Introduction

Paleolithic spatial archaeology studies have been evolving since their inception in the 1950s (Leroi-Gourhan 1950; Laplace and Méroc 1954a, b; Clark and Evans 1954; Dacey 1963; Clarke 1968; Davis 1975; Hodder and Orton 1976; Hietala and Stevens 1977, inter alia), with the 1980s marking a turning point when the concept that sites could change by the action of natural processes after their abandonment by humans and do not remain undisturbed until studied was taken on board (Behrensmeyer 1982, 1988; Schick 1984; Hassan 1987; Walter 1992, inter alia). Moreover, the incorporation of Geographic Information Systems into spatial studies, in combination with new analytic methods derived from other disciplines, has entailed a leap forward in data analysis and the type of results obtained (Benito-Calvo et al. 2009, 2011; De la Torre and Benito-Calvo 2013; García-Moreno et al. 2016; Sánchez-Romero et al. 2016, 2020; De la Torre and Wehr 2018; Spagnolo et al. 2019, 2020; Coil et al. 2020, inter alia). Knowledge of the evolution of the site and the processes that could have taken part in its formation (human activities, animals, and/or natural factors) allow for integral comprehension by combining of data from different disciplines, not only archaeology (Camarós et al. 2013, 2017; Arilla et al. 2020). Therefore, an understanding of the nature of the deposit is essential, because it is very likely that the processes that generated the accumulation of the archaeopaleontological materials and the formation of the host sedimentary unit could have been different. Most of the archaeological remains that we find in a site, to a greater or lesser extent, have been affected by similar post-depositional processes. For this, and prior to any interpretation of the site function, spatial distribution, etc., characterizing these processes and how affected the materials are turn out to be even more important. From an archaeological perspective, the identification and differentiation of significant groupings of archaeological materials can furnish a chrono-cultural coherence to the site under study (Sánchez-Romero et al. 2016, 2020). Thus, spatial archaeology has to consider and work together with other disciplines that analyze the site, positioning it in a middle point where these different fields meet. Given all this, it ought to be possible to evaluate the conservation and preservation of the materials, as well as the processes and organizational patterns of the archaeological site.

Open-air archaeological sites in the Cantabrian region are not common, although in recent years there is rising interest in studying this kind of settlement. This change seems to be due to diverse circumstances, such as the near-completion of the study of known cave deposits, the growing interest in Lower Paleolithic sites, and fortuitous findings during infrastructure constructions or rescue archaeology (Arrizabalaga et al. 2015). Although the interest in these sites is quite recent, the first references date back to the late 1950s, with the discoveries of José Miguel de Barandiarán in the Kurtzia area (Barrika, Bizkaia) (Barandiarán-Ayerbe et al. 1960) and the work by J. M. González-Fernández in Asturias (1968). These studies were followed by others, which resulted in the appearance of sites like Cabo Busto (Rodríguez Asensio 1996), La Verde (Montes 2003), Irikaitz (Arrizabalaga and Iriarte-Chiapusso 2011; Arrizabalaga et al. 2015), Ametzagaina (Calvo et al. 2013; Arrizabalaga et al. 2015), and sites in the Uribe Kosta area (Rios-Garaizar 2015), such as the Aranbaltza archaeological complex (Rios-Garaizar et al. 2012a, 2018, 2020) (Fig. 1). This site was originally discovered by J. M. Barandiarán in 1959 and named Ollagorta (Barandiarán-Ayerbe et al. 1960; Rios-Garaizar et al. 2020). During 2004, in the course of the construction of a sewage trench, the site was partially destroyed, and many lithic artifacts were recovered from the disturbed sediments, including Châtelperronian materials, such as points and bidirectional cores (Rios-Garaizar et al. 2012a). During 2013, the excavation of a trench revealed a well-preserved stratigraphy in Aranbaltza II, including a level (US4b) with abundant Châtelperronian lithic materials, which became the only reference of an open-air Châtelperronian occupation in the Iberian Peninsula. This served as starting point for the successive investigations that allowed the discovery of the other Aranbaltza sites (Aranbaltza I and III) (Rios-Garaizar et al. 2018).

Fig. 1
figure 1

Map of the archaeological complex of Aranbaltza and plan with the location of all sites and trenches

Full size image

The Châtelperronian in the Northern Iberian Peninsula has been identified in several cave sites, such as Cueva Morín, El Pendo, Labeko Koba, Ekain, or Cova Foradada (Andrés-Herrero 2009; Rios-Garaizar et al 2012b; Morales et al. 2019). Aranbaltza II is the first open-air site in this region and it is comparable to other open-air sites in neighboring areas, such as SW France, especially the area between Biarritz and Bayonne, and the area of Bergerac (Chauchat and Thibault 1968; Bachellerie et al. 2007; Grigoletto et al. 2008; Bachellerie et al. 2011). In this sense, Aranbaltza II is a very relevant site since the Châtelperronian presence in the eastern Cantabrian Region has been previously defined as very sporadic, with small cave sites used as hunting camps (Rios-Garaizar et al. 2012b). Aranbaltza II would exemplify a completely different type of occupation, probably representing a site mostly devoted to the production of lithic tools but where, as it happens in other sites such as Vieux Coutets (Grigoletto et al. 2008), other activities have been also developed. This probably means that, during the Châtelperronian, open-air occupations were especially relevant in comparison with regional Middle Paleolithic.

This work aims to address the accumulation processes of the Aranbaltza II site (US4b level), with extraordinary characteristics in terms of the accumulation of materials in a small area. This site is determined by the high-density of remains in singular concentrations, located in a specific zone of the site and with well-defined morphologies. This pattern of high-density accumulations in small areas has already been identified in other Châtelperronian open-air sites, so this study aims to know what factors could generate these accumulations and infer the processes that acted in the formation of similar concentrations in sites of same chronology and context.

Aranbaltza II

The Aranbaltza II site is composed of three excavated areas (Area 1, 2, and 3), with a total excavated extent of 18.14 m2. The stratigraphic sequence has been defined from the excavation of Area 1 and 3 (about 14 m2). The sequence comprises several sterile units in the base (US6, US5 and US4c), a Châtelperronian occupation (US4b), two mixed Châtelperronian-Mesolithic units (US4a and US3), a Chalcolithic occupation (US2), and sediments recently reworked and soil (US1 and US0) (Fig. 2). In this study, we focus on the unit US4b, a deposit mainly dominated by fine sands and with abundant Châtelperronian lithic pieces. This deposit is characterized by the presence of small structures of lenticular shape that comprise lithic material of all sizes and a chaotic spatial arrangement. Regarding the archaeological assemblage, the absence of bone remains and the use almost exclusively of flint for the elaboration of lithic tools stand out, although there are also other raw materials such as sandstone, trachyte, quartzite, ochre and mudstone. The most remarkable aspects are the high number of pieces found in a relatively small and well-delimited area and the presence of blade and bladelets cores, laminar products, and different kinds of waste generated during knapping and tool configuration (Rios-Garaizar et al. 2016, in press). It is also important to emphasize the small number of retouched tools, although there are significant tools like Châtelperronian points, backed blades, burins, and retouched blades and bladelets (Rios-Garaizar et al. 2016, in press). This level has a preliminary dating to ca. 44.000 BP (Rios-Garaizar et al. in press).

Fig. 2
figure 2

Lithostratigraphic sequence of Aranbaltza II and description of the different units

Full size image

Methodology

Databases

The data collection is always a key step in any archaeological study but in the case of spatial studies and the analysis of site formation processes, this issue becomes even more important. In the case of Aranbaltza, all the materials larger than 2 cm have been recorded by total station with respect to a local coordinate system. On the other hand, the smallest pieces were recovered in a variable circular area of about 25 cm diameter and collected in sediment bags which were sieved with small meshes (< 1 mm). An XYZ point is taken in the middle of this diameter with the total station, in order to record the area of collection. The key point is that this criterion is maintained. From 2016, the axes of the pieces began to be recorded with the aim of calculating the orientation and slope of those pieces larger than 2 cm. The longest axes are calculated as between to two points taken by total station that define the axis in each piece. Since the smallest pieces do not contain XYZ information, we generated randomly estimated coordinates for each of them, considering a circular area with a radius of 15 cm whose center is that of the excavated area where these materials were collected. This central point was registered by total station. The use of randomly estimated coordinates is a very useful resource when it comes to incomplete data or data from old excavations (Rios-Garaizar 2012; Blasco et al. 2016; Sánchez-Romero et al. 2020). When this kind of information is handled, it is important to keep the area limits where random coordinates are projected clearly distinguished, as well as the characteristics of the deposit and the site. In the case of Aranbaltza II, we created a polygon shapefile with circumferences of radius 15 cm from the XYZ point taken with total station. Thus, we obtained a map with the dispersion of all pieces, both those with recorded XYZ data and with estimated coordinates. The spatial analysis of Aranbaltza II was carried out taking into account materials with XYZ data (> 2 cm, which have been named PR, Recorded) and smaller (< 2 cm) pieces retrieved from 25 cm diameter circumferences (PS, Simulated).

Patterning analysis

The identification and analysis of the distribution patterns was conducted through the application of kernel density (Silverman 1986) and hotspots analysis, both Getis-Ord Gi* (Getis and Ord 1992) and Anselin Local Moran’s I (Moran 1950). Prior to applying these methods, we analyzed the general distribution of the materials (clustered, dispersed, or random), since the hotspot method only provides relevant results when the distribution is clustered. For this, the chi-square, Kolmogorov–Smirnov, and Average Nearest Neighborhood (ANN) stats were applied to test whether the materials were dispersed, clustered, or randomly distributed (Sánchez-Romero et al. 2021). Kernel density analysis generates a continuous surface that indicates the maximum and minimum areas of concentration of features. In this case, and due to the number of pieces and the extent of the total excavated area (18.14 m2) and the area where the study has been focused (7.14 m2), the search radius was 0.30 m. The density map was classified according to natural groupings inherent in the data (Jenks 1967), deleting intermediate classifications and separating the most important groupings from those less relevant (Sánchez-Romero et al. 2020). Regarding the hotspots analysis methods, both Getis-Ord Gi* and Anselin Local Moran’s I identify the concentration of high and low values. The main difference between them is that Getis-Ord Gi* allows the identification of the clustering according to the statistical significance, while Anselin Local Moran’s I identifies atypical spatial values (Siabato and Guzmán-Manrique 2019; Sánchez-Romero et al. 2020). These methods allow the application of the FDR correction, a filter that reduces the error thresholds of p-values, classifying the values in relation with their statistical significance and eliminating the weaker ones. Apart from applying these methods to individual features (points), we also calculated the distribution patterns according to the number of remains by square. These squares were not excavation squares, but a grid created considering the size of the area and the number of points to for analysis. To do this, we applied the quadrat method (Lee and Wong 2000) to compare the results, both individual and per square. With the aim of evaluating the shape pattern of the materials, we applied ternary diagrams of the representation of particle shape (Sneed and Folk 1958). This lets us observe whether there exists any material selection according to shape and to infer possible post-depositional processes that could have affected the studied assemblage. Finally, the directional patterns of each group were analyzed through the application of the directional distribution method (Mitchell 2005; De la Torre et al. 2013; Sánchez-Romero et al. 2020), which measures the trend of a set of elements by calculating a standard deviational ellipse. The analysis was performed with 2 standard deviations as 95% statistical confidence.

Fabric analysis

The fabric analysis was performed on a total sample of 99-piece major axes. The fabrics are very dependent on the elongation ratio (Drake 1974), so all the pieces considered have Ie =  > 1.6 and are larger than 2 cm (Bertran and Lenoble 2002; Lenoble and Bertran 2004). The orientation pattern calculation was performed through the analysis of the histograms trend together with different statistical tests, considering both axial and azimuthal angles (Sánchez-Romero et al. 2020). The combination of orientation and slope data yielded the 3D fabric projection, in order to discover whether the identified clusters show planar, linear, or isotropic fabrics (Benito-Calvo et al. 2009, 2011). To do this, the eigenvectors method (Woodcock 1977; Benn 1994; Benn and Ringrose 2001; Lenoble and Bertran 2004; McPherron 2005; Lenoble et al. 2008) was applied, obtaining S1, S2, and S3 values and their projections in the fabric triangle (Benn 1994; Bertran and Lenoble 2002; Graham and Midgley 2000; Benito-Calvo et al. 2009, 2011).

Results

Cluster definition

The density analysis applied to both the PR and PS pieces/points did not throw up significant differences between them, since zones of maximum accumulation of pieces are the same in the two projections (Fig. 3). The result obtained with the Jenks classification is practically the same in both datasets, although in the case of PR, the densest zones are contiguous as in the case of PS for kernel analysis (Fig. 3). The differences between PR and PS, with regard to geometries and extent of the high-concentration zones obtained with kernel analysis, are minimal. The selection of the clusters was carried out applying Getis-Ord Gi* and Anselin Local Moran’s I, offering more insight into the results and providing crucial data for defining the clusters.

Fig. 3
figure 3

A Kernel density map and B Jenks classification map for PR data. C Kernel density map and D Jenks classification map for PS data. The green (B) and brown (D) lines delimit the densest area identified by kernel density analysis

Full size image

The type of distribution was analyzed by the chi-square (X2) and Kolmogorov–Smirnov (K-S) tests using the numbers of pieces (n = 5414) and squares (n = 656), with the aim of discovering the relationship between the two in their distribution over the whole area. The results obtained indicate that pieces are clustered and not randomly distributed, the critical values being lower than those calculated for X2 and K-S with a significance value of p = 0.95 (X2 (44) = 3.09416E + 44 > 55.758; K-S (656) = 0.334 > 0.053). ANN and spatial autocorrelation were also performed, to reveal the distributional patterns of the pieces and those contained in each square. The results obtained for PR indicate that the materials are clustered, both in ANN (z-score =  − 12.23) and Global Moran’s I (z = 1.72); meanwhile, spatial autocorrelation shows that there is a peak at 1.12 m with a z-score of 2.47, this being the distance where the assemblage is most clustered. Regarding the results obtained for the squares, we calculated the type of distribution for PR and PS datasets contained in squares. ANN calculates the distribution with respect to the squares, while Global Moran’s I analyzes the distribution through all the squares and the values contained in each square, as well as how these relate to each other. ANN results obtained for PR indicate that the pieces are dispersed (z-score = 44.54), while Global Moran’s I shows a clustered pattern (z-score = 6.15). In the case of PS, the ANN result is equal to that for PR (since the square distribution is the same) and Global Moran’s I indicates a clustering of the materials (z-score = 24.88). Regarding the incremental spatial autocorrelation, there are differences between PR and PS. In PR, two peaks were detected: one at a distance of 1.04 m (z-score = 6.58) and another at 1.13 m (z-score = 6.69); in PS, no distances with statistically significant clustering were detected.

The hotspot analysis of distributional patterns according to length for the PR dataset (XYZ) did not provide statistically significant data, since the groups detected only comprise a few pieces (between 0.18 and 0.66% of the whole assemblage). Similar results were obtained in both analyses (Getis-Ord Gi* and Anselin Local Moran’s I) (Figs. 4 and 5), which detected statistically significant clusters but with a sample size too low to reach any solid conclusion about the clustering pattern in relation to piece length.

Fig. 4
figure 4

A Getis-Ord Gi* projection for the PR points according to the maximum piece length. Application of the FDR correction (B). The percentages are calculated over the whole assemblage (n = 5414)

Full size image
Fig. 5
figure 5

Anselin Local Moran’s I projection for the PR points according to maximum piece length. The percentages are calculated over the whole assemblage (n = 5414)

Full size image

The distributional patterns in relation to the number of pieces per square were determined using the quadrat method. In the case of Getis-Ord Gi* (Fig. 6), the result obtained for PR reveals several dispersed zones whose concentration of remains is statistically significant; meanwhile, in the case of PS, the zones are clearly delimited. In both cases, application of the FDR correction reduces the concentration zones significantly. On the other hand, Anselin Local Moran’s I (Fig. 7) shows very similar results to those obtained in Getis-Ord Gi*, for both PR and PS datasets.

Fig. 6
figure 6

Getis-Ord Gi* map for the squares of the PR (A–B) and PS (C–D) data

Full size image
Fig. 7
figure 7

Anselin Local Moran’s I map for the squares of the PR (A) and PS (B) data

Full size image

The results obtained after density analysis and classification by Getis-Ord Gi* and Anselin Local Moran’s I indicate four statistically significant clusters (Fig. 8). The groups classified show that the main concentrations of materials partially coincide with the areas highlighted by kernel analysis in the case of PR data. Thus, all the analyses applied seem to indicate that the denser concentrations of materials according to statistical criteria are very similar to those areas delimited by kernel analysis for the PR points.

Fig. 8
figure 8

Application of Getis-Ord Gi* (A–B) and Anselin Local Moran’s I (C–D) to kernel density values obtained from the PS data. Clusters resulting after the classification of the kernel density analysis by Getis-Ord Gi* and Anselin Local Moran’s I (E)

Full size image

Defining the clusters is the first step in investigating the composition and characterization of the pieces included in each, as well as for analyzing the formation of the site further.

Cluster composition

All the analyses performed have identified the existence of four statistically significant clusters in terms of number of pieces: AC1 (n = 607), AC2 (n = 235), AC3 (n = 1780), and AC4 (n = 144) (Fig. 8). Two clusters were unified (forming AC3) to constraint the statistical significance of all the clusters and avoid small ones whose sample size could be problematic. All groups are mainly composed of flint pieces, since this is the most common raw material in the Aranbaltza II assemblage. The analysis of each cluster includes the description of the shape patterns, which was only conducted in the PR materials. Thus, the characteristics of the identified groups are as follows:

  • AC1: Materials show a mean length of 26.46 mm (Table 1) and with predominance of Flysch type flint, although there are also some sandstone and quartzite pieces. The most common type of support is cortical flakes (n = 25), followed by flakes (n = 22) and blades (n = 20) (Table 1).

  • AC2: The mean length for materials in this cluster is 25 mm (Table 1). Again, the most common raw material is flint, although there are one piece of sandstone and two ochre fragments. This cluster contains fewer pieces, flakes being the most abundant type of support followed by blocks (n = 6) and cortical flakes (n = 4) (Table 1).

  • AC3: In this group, the mean length is 27.11 mm (Table 1) and flint is the main raw material. However, there are also pieces of sandstone (n = 13), quartzite, quartz, and mudstone. As in the case of AC1, this cluster is characterized by numerous cortical flakes (n = 107), as well as flakes (n = 75) and bladelets (n = 71) (Table 1).

  • AC4: This cluster, together with AC2, contains the smallest number of pieces and with a mean length < 26 mm (Table 1). It is dominated by flint and cortical flakes (n = 8).

Table 1 Lengths and types of support for the lithic pieces in the clusters identified at the US4b unit of Aranbaltza II. The units are expressed in millimeters
Full size table

Rest: This group is composed of the statistically not significant materials, which comprises 2823 pieces and a mean length of 28.48 mm (Table 1). The main types of support are cortical piece (n = 298), flakes (n = 217), and blades (n = 196) (Table 1).

The shape patterning analysis indicates that all clusters have a similar patterning, with some variations depending on the number of pieces included in each. The graphs show a predominance of flat or very flat and platy and bladed shapes in the identified clusters of the US4b assemblage (Fig. 9). There are very few elongated pieces and we do not observe any kind of tendency or selection in their shapes: rather, this is the expected pattern for a Châtelperronian lithic assemblage.

Fig. 9
figure 9

Shape pattern graphs of the lithic materials contained in the identified clusters

Full size image

Analysis of the clusters

The different tools and analytical methods applied have allowed four main clusters to be identified, which could reflect the intervention of different processes in the formation of the Aranbaltza II site. A deeper analysis was performed with each cluster, aiming to ascertain the possible processes involved in their formation and their impact on the archaeological assemblage. Thus, the recording of orientation and slope data has allowed the fabric patterns to be elucidated. The total number of pieces with that information is 99: 12 pieces for AC1, 1 for AC2, 35 for AC3, and 2 for AC4. Thus, we also considered pieces not included in the clusters (Rest, n = 49) and all the available data (All, n = 99). Because the clusters AC1 (n = 12), AC2 (n = 1), and AC4 (n = 2) include a low number of pieces with information about axes, these clusters have not been considered in the analysis. In the context of the number of axes handle, AC3 (n = 35) contains an amount possibly comparable to Rest and All, which are not clusters. Rayleigh and Kuiper tests were performed considering azimuthal (0–360°) and axial (0–180°) angles, with the aim of comparing the two. The results obtained show some similarities among the axial projections, while the azimuthal ones point to certain particularities distinguishing the clusters. The calculations were performed considering the archaeological north, which corresponds to the local coordinate system. The interpretations regarding the site, its location respect to the valley, the Urgozo stream, etc. have been carried out considering the geographical orientation.

For AC3, the results obtained do not allow uniformity to be rejected against unimodal and multimodal tests, either in azimuthal or axial data. Thus, preliminary interpretation seems to suggest a no preferred orientation for the AC3 cluster (Table 2), although the small sample size precludes a solid conclusion. However, when these data are compared with the histogram, it is possible to observe a slight trend to a NW–SE orientation (Fig. 10). The axial results for Rest (n = 49) show high p-values for Rayleigh (p = 0.989) and Kuiper (p > 0.15), evidencing the uniformity of the assemblage. Nevertheless, in the case of azimuthal data, the p-values for both tests are low (Rayleigh, p = 0.055; Kuiper, 0.10 > p > 0.05). This group, even at the limit of 95% (94.5%), could be considered as slight evidence of preferred orientations (Table 2).

Table 2 Results for the azimuthal and axial orientation patterns of AC3 cluster and Rest and All groups identified in Aranbaltza II
Full size table
Fig. 10
figure 10

A Circular graphs obtained after the calculation of azimuthal and axial orientation patterns. B Slope graphs for AC3 cluster and Rest and All groups identified in Aranbaltza II

Full size image

Lastly, the results for All (n = 99) indicate quite different patterns depending on axial or azimuthal calculations. The axial data show a clear uniformity, with high p-values for Rayleigh (p = 0.787) and Kuiper (p > 0.15). However, when these results are compared with the azimuthal data, the pattern changes (Fig. 10). The p-values are lower for Rayleigh (p = 0.009) and for Kuiper (p < 0.01), indicating a clearly oriented pattern.

Regarding the slope, the cluster AC3 contains some pieces in vertical position, something not observed in the rest of the clusters and groups analyzed (Fig. 10). This result matches the type of accumulation, where most of the pieces are overlapped and display a random trend in orientation. On the other hand, most of the assemblage pieces are horizontal and their orientation seems to be influenced by the depositional slope of the sediments where they lay. The combination of orientation and slope data allowed 3D fabric analysis to be performed. This fabric analysis using eigenvectors and Benn’s projection delves into the characteristics of the materials in each identified cluster. The AC3 concentration is situated in the upper part of the triangle (Fig. 11), suggesting an isotropic fabric for these pieces. This is consistent with the uniformity trend obtained after the calculation of orientations and the variability observed in slopes. The features analyzed for this cluster indicate several issues to consider, such as the high values for the isotropy parameter (I = 0.509) and the low elongation of the fabric (E = 0.102) (Table 3), which indicates that there does not appear to exist any shape pattern in the accumulation of the pieces. The K index (Woodcock 1977), which distinguishes planar (0 < K < 1) from linear (1 < K < ∞) fabrics, shows the highest values for all the clusters analyzed (K = 0.189) (Table 3). Although AC3 lies within the range defining planar fabrics, the trend of this cluster is clearly isotropic, located nearby the middle of the Benn’s diagram.

Fig. 11
figure 11

Benn’s fabric diagram with the 3D fabric results for AC3 cluster and Rest and All groups identified in Aranbaltza II

Full size image
Table 3 Fabric indexes calculated for AC3 cluster and the Rest and All groups. K and C parameters have been calculated according to Woodcock and Naylor (1983), while the rest of parameters have been obtained according to the Benn (1994) method, using the triangle of Sneed and Folk (1958). C. L. refers to the confidence level
Full size table

Regarding the Rest group, this is located in the lower left corner of the triangle (planar), far from isotropic and linear fabrics (Fig. 11). This group has the lowest K index (K = 0.01), which indicates that is clearly planar. It is also important to highlight the characteristics of other parameters, such as the low values for elongation (E = 0.018) and horizontal indexes (F = 0.144), as well as for isotropy (I = 0.142) (Table 3). Finally, the analysis for All points to a planar fabric, this group being located in the lower left corner of the triangle (Fig. 11). The K value is low (K = 0.042), but higher than for Rest (K = 0.01), which confirms this planar trend. In general, the pieces are not very elongated (E = 0.050) and the position is mainly horizontal (F = 0.295). The isotropy value (I = 0.281) shows clear randomness of the materials, which would match with the azimuthal and axial results (Fig. 10, Table 3).

According to the results obtained after the application of directional distribution, AC2, and Rest and All groups, shows almost identical eccentricity, something that also occurs with the AC1 and AC3 clusters. The eccentricity value for AC4 indicates that the ellipse is nearly circular (e = 0.454) (Fig. 12). Regarding the orientation, AC3, Rest, and All show very similar patterns, while AC1, AC2, and AC4 are different from each other and in comparison with the remaining clusters and groups. In the case of AC4, since the ellipse is practically circular, this not considered further for the study (Fig. 12). The most similar ellipses are those obtained for the Rest and All groups, and AC3 to a lesser extent, the cluster and groups with the most pieces (Fig. 12).

Fig. 12
figure 12

A Map and table with the values and ellipses obtained for the main clusters identified in Aranbaltza II, as well as for the Rest (B) and All (C) groups

Full size image

Thus, the clusters with the greatest number of pieces (AC3, Rest, and All) show very similar orientation patterns, even though their shapes are different, especially in the case of AC3 when compared to Rest and All. By contrast, AC1 and AC2 seem to form another group of clusters with very similar orientation patterns, and very different from AC3, Rest, and All. This method has allowed us to compare the different accumulations, from smallest to largest, as well as the whole assemblage. The aim of this analysis is to identify processes that could have participated in the formation of the site, from the description and quantification of the inherent attributes of the clusters and the pieces that form them.

Discussion

Several analytical tools were applied to studying US4b level of Aranbaltza II, which has some particular features, such as the absence of faunal remains, a high concentration of pieces, and the existence of lenticular bodies in a specific zone of the site with most of the lithic materials. The identification and classification of the clusters were performed using statistical criteria, combining several analytic techniques and not only considering the maximum accumulation pattern identified by kernel density analysis. The four clusters identified have served to structure the study area and to infer the processes that could have acted in the accumulation of materials. This classification was accomplished based on the total amount of materials recovered at the site (PS, n = 5414) and not only according to the pieces recorded with the total station (PR, n = 1649). This gives us the real projection of all coordinated materials and an approximation for the small pieces (< 2 cm). When all materials were projected, a high concentration of pieces was observed in the west of the excavated area, although after the projections, we realized that the amount of materials in the rest of the area was greater, but it is slightly more scattered. The classification by clusters has allowed us to see there are 1780 pieces in the densest zone, while in the rest there is a total of 3634 pieces. AC3 comprises almost 33% (32.88%) of the whole assemblage.

The main issue to consider is the relation of AC3 with the U-section of well-delimited lenticular bodies (Fig. 13). The resulting materials from any kind of anthropic activity would not show this U-section, unless they are areas previously excavated (such as small basins, cuvettes) that could be dumping zones. However, the morphologies of such areas are not similar to what is seen at Aranbaltza II, since they are smaller and with a narrow entrance that widens, and where the materials are accumulated. The characteristics described for these areas of waste (Yellen 1977; Binford 1978; O'Connell et al. 1991; Stevenson 1991) do not seem to match those found in Aranbaltza II. In addition, the materials are mixed with sands and not deposited as it is expected for unmixed materials.

Fig. 13
figure 13

Morphologies located at the western side of the excavation area, where most of the pieces of this part of the site are concentrated. View from the top (A) and section view (B)

Full size image

The results obtained after the analysis of distributional patterns, in combination with data provided by sedimentology, suggest that the processes causing these accumulations of pieces would not be anthropic activities. Thus, it is important to consider what kind of natural processes can generate these distributions. The site is located on a sandy bank of a fluvial stream, so one of the options is the accumulation of materials as a result of water inflows and deposition from the Urgozo stream, such as the crevasse splay process that has been already described in Unit 4 of Aranbaltza III (Rios-Garaizar et al. 2018). Another option is the deposition of materials by runoff or rills that run down the slope, processes that could be from increases in the river flow and/or precipitation. A rise in precipitation can cause runoff and flows that, as in the case of small streams, displace and mix the materials (Goudie 2004). These rills are small and shallow channels, which erosive action depends on the slope and the water amount transported. Sand deposits are more susceptible to the action of these processes, unlike clays, which are usually more resistant to their formation (Loch and Thomas 1987).

The orientation patterns and the results obtained after the directional distribution analysis have provided useful data for defining the processes that could have affected the site. The patterns of AC3, Rest, and All are clearly random, although with a slight NW–SE tendency, something that has been also detected in the ellipses describing AC3, Rest, and All. As previously noted, and considering the orientation and slope patterns, 3D fabric analysis and ellipses, the fabric of the pieces seems to be influenced by the deposit slope, which dips to the Urgozo stream. The lack of size selection, the large number of pieces, and their preservation are factors that point to a short-distance transport, which would have not allowed a size bias. Regarding the materials, no selection by size has been observed, since the mean is very similar between the clusters (27.11 mm), where there are pieces from 6.3 to 100.7 mm. Furthermore, the shape pattern does not differ from the other materials, because most of the pieces are slightly massive or narrow and quadrangular or rectangular. In light of these results and the low energy of the flow, the deposit could have been affected by channeled flows associated to flood events that would have reworked the materials. These shallow, low-energy channels would have eroded and transported sand, leaving the lithic materials and generating this concentration of pieces we find at the site. All data indicate that the flood processes were of very low energy, where finer particles (sands) would have been transported and the larger materials (lithic pieces) would have remained, experiencing minimal and limited transport. This displacement of materials would not have occurred over long distances, but probably from zones very close by.

The fact that the pieces are so well-preserved is consistent with the process described by Isaac (1967), in which they could more or less maintain their original position following burial, and hence be protected from the water torrent. Apart from this, the shape of the pieces seems to be decisive to the degree of transport and movement (Schick 1984), with spherical pieces (like cores) being the most likely to move. The morphology is also a decisive factor in the stability of the pieces, since the flat ones tend to adopt a more stable position in the substrate until there is enough energy to move them (Schick 1984). As we saw previously, most of the lithic materials at Aranbaltza II are elongated and thin, so water action could have transported them but not far, since it is possible that the pieces had been buried quickly and protected from moving further. Thus, the fact that most of the pieces of Aranbaltza II are thin would have been crucial in terms of remaining more stable against the action of sand and water flows, and therefore against displacement. On the other hand, the distribution of materials is, in general terms, quite homogeneous throughout Aranbaltza II. The maximum length analyses do not evince any significant accumulation pattern. In fact, both Getis-Ord Gi* and Anselin Local Moran’s I show the existence of small clusters composed of just a few pieces, but that these are not relevant in terms of definition of accumulation processes. This would indicate that there has not been a selection by size, but that there is a great variety and no clustering of high or low values, not even in the area of maximum accumulation, where there is no evidence of selection by length or shape.

Regarding the orientation patterns, preliminary results obtained for AC3 (n = 35) show that there is no preferred orientation pattern, either in axial or azimuthal data. However, as previously stated, the small sample size does not allow any solid interpretation. This is the only identified cluster with a sample size close to Rest (n = 49) but is insufficient to accept or discard any firm conclusion about the orientation pattern of this group of materials. We observed that in an area of under 1 m2 (0.76 m2), there is great variability in the slope of the pieces (many of them are even in vertical position), although they are not located in a zone with variations in the slope of the deposit (Fig. 14). However, the results of Rest (n = 49) and All (n = 99) are very similar, since the axial data show a definite randomness and the azimuthal ones indicate a preferred orientation. This disparity in the results obtained may be because Rest and All encompass more pieces, providing a larger statistical sample than AC3, represented only by 35 measurements, which is below the minimum accepted in other studies (Ringrose and Benn 1997; Benn and Ringrose 2001; McPherron 2018). The application of Getis-Ord Gi* to the elevation values (Z) of the coordinated materials of Aranbaltza II offer us a 2D perspective of the lowest and highest zones from the deposit where they lie (Fig. 14). This projection enables us to observe that the materials in the south are lower than those in the north. However, it is not possible to identify where the substrate is more regular, at least in such a small excavated area. The analysis of Rest indicates that the fabric is mainly planar, suggesting that the materials, irrespective of their orientation, show no great variation in slope and they are mostly flat deposited. The result for All shows that, although it is principally planar, it has a more isotropic trend than Rest. This result would be influenced by the fact that this cluster also contains the materials of AC3, which represents a fairly high percentage of the cluster and displays a strong isotropic trend.

Fig. 14
figure 14

Vertical projection of all materials (A), Getis-Ord Gi* analysis according to the altitude value (B), and location of the AC3 cluster (C) where it is possible to observe the variability in the slopes. Note: These projections are approximate, since they have been produced using the PR data

Full size image

As for the directional distributions, the ellipse for AC3 runs NW–SE, an orientation trend completely different from the other clusters. This may be due to two factors. First, the morphology of the identified cluster is not homogeneous. The shape displays two elongated parts (Fig. 12), especially in the northern zone, which could determine the directionality of the ellipse. And second, the shape of the ellipse could be an artifact of the excavation, since it is just on the edge of the trench and points in a similar direction. This same directionality is the one we find for All and Rest, so both ellipses could have been determined by the excavated area. In the case of AC1 and AC2, both exhibit similar ellipses with NE-SW trend. That the ellipses have a different directionality from other groups could be due to the groups of selected materials being more homogeneous and that these really show the direction of entry of materials. However, the azimuthal direction for AC1 points to a NW trend, while for AC2, there are no data to refute this. Although the morphologies are not the same as in AC3, it is possible that the origin of the materials is similar. The processes responsible for the accumulation of pieces at Aranbaltza II could generate displacement of the materials from the west to east; this hypothesis seems to be corroborated by the entry morphologies of the materials that make up AC3.

It is also important to take into account, although unfortunately, this is impossible to quantify, that the activity of the quarry trucks could have caused some deformation in the deposit. So, there is a chance that the slope that we currently find in the level US4b has been modified afterward, due to the weight of the trucks that passed just along the western side of the excavated area. This possibility is supported by the presence of micro-faults generated after the site formed, which seem to be due to the weight of these trucks. Apart from this, Aranbaltza II could have been affected by two directions of material input flows. Considering the morphologies described and the NW–SE slope of the deposit (the same orientation as the Urgozo valley that holds the site), materials could have flowed in from several fronts. The axial orientation of the pieces seems to be parallel or nearly parallel to the direction of the valley, which would correlate with the patterns observed in the pieces from the flooding events already mentioned. The azimuthal one shows a perpendicular trend to the Urgozo riverbed, which coincides with the processes that may have affected the distribution of materials because of rills produced by increases in the river flow and/or precipitation.

Regarding the archaeological characterization, the Aranbaltza II lithic assemblage is clearly dominated by blades, cortical flakes, and flakes (Rios-Garaizar et al. in press). However, in the classified clusters, cortical flakes and flakes predominate as opposed to blades and bladelets, which are also present but in a smaller percentage. The distribution maps by type of support do not indicate any preferred pattern in their accumulation. Thus, the fact that cortical flakes are the most abundant support in all the clusters is mere chance, since the difference between blades and cortical flakes is minimal. In general terms, the size of the pieces is small and they are thin. All the characteristics of this assemblage, besides the results obtained following the spatial analysis, seem to indicate that there was a very intensive flint knapping activity on the site, focused on the preparation of cores to obtain supports for making tools.

Several open-air sites have been located at the Cantabrian region, but due to different issues, mostly related with conservation and the characteristics of the zone in question (Rios-Garaizar 2015), there is not enough consistent information to draw comparisons with Aranbaltza II, at least in terms of patterning and occupation dynamics. However, the distributional patterning observed at Aranbaltza II is similar to other Châtelperronian open-air sites, like Vieux Coutets (Creysse, Dordogne, France), Les Bossats at Omersson (Seine-et-Marne, France), Canaule II (Creysse, Dordogne, France), or Le Basté (Saint-Pierre-d’Irube, Pyrénées-Atlantiques) (Bachellerie et al. 2007; Grigoletto et al. 2008; Bachellerie 2011; Bodu et al. 2017), with high and well-delimited concentrations of material. In the case of Aranbaltza II, this site might be a type of settlement very similar to the one observed at Vieux Coutets (Ortega 2017), where well-defined accumulation zones affected by post-depositional processes have been documented. This site allows some isolated accumulations (loci) structuring the space to be observed, which have been interpreted as lithic resource exploitation areas (Grigoletto et al. 2008). Therefore, and considering the pattern observed at Vieux Coutets and other Châtelperronian open-air sites, the excavated area in Aranbaltza II could be a sort of locus, an accumulation resulting from an intense exploitation of lithic resources. Nevertheless, the relatively small and limited area excavated at Aranbaltza II may be giving rise to significant bias in the interpretation of the assemblage and the function of the site, since the analyzed area could be a part of a larger site with different activity areas. The extent of the locus defined at Vieux Coutets (Grigoletto et al. 2008) is comparable to the main accumulation area of Aranbaltza II, since the characteristics in the accumulation are very similar both in extent (the Châtelperronian locus 2 of Vieux Coutets covers around 25 m2) and composition (predominance of cortical pieces, fragments < 2 cm and resharpening fragments) (Grigoletto et al. 2008). The main obstacle at Aranbaltza II is the small size of the site due to its partial destruction by the construction of a sewage trench, which hinders verifying whether the density of materials is distributed by locus, such as at Vieux Coutets, or the extent is greater and the materials are distributed over this area. Another point to highlight is the location of the site in a protected zone and close to water and raw materials resources, very similar to other Châtelperronian open-air sites, like Bossats, Vieux Coutets, Les Tambourets, or Le Basté (Bachellerie et al. 2011; Bodu et al. 2017; Ortega, 2017). Thus, it is possible that, bearing in mind the characteristics of the archaeological assemblage (Rios-Garaizar et al. in press), the density of materials and the results obtained following the spatial study, Aranbaltza II could be a site where domestic activities took place, including carcass processing, something that currently cannot be confirmed due to the deposit acidity and therefore the absence of faunal remains (Rios-Garaizar 2015). Friction with sand also altered the surface of the pieces, creating a bright patina that hampers any possible use-wear analysis. Thus, and taking the other Châtelperronian open-air sites as models, Aranbaltza II could have served as a central site for different resource processing activities, including the configuration of lithic tools, which seems to have been the main (or at least most visible) activity developed at the site. It is important to highlight the fact that these sites were not widely analyzed considering the formation processes and spatial analysis, which makes difficult their comparison with Aranbaltza II in terms of formation processes and therefore the impact and intensity of the human activities. The lithic tools produced here were probably exported, not only to activities areas located close by, but also to spots further away. On this point, it is important to highlight the existence of different sites, like Labeko Koba (Gipuzkoa) and Cueva Morín (Cantabria), that contain Châtelperronian points also made in Flysch flint, the same that was exploited in Aranbaltza II (Tarriño 2006; personal observation). Aranbaltza II seems to be an occupation site where intense activities were undertaken, but that was affected by natural processes related to low-energy flooding events once it was abandoned. Indeed, apart from the area where the lenticular bodies have been identified and AC3 is located, the material distribution is fairly uniform and with a slope that seems to be influenced by the slope of the deposits where the materials came to rest, as at Vieux Coutets (Ortega 2017). Therefore, this work has allowed us to discern the possible distribution of materials with a certain degree of alteration in the area where Rest and the clusters AC1, AC2, and AC4 are located, while for AC3 it is possible to identify with precision and detail the processes that could have taken place there.

Conclusions

The Aranbaltza II site displays a fairly uniform distribution of material, characterized by well-delimited lenticular bodies that contain large concentrations of pieces. The spatial analysis has revealed a predominance of isotropy in the fabrics, as well as a prevalent randomness in the orientation and slope patterns of the archaeological materials classified in the clusters. However, this could be due to the small sample size handled, since the results obtained using more representative sample sizes considering the entire assemblage (Rest and All groups) indicate a certain preferred orientation. No selection by size or shape has been observed, and the pieces show hardly any alteration. The accumulation pattern observed at Aranbaltza II is very similar to other Châtelperronian open-air sites, also located in valley settings. However, the absence of spatial analyses similar to the one set out here does not allow us to quantify and compare this similarity. The scale factor of Aranbaltza II is defined by the small area analyzed, which constrains interpretation of whether the accumulation of materials at this point of the site is mainly due to anthropic factors. The processes for these concentrations can be interpreted as low-energy alluvial channels or rills with poor capacity for the selecting, sorting, and abrading of the materials. The good state of preservation of the pieces could be the outcome of short-distance transport and context of fine sands that would have protected them, but without enough abrasion capacity to alter the lithic assemblage severely. The short-distance transport hypothesis is supported, not only by the fresh aspect of the edges, but also because there is no selection of materials by size.