GIS and Scale Issues
The issue of 'scale' has become a popular topic in a variety of archaeological domains. This is the result of an increasing awareness of the need to explore human (and landscape) phenomena at a range of explanatory levels from the effects of individual human agency to the operation of large cultural, imperial and/or 'world' systems, and in temporal terms, from contingent event-based history to long-term processes. Integrating varied perspectives of this kind is a chief goal of KIP, and one to which GIS can make a particularly useful contribution, especially with regard to spatial scale. Below we offer a brief discussion of two areas in which we feel insufficient attention has hitherto been given to the problems and potential of spatial scale issues: i) collection scales and resolution in intensive survey, and ii) multi-scalar statistical analysis.
Collection Scales in Intensive Survey
For survey practitioners, the implications of scale are unavoidable. They impact on the collection procedures we use in the field and the degree to which we can recognise coherent patterns in the spatial distributions (e.g. artefacts or sites) we discover. For example, Mediterranean intensive surveys often use line-walking methods (e.g. KIP's Archaeological Survey section) to record surface artefacts, but usually then store, display and analyse the results in the form of larger ‘tract’ units. However, both line-walking and aggregate tract units average their results over distances or areas of varying size and this 'smearing' effect can make statistical analysis problematic. Summarising by zones of variable size raises a series of methodological problems that are by now well-known in sister disciplines such as quantitative geography (the 'modifiable areal unit' problem, e.g. with regard to district-based summaries of census data), but have hitherto gone almost unrecognised in field survey. For tracts, the problems are more acute because areal summarising/averaging occurs twice (first for the distance walked by each surveyor, and second for the whole tract area).
While we would emphasise the continuing reliability of tract-walking as a pragmatic and meaningful way of sub-dividing complex landscape mosaics, more attention should be paid to understanding the impact of collection units (big and small tracts, collection squares, vacuum circles) and observation scales (e.g. time spent looking at the ground surface) on survey results . For example, comparative analysis of resurvey strategies at Kastri (that was both tract-walked and gridded over ca. 11.5ha), Site 057 and Site 096 (that were tract-walked more than once at different walker spacings) shows that variation in recording unit size, sample intensity or observation time do indeed introduce under-appreciated biases into the analysis and visualisation of such things as pottery densities or ground surface visibility estimates (Bevan et al. in prep). Scale-dependent effects of this kind raise problems of interpretation even within a single survey, but become particularly intractable when we attempt to compare different field surveys with varied project designs.
Such challenges can certainly be overcome: one solution for future surveys is to limit the actual range of tract sizes from the outset. Another, more relevant to existing surveys, is to exploit more consistently the underlying 'walker-scale' at which Mediterranean line-survey techniques actually operate in the field. For example, KIP has sought to preserve as much walker-scale resolution as possible. Artefact counts were recorded separately by each surveyor along with the distances they walked. For each tract, the survey team’s bearing and order (e.g. team members identified from left to right) was recorded and a routine was developed to automate plotting each walker line within the digitised tract polygons.
We can use this finer resolution data not only to interrogate intra-tract variability in useful ways (e.g. Bevan and Conolly 2004), but also to create more detailed and intuitive plots (above left) of surface artefact scatters than is possible with basic tract density maps. Such continuous artefact surfaces need to be modelled carefully (i.e. interpolation parameters are very important), but allow us to visualise survey data at the scale of the 'analytical individual' (the walker count) rather than in aggregated units.
Multi-scalar statistical analysis
Because social and environmental processes occur in the landscape at a variety of spatial and temporal scales, we also need statistical tools that allow us to recognise multi-scalar patterning. For example, commonly 'understood' topographic variables such as terrain slope or aspect can vary quite dramatically depending on the scale at which they are observed (Wood 1996). Hence we need to develop methods that test and characterise spatial distributions at different scales. A classic case is terrain 'relief' ('roughness' or texture). The same portion of landscape might be characterised as a small concave channel at one measurement scale, but be part of a larger convex ridge at another. In fact, we can map and exploit this multi-scale variation to our advantage, for example, to correlate with the location of known sites (Bevan and Conolly 2004).
Likewise, KIP GIS research has explored ways (e.g. edge correction methods for irregular landscape samples, Ripley's K function) to make sense of point patterns (artefact or site distributions) in survey contexts where the sampled areas are often irregular, site dates sometimes imprecise and the underlying landscape processes complex (Bevan and Conolly in press). Consider the figure to the right: the left panel shows a hypothetical distribution of 56 sites. Ordinary point pattern analysis (e.g. nearest neighbour) would detect the presence of site clustering and could perhaps suggest an optimum number of 8 clusters of sites, but would be unable to identify the fact that there is also a higher-order scale producing 3 clusters. Furthermore, if we include the finer artefact-scale resolution represented on the right panel (rather than just an approximation of the centre of the artefact distribution), then clustering can be shown to exist at three different spatial scales: (i) artefacts forming sites (clusters i to x); (ii) sites forming primary clusters (clusters 1 to 8), and (iii) primary clusters forming secondary clusters (clusters A to C).
A good real-world example of the importance of discerning this sort of multi-scale structure is the recent Greek settlement landscape. The typical modern pattern, and Kythera is no exception, is one of buildings clustered into discrete villages. However, beyond these villages, outbuildings and fieldhouses within field systems are often spaced apart from each other fairly regularly. Moreover, villages themselves are sometimes evenly-spaced, a consistent separation which probably reflects the impact of out-colonisation episodes, in-field landholding catchments, refuse disposal zones and local political units. At a larger scale, villages can also cluster within more favoured agricultural areas (such as the inland basins on Kythera). These multi-layered patterns of clustering and regularity have important implications for the overall organisation of the landscape: we can detect them statistically but require explicitly multi-scalar techniques and a willingness to embrace scale-sensitive explanations.