One of the things I’ve been working on for the last 2 (or 10 or 15) years is how do we understand the distribution of artifacts produced through intensive pedestrian survey. In fact, thinking about the distribution of artifacts and how our methods of survey and analysis shape the kinds of conclusions we can reach has been a major element in my career.
For my current project, the Western Argolid Regional Project (and any of these survey really), there are a few obvious challenges. First, the basic spatial unit is the survey unit, but in our survey (and many surveys) these units are irregularly shaped and sized. Second, the units produce a good many interdependent variables that, in turn, shape artifact recovery. These range from surface visibility and vegetation to less obvious influences on site formation like terracing, ploughing, or various kinds of fills. Third, there are a good many variables that impact the distribution of artifacts that occur outside of what we can understand and record within the units themselves. These includes the presence of ephemeral paths and roads through an area, access to water or other changeable resources, the proximity to the edge of the survey area or highly disturbed units, and many other archaeological and historical features that might impact how we understand an artifact scatter. Finally, we understand that various periods and types of artifacts have different levels of visibility in the landscape. This is the result of different historical processes that produce horizons visible on the surface as well as the character – and visibility – of the artifacts themselves.
In short, there are many variables that shape the distribution of artifacts on the surface and it’s hard to imagine a statistical model that would accommodate all these various.
There are, however, ways to start to smooth the distribution of artifacts and in an article that’s due in October, I’ve proposed the following method, which both attempts to produce understandable clusters of units with artifacts from the same period plotted across the landscape. These clusters are based on two measures of proximity between units with material from the same period: 20 m buffers and near analysis based on the “near” function in our project GIS. More important, however, this work is only the start. Because buffers and the near function do not adapt to changes in the landscape that range from steep slopes, modern roads, the course of the Inachos river, fenced plots, or the edge of our survey area, we need to scrutinize these simple clusters, particularly the discontinuities between clusters, to determine their historical and geographical probability.
Here’s what I wrote:
The following analysis is based on clusters of units from across he wider WARP survey area that produced Late Roman material. We identified groups of units on the basis of analysis done using the projects GIS platform, ESRI’s ArcGIS. We produced aggregated clusters of units by grouping any units that fell within a 20 m buffers of a unit with Late Roman pottery. We then assessed the relative isolation of the clusters using the “near” function in ESRI ArcGIS. Groups of clustered units that were statistically “near” one another could be aggregated further. Finally, we also allowed our familiarity with the topography of survey area to shape how we defined the clusters described below. Buffering, near analysis, and familiarity with the survey area helped to smooth some of the variations in surface visibility, local site formation, and recovery rates. These clusters also produced larger and more complex assemblages of artifacts than would appear in single or adjoining units, and these larger assemblages offered the opportunity for more nuanced reading of the material. These clusters, however, should not be confused with sites and their attendant assumptions regarding function or settlement rank. Instead, the larger assemblages allow us to retain the ambiguity inherent in the functional analysis of surface assemblage, while also constructing arguments for chronological and spatial differentiation at the scale of our survey area.