CHAPTER 3. METHODOLOGY
III. Sources and methods
2. Geographic Information Systems
2.2. Visualizing distributions
The techniques in question, to be more specific, are largely concerned with visualization. This begins, prosaically enough, with data entry. The physical landscape within which the MHS survey universe is located is described by a variety of data, the most basic of which concern the appearance of the surface and its topography. Both sorts of information were supplied by a raster layer comprising one tile of Standard, four-band color satellite imagery acquired by a QuickBird satellite and purchased by MHS from DigitalGlobe.385 This raster was joined to a Digital Elevation Model (DEM), which contains information concerning the elevation and slope of the region.386 The effect is to build up a digital landscape whose characteristics reflect, and for purposes of analysis serve to proxy, those of the 382. E.g. De Silva & Pizziolo 2005; Harris 2000; Llobera 2000; Madry & Rakos 1996; Vermeulenet al.
2005.
383. Bellavia's "natural pathways" algorithm, and the developments it has inspired, may represent a recent exception: Bellavia 2006. On the possible utility of this and similarly innovative tools in a later phase of this research, see Chapter 6, Section V.3.
384. van Leusen 2002b: 5.5. Llobera's (2000: 72-73) attempt to model the influence of cultural features on patterns of movement is good example. He postulates, as a thought experiment, the presence in a landscape of a tomb to which travelers will be attracted; perhaps, we may guess, because visiting is good luck. Specifically, he assumes that the presence of the tomb imposes a "cost"c(di) on movement, at all places i from which the feature may be seen, which decreases linearly with proximity to the tomb:
where "magnitude" is the rank of the tomb, in respect of its importance as a factor influencing movement, vis-à-vis other variables affecting cost. The author also provides equations for cost influence modeled as decaying at an exponential rate, "when the effect of the monument is is felt only at a very close range", and as a stepwise function where "the full impact of the feature (i.e. all its magnitude) is felt within the immediate surroundings of the feature": Llobera 2000: 74–75.
385. Standard imagery products are provided with embedded geographic positional data—as reckoned, in the case of the image utilized by MHS, in the UTM coordinate system on the basis of the WGS1984 datum and spheroid—which obviates the need to manually locate the image in Cartesian space.
386. As acquired free at http://www.sinanet.isprambiente.it/it/sia-ispra/download-mais/dem20/view
physical space of which it is a model. It is within this space that information generated by the survey is located.
Such information includes, first of all, the areas which have in fact been surveyed. DUs and GCUs are represented by several layers of polygon Shapefiles, the definition of which begins in the field. A hand-held GPS unit, accurate to ±4 m and set to the coordinate system utilized by the project GIS, was used to record a series of at least five387 points corresponding, in order, to the NE, SE, SW, and NW corners of the unit. The fifth corresponds to the centroid, or presumed center of the survey unit. These five points were entered in the project GIS as a point Shapefile, and, on the basis of comparison with a sketch made in the field, the unit borders were drawn. (The purpose of the sketch, which was not made to scale, was to allow for correction of the putative unit borders on the basis of comparison with the landscape visible in the satellite image.388) These borders define the polygons referred to above, which are grouped as layers corresponding to each of several areas surveyed. These are eight in number. Units in the areas of Borso, Genna, and Marcanzotta were surveyed during the 2008 field season, and those at Timpone Granatello, Casa Abbadessa, and Piscitello were walked in 2009. San Leonardo was a target of research during both seasons. Information concerning the artifacts collected in each survey unit is then linked to the relevant polygon.
As explained in Section III.1.2.1, I view these remains as proxies for various sorts of historical presence and behavior. Their distribution is related, albeit incompletely, to aspects of human dwelling in the landscape. Mapping the density of e.g. finewares versus cooking wares, or Early Imperial ceramics against those of a Late Roman date, may suggest the existence of patterns which can subsequently be interrogated in light of a variety of characteristics. The process of doing so is as follows. Sherds counts for each of my five functional classes, which are recorded at the level of the DU or GCU, are standardized as a measure per Hectare.389 The densities which result are useful for a variety of purposes. First, they allow for the comparison of results between survey areas—and, by more or less imperfect proxy, the landscape in which they are situated.390And second, they provide a window into
387. The delineation of DUs coincident with modern agricultural fields, as per MHS project methodology, resulted in the definition of several with irregular polygonal outlines. In such cases, points were taken at every angle.
388. This practice allowed for the correction of a peculiar sort of research bias identified by van Leusen 2002b: 4.11-4.12. He explains: "the raw counts and weights per collection unit resulting from a survey will be ‘normalised’ to account for any differences in the size and coverage rate of the collection units. It was found during analysis of the Ostuni survey data…that the digitised areas of many collection units were approximately 10% smaller than those mapped in the field on topographic maps at scale 1:10,000", which discrepancy was in turn reflected in the standardized measures of density.
389. A procedure which, it is worth noting, is not without its problems, notwithstanding its necessity. The reason is to be found in the so-called "modifiable areal unit problem (MAUP)", which states that patterns observed in spatially aggregated data are influenced by the process and scale of aggregation: Amrhein 1995; Frotheringham & Wong 1991; Openshaw 1984. Bevan and Connolly's (2009: 957) solution—to work exclusively from walker transects of similar size—is not feasible in the present case, as only in some instances were MHS counts kept separate at the level of the tract. Correcting for the bias introduced on this account is, without a doubt, a desideratum, but not one I effect at present.
390. For a discussion of the ways in which MHS survey areas relate to the region of the project's activity, see the discussion of individual areas in Chapter 4.
the patterning of finds across a given area. To the degree that any trends can be discerned therein—are the finds especially numerous? Are they dispersed, or do they cluster in an appreciable way?—this is noted, both with respect to group of all Roman sherds and, subsequently, the remains belonging to each of the classes described above.391Where discrepancies exist between the number or distribution of these groups—either between the collection of all sherds and some subset thereof, or between different functional groups—these are signalled.
As will be clear in the following chapter, GIS is a significant aid to visualizing these sorts of relationships. It is also, however, useful in contextualizing them. The ready availability of topographic information, as contained in the project DEM, makes it easy to determine whether the patterns observed in the material can be mapped onto variation in the natural environment, or, conversely, other explanations need be marshalled.