Three-dimensional surface morphometry differentiates behaviour on primate percussive stone tools

Abstract

The Early Stone Age record preserves a rich behavioural signature of hominin stone tool making and use. The role of percussive technology in the daily subsistence strategies of our earliest ancestors has seen renewed focus recently. Studies of modern primate tool use highlight the diverse range of behaviours potentially associated with percussive technology. This has prompted significant methodological developments to characterize the associated damage marks (use-wear) on hammerstones and anvils. Little focus has, however, been paid to identifying whether these techniques can successfully differentiate between the damage patterns produced by specific and differing percussive behaviours. Here, we present a novel workflow drawing on the strengths of visual identification and three-dimensional (3D) surface quantification of use-wear. We apply this methodology firstly to characterize macaque percussive use-wear and test the efficacy of 3D surface quantification techniques in differentiating between percussive damage and natural surface topography. Secondly, we use this method to differentiate between use-wear associated with various wild macaque percussive behaviours. By combining analyst-directed, 3D surface analysis and use-wear dimensional analysis, we show that macaque percussive behaviours create specific diagnostic signatures and highlight a means of quantifiably recording such behavioural signatures in both primate and hominin contexts.

1. Introduction and background

Behaviours do not fossilize, leading to our understanding of how early hominins used technology being derived primarily from stone tools [1–4], modified bones [5,6] and the remains of hominins themselves [7–9]. Until recently, our understanding of technological behaviours of the earliest hominins revolved around the intentional production of sharp-edged flakes used as cutting tools [10]. The past decade has shown that percussive behaviours also played an important role in hominin subsistence strategies [11–17]. The new field of primate archaeology [18] has highlighted the potential for developing a greater understanding of hominin percussive technology through the study of modern primate percussive tools. Drawing robust corollaries between primate percussive behaviours and observed damage traces would undoubtedly shed new insight into hominin behavioural diversity.

Despite the presence of multiple forms of percussive tools identified in the Earlier Stone Age (ESA), their underlying function(s) are not well understood. Localized areas of intense surface battering on convex surfaces of rounded cobbles have been interpreted as knapping hammerstones, used to make stone tools. However, behavioural interpretations for other percussive artefacts, including hammerstones with fracture angles, spheroids, sub-spheroids and pitted stones still pose a difficulty for the interpretation of the underlying behaviour [15] as different activities can leave similar use-wear patterns.

High-resolution photography and three-dimensional (3D) scanning allows quantifiable characterization of percussive damage (use-wear) on stone tools [13,19] beyond qualitative descriptive methods [20]. Recent studies have primarily focused on either two-dimensional (2D) or 2.5D analysis. These involve quantifying (i) the area and perimeters of use-wear (2D) [19,21], or (ii) the topographic structure of percussive use-wear on flat planes (2.5D) [19,22]. Such methods have been used to characterize the percussive damage produced by wild chimpanzee nut cracking [19]. These techniques have also been shown to be useful in quantifying residue distribution on wild capuchin hammerstones from Serra da Capivara National Park (Brazil) [23]. By applying the same methods to experimental anvils [13] or knapping hammerstones [22], specific use-wear patterns can be compared to archaeological artefacts to allow for more robust identification and interpretation of behaviours.

These methods, however, largely require the projection of a 3D surface in 2D. While these methods work well when damage occurs on flat surfaces, active regions of percussive tools are often not located solely on flat surfaces [23]. Percussive tools accrue damage on multiple faces over time [24]. As a result, there is a limitation on the range of percussive artefacts where these methods can be applied. To potentially overcome this issue, some studies have segmented active [19,22] surfaces or focused their analysis on visible use-wear [25], restricting the holistic scope of the analysis.

The application of open source 3D point cloud analysis software has the capacity to quantify, in high resolution, the surface morphometry of percussive tools in their entirety [26], potentially overcoming the weakness of 2D and 2.5D methods. These analytical techniques can be powerful when assessing overall changes to use-wear in experimental datasets where surfaces are quantified before and after use [25–28]. However, this method does not differentiate between damaged and un-damaged areas of a tool's surface and as such does not provide an accurate measure of the use-wear alone.

The material culture of non-human primates highlights the potential diversity of behaviours associated with percussive technology [18]. West African chimpanzees (Pan troglodytes verus), capuchins (Sapajus libidinosus and Cebus capucinos) and long-tailed macaques (Macaca fascicularis) carry out a variety of percussive foraging activities using stone tools. These include nut cracking, digging, stone on stone percussion, gastropod processing, oyster processing, seed and fruit processing [29–39]. Primate stone tool behaviours create differentiated identifiable signatures [19,21,23,36], leaving an archaeologically durable record [40–44]. These archaeological records of behaviour can be used to identify and reconstruct past primate tool use [20,43,44].

Long-tailed macaques perform a variety of percussive behaviours, using a wide range of stone tool morphologies to access inland and coastal food sources [45]. These include oil palm nuts [36,46], sea almonds [47], sessile oysters, marine gastropods as well as crabs [37]. Each target food group requires a different action pattern and tool behaviour [37]. This variety of tool use positions macaques as a unique species for characterizing the diversity of damage patterns across a range of percussive behaviours. Previous studies have shown that macaques alter the size of their hammerstones depending on the percussive behaviour being performed [37]. Furthermore, qualitative use-wear analyses have shown that the location of damage can be correlated to specific percussive activities [20]. To date, however, no quantitative analysis of macaque percussive damage has been undertaken which seeks to differentiate behaviours based on use-wear alone. The ability to infer specific behaviours from percussive tools themselves would represent an important toolkit to assess primate stone tool behaviour where direct observations are not feasible. This may include unhabituated or extinct primate groups, as well as allowing behavioural inferences of primate archaeological assemblages.

Here, we apply two related but separate analyses: the first uses 3D surface morphometry variables to measure damaged and undamaged areas of macaque hammerstones, and the second uses a combination of dimensional (area, perimeter and density) and 3D surface morphometry variables to investigate specific areas of damage on these hammerstones to distinguish between specific uses. Here, we combine analyst-directed identification of percussive damage with 3D surface analysis in a novel workflow. This allows specific areas of use-wear as well as undamaged regions to be quantified in terms of both dimensions and surface morphometry. We test the efficacy of using 3D surface morphometry alone in differentiating between damaged and undamaged regions on macaque hammerstones. This is a critical first step in assessing the utility of this method to the hominin archaeological record, as the linking of static percussive tools to dynamic behaviours requires the accurate identification of use-wear. By limiting the scope of the analysis to the visually identifiable areas of percussive damage only, a finer degree of comparable data associated directly with behaviour can be achieved. Ultimately, we apply this methodology to quantify and differentiate between different percussive behaviours undertaken by wild long-tailed macaques based solely on observable surface macro use-wear characterizations.

2. Material and methods

We analysed 18 hammerstones associated with four macaque percussive foraging behaviours: oil palm nut (Elaeis guineensis) cracking (n = 5), sessile oyster (Saccostrea cucullata) processing (n = 5), large marine shell (Thais bitubercularis) (n = 4 and small marine shell (Nerita spp.) (n = 4) cracking (table 1 and figure 1) (electronic supplementary material, figures S1 and S2). The distinction between small and large shells in this study is based on the size difference between the two target species. Thais bitubercularis is significantly larger and more robust compared to Nerita spp. [48]. All hammers were collected between 2017 and 2019 in the Ao Phang Nga National Park in Southern Thailand and chosen based on their behavioural association through direct observations and identifiable prey remains. Due to the underlying geology of different behavioural locations, all marine prey hammerstones are sandstone, while nut cracking tools are limestone. All have a coarse uniform grain structure.

Table 1.

Dimensions (mm) and weight (g) of all hammerstones included in this study.

food processed	tool ID	max length (mm)	max width (mm)	max thickness (mm)	weight (g)
oil palm	TPA2	93.91	67.47	43.33	167.70
oil palm	TPA4	96.32	74.71	34.35	240.00
oil palm	TPA7	78.89	51.25	21.47	111.80
oil palm	UNK1	63.82	39.76	27.41	66.00
oil palm	UNK3	85.75	71.08	38.91	179.20
mean		83.74	60.85	33.09	152.94
min		63.82	39.76	21.47	66.00
max		96.32	74.71	43.33	240.00
s.d.		11.71	13.25	7.84	59.56
oyster	TOB13	46.31	41.00	34.25	66.40
oyster	TOB16	57.09	43.00	40.72	92.30
oyster	TOB18	40.98	31.83	28.17	41.60
oyster	TOB5	47.61	42.33	26.14	59.20
oyster	UNK2	54.89	36.06	21.66	63.30
mean		49.38	38.84	30.19	64.56
min		40.98	31.83	21.66	41.60
max		57.09	43.00	40.72	92.30
s.d.		5.88	4.27	6.64	16.31
small shell	TLA3	70.88	61.09	28.83	96.30
small shell	TMB2	58.52	46.33	31.76	101.10
small shell	TNB11	42.59	28.06	25.44	42.90
small shell	TNB14	100.77	56.19	51.82	337.00
mean		68.19	47.92	34.46	144.33
min		42.59	28.06	25.44	42.90
max		100.77	61.09	51.82	337.00
s.d.		21.32	12.64	10.27	113.56
large shell	ETTB	121.79	82.46	63.84	857.10
large shell	ETTB7	165.00	103.17	65.05	1000.00
large shell	TTB11	140.60	84.83	45.34	641.90
large shell	TTB9	160.00	87.10	61.82	992.10
mean		146.85	89.39	59.01	872.78
min		121.79	82.46	45.34	641.90
max		165.00	103.17	65.05	1000.00
s.d.		17.10	8.12	7.98	144.89

Figure 1.

Examples of typical percussive damage patterns and depth (mm) analysis on hammerstones used for (a) oil palm processing, (b) large shell cracking, (c) small shell cracking and (d) oyster processing by long-tailed macaques. See electronic supplementary material for further images of the hammerstone sample used in this study.

2.1. Three-dimensional data capture

Three-dimensional models of all hammerstone were generated using the 3D structure from motion software Metashape (v. 1.6.5), generating high-resolution photorealistic scaled 3D models with a point density of 25 000 points cm⁻². To generate these data, a minimum of 216 photographs using a Nikon D850 with a 45.7 megapixel sensor were taken of each hammerstone. Full coverage was ensured by using Foldio360 automated turntable with a 10° interval per photograph at three separate elevations. The workflow for this methodology and associated Python code is included in the electronic supplementary material.

2.2. Extraction of use-wear

Currently, there are no computational methods for automatically identifying and extracting areas of percussive damage from non-damaged regions. All use-wear traces on each textured hammerstone model were extracted by hand using the open access software Blender (v. 2.90.1). Individual contiguous regions of percussive damage were identified visually and extracted by a single analyst with constant reference to the original hammerstones. Percussive damage on all macaque tools is fresh and readily identifiable due to a clear difference in surface colour between damaged and undamaged regions (electronic supplementary material, figures S1–S3). As such, it is unlikely that significant error in misidentifying use-wear occurred. However, to investigate whether such a misidentification error would affect the results, a comparative analysis was conducted on the 3D surface morphometry variables (depth, surface roughness, mean curvature and gradient) associated with all use-wear areas and the largest single use-wear area for each hammerstone. This comparison was predicated on the assumption that it is unlikely that a misidentification would occur for the single largest use-wear area on each tool, i.e. the most conservative assessment of use-wear. The results of a Kruskal–Wallis test show no significant difference (electronic supplementary material, table S1) between the mean value of each 3D surface morphometry variable between all use-wear and largest single use-wear area only (electronic supplementary material figure S4). All use-wear regions and remaining undamaged surface were exported for analysis (for full workflow, see electronic supplementary material). Furthermore, in the case of some large shell hammerstones, the remnants of naturally attached sessile oysters were removed digitally from the surface so as not to interfere with surface analysis (figure 1).

2.3. Data collection

General measurements and weight of each hammerstone were recorded (table 1) and a qualitative visual inspection of percussive damage was undertaken following protocols set out by Adams [49]. Surface area and perimeter for each use-wear segment and total surface area for each complete hammerstone were calculated using the 3D modelling library PyVista [50] in Python (see electronic supplementary material). Maximum dimensions of all use-wear segments were calculated from the X and Y values of a fitted 2D plane through their geometric centre using Cloud Compare [51]. Following methods outlined by de la Torre et al. [13], a number of variables were calculated. These include the percentage of use-wear area in relation to the total area of the tool surface (PA), the largest single use-wear area (LUW), the density of use-wear on the tool (D) and the density of use-wear perimeters in relation to the total surface area (ED). Additionally, use-wear elongation (maximum width/maximum length of discreet use-wear regions) was also calculated.

2.4. Three-dimensional surface use-wear characterization

Following methods outlined by Benito-Calvo et al. [26], the 3D surface morphometry was calculated directly on each hammerstone point cloud using CloudCompare [51] and Meshlab [52]. These data include depth, roughness, mean curvature, surface gradient and topographic position index (TPI). The mean curvature and roughness were calculated using a neighbourhood radius of 0.5 mm. TPI was used to extract the percentage of surface depressions and ridges on each hammerstone. Additionally, a new variable, plane morphology (PM) was calculated to quantify the relative surface curvature on which use-wear is located. PM provides a relative measure of surface curvature, with convex surfaces possessing greater mean curvature values compared to flatter regions (for a detailed description of all variables used, see electronic supplementary material). The Python libraries Scipy, and Pyvista were used to batch extract all computed variables for each use-wear and undamaged region (electronic supplementary material).

2.5. Statistical analyses

To differentiate between percussive behaviours based on use-wear, descriptive, exploratory and classifier statistical tests were used. Descriptive statistics were calculated for all variables to describe the overall percussive signature of each behaviour. The Kruskal–Wallis tests were used to compare variables between behaviours with additional Bonferroni-corrected Dunn's post hoc tests applied to identify the source of significant variation. The degree of accuracy by which 3D surface morphometry variables differentiates between behaviours was assessed through a quadratic discriminant analysis (QDA). QDA was chosen due to the non-parametric nature of the data [53]. Given that non-damaged areas always encompass a greater amount of surface area than damaged regions, additional steps were taken to balance the datasets. Each QDA is performed on an equal random number of observations for each behaviour to minimize the effect of potential sampling error introduced by subsampling. The level of accuracy is reported as the mean value derived from 1000 iterations of the QDA. Multiple variables on complete tools were used to differentiate between behaviours. The mean values for each use-wear measure were examined with a principal component analysis (PCA) to identify variables which differentiated between behaviours.

To discriminate between damaged and undamaged surfaces, tools were grouped by behaviour. Descriptive statistics were calculated for each surface variable for damaged and undamaged regions. These included mean, minimum, maximum, median and standard deviation. QDAs were performed for each behaviour individually. All statistical tests were calculated using Excel and R (v. 3.6.3) [54].

3. Results

3.1. Description of damage patterns

Hammerstones used to process oil palm nuts measure on average 83.74 × 60.85 × 33.09 mm, possessing a mean weight of 152.94 g. Damage is distributed across multiple active planes often located opposed to each other resulting in an average of 13.4 (s.d. = 11.3) discreet areas of use-wear, producing a mean density value of 0.09 cm². These represent a mean area of 175.27 mm² and an average PA value of 17.55% (s.d. = 6.05%). Percussive marks possess an average length and width of 13.3 × 8.89 mm. Use-wear dimensions, however, represent a range of values from 0.79 × 0.67 and 79.99 × 58.79 mm in the minimum and maximum length and width resulting in a mean elongation value of 0.62.

The active planes on which damage is located have a relatively flat morphology, resulting in a mean PM value of 24.895 mm⁻¹ (s.d. = 18.267 mm⁻¹). Percussive use-wear is characterized by an increased frequency of damaged ridges (57.95%) compared to depressions (42.05%). The ridges, located on active planes and in some cases across other regions of the tools, consist of smaller areas of battering often associated with fractures. In all tools, a depression is located towards the centre of each active plane measuring on average 30.3 mm (s.d. = 2.8 mm) × 23.1 mm (s.d. = 5 mm) in the maximum diameter. In most cases, central depressions have relatively shallow sides and uneven convex bases with a broadly U-shaped profile (figure 1a). Oil palm use-wear surface morphometry is characterized by a mean depth of −1.6058 mm (s.d. = 1.8304 mm), a mean surface roughness of 0.0217 mm (s.d. = 0.0176 mm), a mean curvature of 0.0063 mm⁻¹ (s.d. = 0.0145 mm⁻¹) and a mean surface gradient of 0.1719 mm (s.d. = 0.1045 mm) (electronic supplementary material, table S2).

Large shell hammerstones are substantially larger compared to all other macaque percussive tools, measuring on average 146.85 × 89.39 × 59.01 mm and weighing an average of 872.77 g. Damaged regions are generally large, possessing a length range from 3.31 to 139.38 mm and a width range from 1.78 to 113.47 mm. This results in the greatest degree of percussive damage compared to all other macaque hammerstones with an average of 1166.38 mm² (s.d. = 2634.2 mm²) and a mean perimeter of 159.51 mm (s.d. = 274.4 mm). Resulting in an average PA and LUW value of 27.08% and 15.67% respectively. The number of individual use-wear regions is, however, low with a mean number of 9 (s.d. = 5.89). Overall, damage is located across two flat active planes on opposite sides of the tool, with a mean PM value of 12.58 mm⁻¹ (s.d. = 6.51 mm⁻¹). A higher proportion of ridges (59.81%) possess areas of damage compared to depressions (40.19%). In general, however, use-wear is characterized by the development of a single large depression on each active plane measuring on average 52.11 × 39.37 mm. These depressions are broadly circular in plan and possess both rounded and irregular convex bases with a general U-shaped profile, often surrounded by heavy battering along the topographical high regions of the active planes (figure 1b). The surface morphometry shows that use-wear is characterized by a mean depth of −1.5871 mm (s.d. = 2.1194), surface roughness of 0.0142 mm (s.d. = 0.0105 mm), mean curvature of 0 mm⁻¹ (s.d. = 0.004 mm¹) and a mean surface gradient of 0.1195 mm (s.d. = 0.0804 mm) (electronic supplementary material, table S3).

Small shell hammerstones measure and weigh on average 68.19 × 47.92 × 34.46 mm and 144.32 g. They possess the greatest mean number of discreet use-wear regions (27.25), however, varying within the sample from 6 to 37 (s.d. = 14.5). Overall damage is very slight, compared to all other behaviours, consisting primarily of loosely clustered depressions measuring on average 3.46 mm × 2.43 mm (figure 1c). This results in an average use-wear area and perimeter of 11.75 mm² and 15.74 mm, respectively. Consequently, small shell hammerstones possess the lowest mean PA (2.89%) and LUW (1.4%) value of all macaque hammerstones. Percussive damage is primarily located across flatter surfaces with an average PM value of 22.61 mm⁻¹ (s.d. = 17.974 mm⁻¹). Overall, depressions (47.43%) and ridges (52.57%) are roughly equally split within use-wear regions. The use-wear surface morphometry is characterized by a mean depth of −0.8874 mm (s.d. = 0.7078 mm), surface roughness of 0.0225 mm (s.d. = 0.0105 mm), mean curvature of 0.0047 mm⁻¹ (min = 0 mm⁻¹, max = 0.198 mm⁻¹) and surface gradient of 0.1351 mm (s.d. = 0.0826 mm) (electronic supplementary material, table S4).

Oyster tools measure on average 49.38 × 38.84 × 30.19 mm with a mean weight of 64.56 g. In general, damaged consists of intense concentrated battering and surface macro pitting on the axial ends of each hammerstone. As a result, the active surface possesses a rounded and convex morphology (figure 1d). This is highlighted by a high mean PM value of 68.92 mm⁻¹ (s.d. = 29.188 mm⁻¹). Within this sample, oyster tools possessed an average of 22 discreet use-wear regions (s.d. = 17.7), with mean measurements of 4.92 mm × 3 mm; however, a greater range of length (min = 0.35 mm, max = 42.43 mm) and width values (min = 0.09 mm, max = 27.09 mm) are also recorded. Damaged areas occupy an average area and perimeter of 31.03 mm² and 27.74 mm. This results in oyster tools possessing the mean PA and LUW values of 10.49% and 1.4%. Depressions (27.85%) within use-wear are present; however, ridges (72.15%) dominate due to the location of active planes. For the same reason, the depth values of use-wear are low, with a mean value of −0.176 mm (s.d. = 0.237). These heavily battered regions possess a mean surface roughness of 0.0177 mm (ranging from 0 to 0.115 mm), and mean curvature and surface gradient of 0.0027 mm⁻¹ (s.d. = 0.0088 mm⁻¹) and 0.0797 mm (s.d. = 0.055 mm), respectively (electronic supplementary material, table S5).

3.2. Identifying use-wear from three-dimensional surface morphometry

A Kruskal–Wallis test show no significant variation in the mean values of all 3D surface morphometry variables when hammerstones are combined (electronic supplementary material, table S6). At a behaviour level, use-wear depth differs significantly from the natural surface morphology for both large shell and oyster use-wear. Large shell use-wear is significantly deeper, while oyster use-wear is significantly more shallow. No significant differences are found for other surface morphometry values in each behaviour (electronic supplementary material, table S6 and figure S5).

When all hammerstones are combined, a QDA classifies damaged and undamaged regions with an overall accuracy of 57.7%. Areas of use-wear are accurately classified in only 26.1% of cases. The degree of accuracy which use-wear is classified also varies between tools used for different behaviours (from 55.4 to 64.2%) (electronic supplementary material, table S7). The level of accuracy in positively identifying areas of use-wear for each behaviour varies significantly. Use-wear on oyster tools was accurately classified in 91.9% of cases. This is likely the result of low depth values associated with oyster use-wear due to its specific location. Small shell use-wear was classified correctly in 59.1% of cases, large shell tools in 36% of cases and oil palm tool use-wear was classified with the lowest accuracy (29.8%).

3.3. Comparative analysis between behaviours

A Kruskal–Wallis test shows significant differences in use-wear area (χ² = 94.41, d.f.(3), p ≤ 0.001) and perimeter (χ² = 77.11, d.f.(3), p ≤ 0.001) between behaviours. A Bonferroni-corrected Dunn's post hoc test shows differences in absolute use-wear area between each behaviour (electronic supplementary material, table S8). Large shell and oil palm hammerstones develop the greatest area of use-wear, with small shell and oyster hammerstones possessing significantly less (electronic supplementary material, tables S1–S4). Additionally, differences in terms of use-wear perimeter exist between all tool types apart from between oyster and small shell, and palm oil and large shell hammerstones (electronic supplementary material, table S8). The relative surface area affected by use-wear (PA) for each behaviour, however, differs to the absolute measures. A significant distinction is identified (electronic supplementary material, table S10), with a post hoc test indicating that large shell hammerstones possess significantly greater relative use-wear areas compared to small shell hammerstones, however, with no difference between other behaviours (electronic supplementary material, table S10). Significant variation exists between behaviours in elongation values of discreet use-wear regions (χ² = 12.43, d.f.(3), p = 0.006). A post hoc test (electronic supplementary material, table S8) indicates this variation is derived from use-wear elongation being greater for small shell tools compared to both oil palm and oyster damage. Small shell tools, also, possess the greatest density of individual use-wear areas, resulting in a scattered pecked appearance as opposed to a single connected larger pitted region as is seen with the large shell and oil palm tools (electronic supplementary material, figure S2). ED values differ significantly between all behaviours (electronic supplementary material, table S9); however, a Dunns's post hoc test shows that only oyster and large shell tools differ. Oyster tools possess the greatest mean ED values, with large shell possessing the smallest.

A Kruskal–Wallis test indicates significant differences between behaviours for the mean values of all 3D surface morphometry variables (electronic supplementary material, table S11). A Dunn's post hoc comparison shows that there is no significant difference in terms of use-wear depth when comparing large shell with both oil palm and small shell hammerstones (electronic supplementary material, table S12). Surface gradient differs significantly when comparing oil palm with both oyster and small shell tools, and when comparing oyster with large shell tools (electronic supplementary material, table S12). Surface roughness differs significantly when comparing oil palm with large shell tools, oyster with small shell tools and small shell tools with large shell tools (electronic supplementary material, table S12). The mean curvature differs significantly when comparing oil palm with all other tools, oyster hammerstones with small shell hammers and small shell hammers with large shell tools (electronic supplementary material, table S12). Oyster tools differ significantly in PM values from both small and large shell tools, with no difference between the other behaviours (electronic supplementary material, figure S6)

The QDA results show that classification of behaviours based on 3D surface measures (depth, roughness, curvature and gradient) of use-wear alone is low overall, at 49.22%. Oyster tools are classified with the highest level of accuracy (88.52%), with lower levels of accuracy classifying all other behaviours (electronic supplementary material, table S13). This is caused by the significant difference in use-wear depth values on oyster tools, driven by the location of the active plane. If oyster tools are omitted, large shell tools are classified with a far greater degree of accuracy (86.79%), while the classification accuracy levels for both small shell tools (43.65%) and oil palm tools (30.83%) remains low (electronic supplementary material, table S13).

Considering only 3D surface morphometry data, therefore, does not provide adequate characterization of behaviour-specific use-wear patterns to differentiate successfully between behaviours. When comparing multiple variables on complete tools, combining mean values for area and perimeter measures and 3D surface morphometry data, better behaviour-specific patterning is achieved (figure 2). The resulting PCA differentiates between the majority of tools from each behaviour. Principal components 1, 2 and 3 account for 81.87% of all variation between behaviours. The primary overlap between behaviours is associated with a single small shell hammerstone (TNB14) which possess more prominent use-wear, potentially associated with prolonged or multiple uses. Oil palm use-wear is positively correlated along PC1 with increased mean surface gradient, elongation and depression values. Oyster use-wear is negatively correlated along PC1 with PM, percentage of ridges, ED, use-wear density and depth. Small shell use-wear is positively correlated along PC2 with surface roughness, and mean curvature, while large shell use-wear is negatively correlated along PC2 with LUW and PA (electronic supplementary material, figure S7).

Figure 2.

Multivariate analysis of all use-wear variables assessed for all hammerstones from each behaviour. PCA and associated biplot showing (a) PC1 and PC2, and (b) PC2 and PC3.

4. Discussion

This study addresses two related questions regarding the analysis of percussive stone tools using hammerstones used by wild long-tailed macaques. Firstly, can 3D surface morphometry be used to differentiate between damaged and undamaged regions on percussive tools? Secondly, is it possible to differentiate between specific percussive behaviours based on 3D surface morphometry data alone?

Our results show that although all percussive behaviours leave a distinct damage pattern, some are more extensive. This is clear when considering large shell and oil palm tools, which leave significant deep, central pits compared to small shell hammerstones. The accuracy in which percussive use-wear can be differentiated from unmodified surfaces varies significantly depending on the degree and extent of the damage. Archaeological hammerstones, pitted stones [12,55], spheroids and sub-spheroids [56–58], hammerstones with fracture angles and anvils [15,55,56] have all been associated with multiple behaviours, including freehand or bipolar knapping for flake production [11,12,55–57,59,60], plant processing [11,58], nut cracking [61] or bone breaking [15]. In all cases, however, percussive damage is readily identifiable and often used as the primary diagnostic attribute. The methods outlined in this study represent a quantifiable means by which the behavioural signature on these tools can be characterized. The use of 3D surface morphometry alone, however, does not provide a robust means by which to discriminate percussive damage from undamaged surfaces. Currently, this method still relies on the visual demarcation of damaged areas in order to target subsequent analyses.

When confined to use-wear areas and assessed individually, neither 3D surface morphometry nor area and perimeter measures alone successfully differentiate between percussive behaviours. By combining both approaches, however, a far greater degree of behaviour-specific identification is achieved, with different prey targets leaving distinct, durable and quantifiable use-wear signatures.

No single measure of density, area or perimeter (mean area, mean perimeter, mean elongation value, PA, LUW, density and ED) differentiates between all behaviours in a satisfactory manner. Although absolute use-wear area differs significantly between all behaviours and absolute use-wear perimeter, between the majority of behaviours. This variation can be associated with relative hammerstone dimensions. Primates, including macaques, choose larger hammers for harder targets [37,62]. As such, it is reasonable to expect behaviour-related differences in absolute use-wear extent [20]. As a comparative measure, however, relative measures for use-wear area and perimeter are more applicable across species and behaviours [13,19]. From this perspective, these measures show few significant behaviour-related variation. Significant differences are observed, between large shell and small shell hammerstones in terms of percentage area of use-wear (PA), largest single use-wear (LUW), and between large shell and oyster tools in terms of use-wear density and edge density.

The 3D surface morphometry analysis alone shows a similar pattern of heterogeneity, although significant differences between behaviours are present in some cases. Use-wear depth and gradient separates all behaviours apart from between oil palm and large shell and large shell and small shell. Surface roughness differentiates between all behaviours other than between oil palm and small shell, and oyster and large shell hammerstones. Use-wear mean curvature differs significantly between all behaviours other than between oyster and large shell tools. Finally, the morphology of the active plane of each tool significantly differs between oyster and shell (both large and small) tools, however, does little in differentiating between behaviours where flatter active planes are preferred over convex edges.

Although often showing differences between behaviours, the measures derived from 3D surface morphometry alone are largely inadequate at classifying different percussive behaviours beyond oyster processing. This discrepancy can be explained by the differing location of oyster processing use-wear on the tool itself, brought about by how it is handled during use [38]. Oyster tools possess heavy battering along their axial convex ends as opposed to the larger flatter surfaces used in all other macaque percussive behaviours [20]. During use, these narrower ends allow for a more precisely located impact on the shells of sessile oysters adhering to large boulders. In many cases, oysters are located on the underside of boulders, and as such, oyster tools are manipulated with higher degrees of dexterity, often using precision grips, compared to other pounding hammerstones [39]. If oyster tools are removed from the sample, the behaviour that results in the greatest degree of percussive damage, large shell processing, is classified with far greater degrees of accuracy compared to both, oil palm and small shell processing. The hardness of the large marine (Thais) shells requires macaques to perform repeated heavy percussive actions compared to smaller shells. When coupled with the nature of the hammerstone raw material, a significant increase in percussive damage is observed.

As no single analysis type, neither use-wear area and perimeter measures nor 3D surface morphometry alone can successfully and consistently differentiate between percussive behaviours, both must be taken together to achieve better results. Our study has shown that, when combined, these datasets can differentiate between primate percussive behaviours in most cases (figure 2).

The type of percussive behaviour performed has a direct bearing on the surface morphometry variables included in this study. This corroborates previous studies, showing that macaque percussive behaviours produce broad use-wear differences [20,37]. Hammerstones used to process loose prey (gastropods and nuts) develop damage primarily on one or two flat active planes [20,39]. This pattern may be associated with a need to maximize the surface areas of the hammerstone in contact with the target prey. Conversely, percussive damage on hammerstones used to access sessile oysters is largely located on narrow convex axial ends and edges. The smaller morphology of these active planes may aid in more precisely directing impacts [38,39] towards preferential areas of the oyster shells. As such, percussive damage located on significantly differing active plane morphologies (seen here between oyster and all other behaviours) may, therefore, represent a signature of differing mechanical use patterns.

Differences between hammerstone damage patterns in terms of use-wear depth and extent are also present. Both mechanical [49] and behavioural factors can affect the depth of percussive damage [24]. Mechanical factors include raw material hardness or the relative hardness and morphology of the target. The latter is especially apparent when considering the disparity between the large shell and small shell damage patterns. Large shells are especially robust compared to the range of small gastropods included within the small shell category (Nerita spp.); as such, more force is needed to break them open having a significant effect on the degree of hammerstone damage.

Behavioural variation, however, may also affect the differing degrees of damage. Deep depressions on one or two planes of primate hammerstones can be taken as a signature of repeated use over time [24,63]. The restricted nature of the use-wear associated with small shell processing can be used as evidence of a shorter term, opportunistic percussive behaviour. In this case, locally available raw material is used as a hammerstone for the processing of a small number of shells and then discarded after use. Primate nut processing, however, is a behaviour that is largely structured within the landscape by the location of trees and adequate large anvils [24,46]. In these cases, hammerstones which are initially found within the vicinity of a tree and anvil are left at that location and re-used over multiple nut cracking events, potentially over multiple years [40,64]. This behaviour results in a developed damage pattern, with nut cracking hammerstones possessing deep depressions (this study) or heavy battering [36] depending on the raw material. Both diagnostic damage patterns are clearly different to the damage patterns observed for dispersed small shell processing.

An additional factor, which affects the development of percussive damage, is the degree of force employed during each behaviour. As tool-using primates vary the size and weight of hammerstones depending on the hardness of the target, the degree of force applied also changes between behaviours [37,62,65–67]. This is apparent when considering the differing damage patterns on small shell, oil palm and large shell tools (electronic supplementary material, figures S1 and S2). The target species, all inflict use-wear towards the centre of flatter planes but vary significantly in relative hardness [37] and as a consequence result in markedly different use-wear patterns. Force employed during percussive behaviour, therefore, plays a significant part in the production of differential percussive damage and has a direct effect on our ability to identify differing behaviours from their archaeological signature.

The ability to quantify and differentiate reliably between primate percussive behaviours based on their use-wear signature alone has implications for future primate archaeological studies, especially those concerned with identifying behaviour in the absence of observations. Previous primate excavations have used the relative size difference of hammerstones to infer varying percussive behaviours [43] and indeed to identify primate tool use itself [40]. By employing, the combination of quantifiable measures presented in this study to characterize damage patterns our ability to identify and understand the range of stone tool behaviours practised by either non-habituated or now extinct non-human primate populations can be greatly increased.

Linking the specific percussive behaviours with quantifiable use-wear may also provide avenues to interpret hominin archaeological percussive tool variation. Furthermore, the ability by which it is possible to identify percussive damage as opposed to the natural surface geometry of an artefact has direct implications for identifying percussive tools used by early hominins.

Both oil palm and large gastropod macaque tools develop significant percussive damage due to repeated use over time [66]. Indeed nut cracking pitted stones produced by primates [24,66] or experimentally by humans [68] are rarely associated with single instances of use and develop substantial depression over multiple behavioural events [24,66]. The sedentary and structured nature of these behaviours may suggest that hominin assemblages bearing only percussive artefacts may become visible in the Pliocene archaeological record once hominins began routinely revisiting locations within the landscape [12,69].

The use of modern analogies to develop behavioural inferences for ESA archaeological sites and material is an important avenue of research [70,71]. A primary aim of the field of primate archaeology [18] is to better understand the breadth of different primate tool use activities which in turn may provide insight into the potential variability present within the hominin percussive record [18,72]. This will only be achieved if strong referential frameworks are developed which account for a wide range of percussive behaviours, in turn allowing for greater insight into hominin tool-assisted foraging beyond the use of sharp cutting flakes [13,21,69]. By applying quantifiable methods to modern primate percussive material, primate archaeological research such as is presented here may allow new behavioural inferences for hominin percussive technology. Applying these methods directly to archaeological contexts will, however, require further considerations. The study present here is limited to two raw material types, with similar internal structure; however, a far wider variety of raw material types were used during the ESA, which in turn may affect the outcomes of such 3D surface analyses. Although individual variables such as mean depth and mean gradient can differentiate between broadly different behaviours such as oyster and oil palm processing (electronic supplementary material, figure S8), differentiating between subtle behavioural differences in percussive artefacts may require more in-depth multi-scalar approaches. Furthermore, the early archaeological record is not pristine. Percussive artefacts may have undergone post-depositional alteration, presenting limitations in determining activities beyond flaking. In these cases, identifying and distinguishing more ephemeral use-wear, similar to that produced during macaque small shell processing, from post-depositional alteration will be difficult. Additional future efforts to quantify such light percussive damage and to overcome the issues of taphonomic decay are needed. In doing so, greater insight into the variety of potential percussive behaviours undertaken by hominins might be achieved. Where percussive damage is evident on archaeological artefacts, using the methods outlined in this paper, there is potential to differentiate, at a fine scale, the behaviour behind percussive damage patterns.

5. Conclusion

By linking observed macaque percussive behaviours to quantifiable surface modifications on hammerstones, we have shown that an analyst-directed approach to 3D surface morphometry distinguishes between specific percussive behaviours. Using only the 3D surface morphometry measures in this study, however, currently cannot achieve the same results and in some cases does not successfully distinguish between damaged and undamaged areas of hammerstones without analyst interpretation. This study, however, does provide a quantifiable method applied at a macro scale, by which identifiable use-wear either on primate or hominin hammerstones can be repeatedly characterized. The development of standardized quantifiable experimental, ethnographic and primate references is a fundamental stage in developing a framework by which to compare percussive technologies of both primates and hominins.

Data accessibility

Data are available in the electronic supplementary material, and all custom scripts are maintained on the following github repository: https://github.com/reevesj191/Proffitt_et_al_3D_Surface_Morphometry. Additional sample data are available at the following Figshare.com links: Small Shell_TLA3_data.csv. https://doi.org/10.6084/m9.figshare.16573250; Oil Palm_TPA2_data.csv. https://doi.org/10.6084/m9.figshare.16573253; Large Shell_ETTB_data.csv. https://doi.org/10.6084/m9.figshare.16573256; Oyster_TOB13_data.csv. https://doi.org/10.6084/m9.figshare.16573259.

The data are provided in electronic supplementary material [73].

Authors' contributions

T.P. conceived of the study, designed the study, coordinated the study, undertook formal analysis and wrote the manuscript. J.S.R. designed the study, undertook formal analysis and wrote the manuscript. A.B.-C. undertook formal analysis and wrote the manuscript. L.S.-R. undertook formal analysis and wrote the manuscript. A.A. undertook formal analysis and wrote the manuscript. S.M. provided resources and wrote the manuscript. L.V.L. conceived of the study, designed the study and wrote the manuscript. All authors gave final approval for publication and agree to be held accountable for the work performed therein.

Competing interests

We declare we have no competing interests.

Funding

 This research was funded by the Max Planck Society. A.A. is supported by the Spanish Ministry of Science and Innovation (IJCI-2017-33342, Subprograma Juan de la Cierva- Incorporación). The Institut Català de Paleoecologia Humana i Evolució Social (IPHES-CERCA) has received financial support from the Spanish Ministry of Science and Innovation through the ‘María de Maeztu’ programme for Units of Excellence (CEX2019-000945-M).