The Management of Space in a Palaeolithic Rock Shelter: Defining Activity Areas by Spatial Analysis. (Method)

Article excerpt


Grotte Mandrin (Malataverne, Drome) is a rock-shelter overhanging the river Rhone in south-eastern France, and is so far the only Mousterian (middle Palaeolithic) site known on the east bank of the Rhone (Giraud et al. 1998). The rock-shelter opens to the north, is roughly triangular in plan and over several years the greater part of it (24 square metres) has now been excavated (Figure 1). These excavations have yielded a rich assemblage of flint objects, and it was quickly noted that they had a patterned distribution at the lowest occupation level (level 1). The 1514 objects plotted at this level were broadly concentrated in the western part of the shelter, with zones of different densities (in numbers per square metre) towards the east (Yvorra & Slimak 2001; Figure 2). Different kinds of object (including blades, flakes and cores) were represented in different quantities, and it was evident that these variables could also have significance for the use of space in the rock shelter. Accordingly it was decided to apply statistical methods to define groups of objects which might represent different activities deployed in different parts of the shelter, and so throw some light on domestic practice in the middle Palaeolithic period.


It was first necessary to establish that level 1 was a primary context without admixtures from above which would invalidate a contemporary spatial pattern. Even if not disturbed by post-depositional activity, occupation layers may erode, concentrating objects of different periods in apparently the same layer. The creation of such a 'taphonomic palimpsest' is always possible in the middle Palaeolithic, but in this case there were reasons for having confidence in Level 1 as representing a single occupation. Distinctions between levels were observable, the lithics are related to the same production processes and the spatial patterning is heterogeneous. Conjoining flint and association with animal bone also suggests that there had been little post-depositional re-ordering.

Defining clusters

The use of sophisticated statistical analysis therefore appeared to be justified and k-means analysis already used in similar studies, seemed to answer our objectives (Simek 1984, 1987; Simek in Rigaud 1988; Vaquero et al. 1998; Vaquero 1999). K-means analysis finds central points (centroids) for groups of objects distributed in two dimensions by calculating the mean distances between them. The number of defined centroids can be increased, so as to resolve the distribution pattern into more spatial units or clusters, the most useful results coming when the variation within each cluster is low. J. Simek has devised a diagram which shows the degree of variation within clusters when the number of clusters defined ranges from two to 15. This may be seen in Figure 3a, where the graph of variance versus numbers of dusters is shown logarithmically for convenience. In our analysis, the optimal numbers of clusters was considered to be those corresponding to breaks seen on the diagram (marked with arrows). The material was accordingly mapped first in three clusters (Figure 3b), then into 10 clusters (Figure 3c) and then into 12 (Figure 3d). These seem to present the best solutions to the partitioning of lithic distributions in level 1. Each cluster is represented by a circle based on the centroid, whose radius is the standard deviation. The density of objects is represented by a grey scale (darkest equals most dense).


The character of the spatial units

The three spatial models shown in Figure 3 were then analysed further to investigate how far each cluster might be characterised by a particular kind of debris, from which particular activities might be inferred. The lithics of the first occupation level were classified according to form (flakes, blades, bladelets, cores, discards, tools and micro flakes of < 15 mm length), their integrity (whole or fragmented), and the presence of cortex. …