56 and 100 mm (Table 11). Cores from Al-Jawf are generally the smallest (with the exception of DAJ-51, although with a small sample of 3), with an average length of 62 mm, comparably elongated, but flatter. Indeed, DAJ-51 and DAJ-110 samples are the flattest NLCs overall (ICF, 3.3 and 3.5, respectively) (Fig. 7). Small, short and flat cores at DAJ-110 (mean, 56 mm; ICE, 1.1; ICF, 3.5) may be accounted for by the absence of a primary outcrop close to the site, although chert nodules 10–30 cm in diameter are available in secondary context (Hilbert et al., 2017). In contrast, Al-Kharj NLCs show comparable shape, in terms of elongation (ICE, 1.3 to 1.5) and flattening (ICF, 1.9 to 2.5) with Dhofar and Hadramawt samples, and are among the thickest cores overall (Fig. 7). Hilbert et al. (2016) note that the reduction of Al-Kharj quartzite cores frequently ceased due to knapping errors such as siret and hinge fractures on the preferential removal, which may explain their relative thickness.

An interesting question relates to the age of the smaller cores observed in Eastern and Southern Africa, attributed to MIS 3 in both regions. In Dhofar, a “micro-Nubian” core variant has been noted in assemblages assigned to a separate industry, the Mudayyan (Usik et al., 2013). As well as diminutive Nubian cores (< 80 mm; mean, ~ 70 mm), these assemblages contain small bidirectional cores, and show a much lower degree of weathering (patination, chemical dissolution, rounding of arêtes), suggesting they are a younger—though currently undetermined—age than the “Classic Dhofar Nubian” (MIS 5) assemblages (Usik et al., 2013). This observation could suggest a wider pattern of miniaturisation of Nubian cores after MIS 5, though the meaning and significance of this are currently unclear.

Metric Variability in Environmental Context

A clear environmental trend in core metrics is that core size decreases as rainfall increases. This is largely driven by K’one with the smallest cores and highest rainfall (> 770 mm), which contrasts with the largest cores at Dhofar sites that have relatively low (but not the lowest) rainfall of < 210 mm. These two extremes are also expressed in core shape: elongated cores may occur in areas with higher temperatures and higher ecological productivity, with lower roughness (i.e. Dhofar cores are the most elongate as well as the largest); and flatter cores may occur in areas with wetter environments and higher elevation (i.e. K’one).

As the only assemblage with primary metric data in Eastern Africa, K’one could be an outlier, given its extreme values for both rainfall (all otassemblage with primary metric data inher sites receive less than 350 mm) and size, although the two Southern African sites are similarly small. To test this, additional data from four NLCs from Gademotta were considered (Table 11), showing a similar size (mean, 42 mm at Gademotta; 40 mm at K’one), but higheassemblage with primary metric data in

r ICE (1.4 at Gademotta; 1.3 at K’one) (ICF could not be calculated). Gademotta is a highland site that experienced MIS 5 precipitation of over 1100 mm, therefore follows the same trend as K’one for small core size and high rainfall. Besides K’one and Gademotta, all other Eastern African sites have (where stated) one reported obsidian NLC, with limited accompanying technological information or illustration. Given the mixed chronological affinities, relating to both MIS 5 (only Gademotta) and MIS 3, currently, it is difficult to determine whether NLCs in Eastern African contexts are technologically comparable, or a convergent result of Levallois reduction (e.g. Groucutt, 2020). Furthermore, from the current limited data, it is not possible to determine whether high rainfall, or rather other factors, such as the raw material properties of obsidian (see the section below), lie behind smaller cores at K’one and Gademotta.

Metric Variability and Raw Material

Raw material is an important factor to consider as an explanation for core size, in terms of its original nodule size, shape and availability (i.e. transport distance). All Northern African and Levantine sites exclusively used chert, as well as the Dhofar sites. In Central Arabia (Al-Kharj sites), quartzite was used, in Eastern Africa, obsidian or pitchstone, and in Southern Africa, mostly hornfels and silcrete. ‘Distance decay’ models in lithic studies follow that artefacts will be more heavily reduced, and thus smaller, the further away from a raw material source they are when discarded (Brantingham, 2003; Henry, 1989; Marks et al., 1991).

Only one study has explicitly considered NLC raw material and size variation previously, comparing cores from Al-Kharj made on quartzite—generally regarded as a lower-quality material than chert—with chert cores from Dhofar (Hilbert et al., 2016). In each region respectively, quartzite blocks were available in dimensions up to 50 cm, and the most widespread chert in Dhofar occurs in banded slabs up to 40 cm (Hilbert et al., 2016). However, AK cores are generally smaller at discard than Dhofar cores. The authors ask whether the larger and more elongated size in assemblages TH.383 and TH.377 were due to starting nodule size, concluding that while there is insufficient information on unknapped raw materials to contextualise core size at discard, the high elongation is likely to reflect deliberate production of more elongated end-products rather than raw material constraints (Hilbert et al., 2016).

Almost all Arabian sites have raw materials available at or close to the site; the exception is TH.69 which shows anomalously small core sizes (and volumes) compared to other Dhofar sites (Hilbert et al., 2016). The closest recorded chert outcrop is 250 m from TH.69. NLCs from AK-40 are also comparably small among the Al-Kharj sites, noted to be at the top of a jebel with no observed quartzite source nearby (Hilbert et al., 2016). The cores in the TH.69 assemblage are elsewhere regarded as highly standardised, although the shape (i.e. elongation) of preferential end-products is less standardised (Groucutt & Rose, 2023). In the samples overall, TH.69 shows the lowest flattening value of 1.8, meaning that cores are narrow and thick (Fig. 7), but elongation is similar to the other Dhofar assemblages. The TH.69 core convexities are maintained by very steep lateral or distal removals, the majority forming a median distal ridge of 45–75° (Groucutt & Rose, 2023). Given the smaller overall dimensions of the cores, these removals often affect the whole core shape rather than just the distal portion, while still preserving core length (EH, personal observation). The forthcoming results of whole-core 3D geometric morphometric analyses comparing TH.69 with other Dhofar assemblages will add an interesting dimension to understanding reduction intensity and core shape in this regard. Overall, this supports the earlier suggestion that elongated end-products were the goal of Nubian reduction in the Dhofar region (Hilbert et al., 2016).

Chert was also the preferred raw material in Northern Africa and the Levant. At Dimona South, rounded flint cobbles of 10–30 cm in diameter are available at the site (Oron et al., 2024). Smaller nodules (maximum dimension 5–18 cm) often show a few test removals by knappers (mean, 5 cm), the refit sequences show larger cobbles (12–25 cm) were selected for Nubian reduction (mean, 17.5 cm) (Oron et al., 2024). For the largest refitted nodule, this was split along its middle and each half reduced separately, showing that in fact smaller blocks were preferred. Reduction was relatively limited, given the discarded NLC size range of 8–10.5 cm, producing two to three preferential blanks with little surface reshaping in between (Oron et al., 2024).

In Southern Africa, the natural size and shape of raw material nodules may have had a strong influence on NLC morphology, available either as small silcrete nodules (less than 10 cm) or tabular hornfels cobbles (generally less than 25 cm) (Hallinan & Shaw, 2020; Will et al., 2015). At Tweefontein, silcrete NLCs uphold a ‘distance-decay’ model since these are the smallest cores (30–44 mm), using silcrete from sources further than 10 km away, in contrast with hornfels (the largest cores) available next to the site (Hallinan & Shaw, 2020). However, at the Eastern African site of Gademotta, obsidian nodules were available in a variety of sizes, from small pebbles to large blocks of 40–50 cm, yet NLC dimensions are very small (37–48 mm) (Douze & Delagnes, 2016). While both chert and obsidian are materials rich in silica and have homogeneous, microcrystalline structures, the latter requires less striking force during knapping (Dogandžić et al., 2020; Suga et al., 2023). Therefore, reduction of obsidian cores may be extended further than chert because it is less costly in terms of raw material volume economics and striking force. However, individual rocks vary in quality, which can affect the ease of fracture mechanisms, and no sites show reduction of both chert and obsidian to allow direct comparisons.

Given the wide range of raw materials between regions, strong geographic patterning, and the fact that NLCs are the only artefact type considered in our analysis, it is difficult to draw firm conclusions on the role of raw material on core size. Furthermore, other attributes, such as cortex percentage and number of scars, are more useful co-indicators of core reduction intensity than size alone (Clarkson, 2013; Douglass et al., 2018; Lombao et al., 2023). Core shape may be a better indicator of variation in human behaviour than size (Hilbert et al., 2016), supported by our findings that no specific environmental variables show a significant correlation with core elongation. Numerous studies observe that a characteristic of NLCs, compared with other Levallois production methods, is the elongation of the end-product (Oron et al., 2024; Usik et al., 2013; Van Peer, 1992). Our results show that while overall core elongation is high, it is especially notable in Dhofar. At TH.69, core elongation is 1.4 (lower than the other Dhofar sites) whereas product elongation is 1.9 (Groucutt & Rose, 2023). At Dimona, the ICE of cores is 1.3 and products 1.7 (Oron et al., 2024). In the Nazlet Khater assemblages, core ICE is 1.2 and product ICE is 1.7 at NK1, with respective ICEs of 1.3 and 1.9 at NK3 (Van Peer, 1992). These compare with Levallois point elongation values of 1.2 to 1.5 at Mediterranean MP sites (Oron et al., 2024), and 1.6 for Tor Faraj in Jordan (Groucutt & Rose, 2023). Since product shape is likely to be strongly related to point function (e.g. Shea, 2006; Sisk & Shea, 2009; Yaroshevich et al., 2021), the scarcity of points from Nubian cores at most sites currently limits further discussion (Hallinan et al., 2022a) (although see discussion below).

Interpreting Nubian Cores and Human Behaviour

The association between Nubian technology and arid environments has been suggested previously (Hallinan & Shaw, 2020) but is now quantitatively supported by the results presented here. The resultant question that research must now address explicitly is why this Levallois reduction method was preferred over non-Nubian Levallois point production under these environmental conditions. While cultural and demographic factors will play some role in the distribution and interaction of populations using this technology, the environmental context is central to identifying the potential drivers behind hunter-gatherer behavioural adaptations (Barzilai et al., 2022; Goder-Goldberger et al., 2016; Scerri et al., 2014).

Arid environments can be characterised as high risk for hunter-gatherers in terms of the restricted spatio-temporal distribution and abundance of resources, most critically, water. Thus, in water-poor environments especially, hunter-gatherers must employ various strategies to mitigate the risk of missing out on these resources which may be manifested in behaviour such as mobility, time-budgeting, technological organisation, and tool design (e.g. Bamforth & Bleed, 1997; Bousman, 1993; Kuhn, 1994; Torrence, 1989). In the case of Nubian technology, it is important to address whether it is the system of point production or the features of the points themselves that might confer an advantage on hunter-gatherers foraging in a high-risk environment.

For a number of reasons, there is currently insufficient evidence to reach any firm conclusions on the behavioural implications of the patterning observed; therefore, we propose a series of hypotheses to be tested in future work.

H1: Nubian Levallois reduction was more technologically efficient than other reduction methods

NLCs are the residual by-product of Levallois point production and the points themselves are scarce in most assemblages. Conversely, Levallois point cores that follow other reduction methods are generally infrequent in assemblages with abundant Levallois points (e.g. Qafzeh (Hovers, 2009); ‘Ain Difla (Clark et al., 1997), Klasies River (Wurz, 2002)), meaning that direct comparison of these methods is hampered by unequal sample size. Useful comparative data would therefore be generated through experimental production of Levallois points using different Levallois methods to test variables such as reduction efficiency, the relationship between point length and core length and mass, and the number of points that can be produced from a core (Eren et al., 2008; Picin & Vaquero, 2016). One scenario is that the distal median ridge may have provided knappers with a higher level of predetermination for pointed end-products than other Levallois methods. An alternative is that Nubian reduction was a more efficient strategy in terms of time invested in raw material procurement, lithic reduction and tool maintenance (Torrence, 1989). Given the high risk for hunter-gatherers associated with marginal, resource-restricted environments (Bamforth & Bleed, 1997; Barker, 2022; Bousman, 1993), enhanced technological efficiency may explain the adoption of this method in arid contexts.

H2: Products from Nubian Levallois cores had functional advantages over points produced by other methods

The products from NLCs in various regions are observed to be elongated pointed forms (Hilbert et al., 2016; Oron et al., 2024; Usik et al., 2013; Van Peer, 1992). Convergent flakes or points are generally interpreted as hunting weapons (Boëda et al., 1999; Iovita et al., 2014; Shea et al., 2001; Sisk & Shea, 2009) or knives (Goval et al., 2016), therefore variations in shape are attributed to either functional or cultural differences (Dibble et al., 2017). While technological and functional studies have been carried out on Levallois points produced using varied reduction strategies (Bonilauri et al., 2023; Crassard & Thiébaut, 2011; Meignen, 1995; Munday, 1976), these have been limited on Nubian Levallois products (Van Peer, 1992). Forthcoming geometric morphometric studies comparing points produced using varied Levallois methods, occurring in different environmental contexts, will provide one method to quantitatively compare morphological differences. Given the low numbers of products from NLCs in many assemblages, functional analysis on experimentally-produced points may offer further insights into the specific morphology of these points; for example, understanding the role of the steep distal median ridge in determining point shape, profile and robustness. Risk may be one explanatory factor, with Nubian points being more resistant to breakage and therefore minimising the risk of missed opportunities when hunting game. Alternatively, properties such as penetration power or aerodynamics (Eren et al., 2020; Sisk & Shea, 2009; Sitton et al., 2020; Yaroshevich et al., 2021) may have been suited to particular prey animals that inhabit(ed) arid environments across this range, such as oryx or ostrich (Blinkhorn et al., 2015; Bonilauri et al., 2007; Durant et al., 2014; Le Houérou, 1992; Stewart et al., 2019).

H3: There are no significant technological or functional advantages of Nubian Levallois reduction; therefore, cultural and demographic factors account for the occurrence of NLCs

The analyses presented in this paper have exclusively focused on the environmental context and metric attributes of NLCs. The notion of a widespread Nubian technocomplex (Van Peer, 1998) introduces a cultural and demographic dimension that is crucial for contextualising human behaviour; however, this remains a contentious issue at present. Elsewhere, scenarios of population dispersal (Rose et al., 2011) and cultural diffusion (Goder-Goldberger et al., 2016) have been proposed, but this still remains to be tested with artefact- and assemblage-level data at an inter-regional scale. Therefore, alongside the test of functional scenarios, future work must formally examine the complex constellations of technological traits observed across assemblages with NLCs to address its role in human cultural behaviour (Hallinan et al., 2022a).

The research history of Nubian cores has strongly tied their presence to Northeastern Africa (Guichard & Guichard, 1965; Hallinan & Marks, 2023; Van Peer, 1998; Van Peer & Vermeersch, 2000; Vermeersch, 2001). However, the broader spatial distribution of NLCs that has emerged since they were first named and interpreted in cultural terms, together with the environmental patterns demonstrated in this paper, warrants a shift in how they are understood in the context of human evolution. The occurrence of NLCs in two distinct time-periods, MIS 5 and MIS 3, in both contiguous and separate regions, strongly suggests that there are multiple factors that explain the adoption and spread of this technology. Thus, we anticipate that the hypotheses outlined above will be variably supported in different spatial and temporal contexts. While the low chronological resolution of surface assemblages containing many NLCs imposes some limitations on how behavioural patterns can be interpreted at a broad scale, we argue that a systematic, bottom-up approach to Nubian cores, their products, associated assemblages and spatial setting, offers a constructive pathway to identifying behavioural variability and wider cultural, demographic and adaptive trends.

Conclusion

Nubian Levallois technology is a current and heavily debated topic Palaeolithic research, particularly surrounding its definition, spatial and temporal occurrence, and demographic and cultural implications (e.g. Groucutt, 2020; Hallinan & Marks, 2023; Hallinan et al., 2022a, 2022b). Our analysis presents the first systematic inter-regional metric and environmental comparison of NLCs, identifying various statistically significant patterns. Based on the data available for our study, desert or xeric biomes and low rainfall appear to be strong environmental factors related to the presence of NLCs in MSA or MP assemblages across space and time. Furthermore, there are distinct size trends between cores dating (or presumed to date) to MIS 5, which are much larger than those associated with MIS 3. However, at the moment it is difficult to determine whether raw material type, nodule size or environmental factors are related to significantly smaller core sizes at the MIS 3 sites in Eastern Africa and Southern Africa.

Our analysis has identified broad-scale trends in NLC metrics and their distribution in space and time; however, studying only one artefact type in isolation presents numerous limitations for understanding past behaviour at a finer scale. Cores, especially Levallois, are highly informative, but other information is provided by refitted reduction sequences, toolkit size and composition, and the distribution of human activity across the landscape. Further dimensions to NLC variability are the quantitative and qualitative differences in the assemblages in which they occur (examined in other work in this volume). Sites in the Nile Valley (Nazlet Khater), Dhofar area, and Tweefontein in Southern Africa have over 100 NLCs in a single assemblage, in contrast with 38 sites with fewer than five reported cores (Supplementary Information 1). Furthermore, in some assemblages, Nubian reduction represents the dominant or the only Levallois strategy (e.g. in Dhofar), whereas in others it is accompanied by other reduction methods (e.g. the Negev). While surface site taphonomy, sampling and research bias will account for some of this variation, behavioural—and potentially cultural—factors also play some role. Our results provide a spatial and environmental foundation for future analyses examining technological variability in more detail, and testing scenarios surrounding cultural and demographic relationships.

The widespread spatial picture of NLC distribution is largely due to their visibility in surface assemblages, yet this comes at the cost of temporal resolution. The inferred—and increasingly demonstrated (Oron et al., 2024; Rose et al., 2011; Smith et al., 2007)—association of Nubian technology with MIS 5 in Northern Africa, the Levant and Arabia has implicated it in debates surrounding human expansions and dispersals. However, the apparent independent development of the technology in Southern Africa in MIS 3 highlights an important alternative adaptive scenario, given its comparable ecological setting of low rainfall and a xeric biome (Hallinan & Shaw, 2020). Whether the occurrence of Nubian technology is an indicator of shared cultural traditions, or a technological strategy geared towards a specific ecological niche, future study will offer important insights into past human adaptations at a global scale.