4.2.2 External testing

Subsequently, in order to collect new data to test the Pisa coastal plain model, a surface survey was carried out in a small sample area that led to the discovery of new archaeological sites.

The final strategy was to survey a small sample area, corresponding at least to 10% of the total study area, which in terms of sample size was sufficient to give reliable information about the total population (see Fletcher and Lock 2005). The choice of the sample was based on unsurveyed areas and the three different levels of archaeological risk or probability on which the study had been divided. In field walking, the decision was made to collect all the archaeological materials, dating from prehistory to the modern age.

When defining the collection unit, we referred to the Archaeological Topographical Unit as the primary standard employed by Italian superintendence to record archaeological presence. In landscape analysis, this consists of the smallest trace of a past human identifiable presence (Cupitò 2007). Basically, we prefer to consider the concept of topographical unit or 'presence', rather than considering the concept of 'site', since we identified as primary analysis units those locations where we could find traces of human presence, with at least two different artefacts dating to the same time period, using the same criterion in making the archaeological dataset as was used in the predictive model (Mazzanti 1994).

One of the main issues to be addressed in the field walking planning was surface visibility. In order to test the predictive model we looked for arable lands, to have a homogeneous view of the testing sample, in which surface visibility was not affected by a biasing factor such as vegetation cover (Terrenato 2000). For this reason, from the expected 10% initial sample (about 11km2), field walking was reduced to 2km2, corresponding to an approximate 2% sample. As a matter of fact, this sample was the only ploughed portion of land available to make the survey (Figure 5).

Figure 5: Sample areas chosen to carry out an external testing

Figure 5

Despite the small size of the sample, which did not allow us to obtain statistically reliable conclusions, we think that the results obtained are still worthy of attention.

As we have seen, the sample area had been allocated to three fairly equal portions of land corresponding to the three different probability classes, with a slight prevalence of the low risk one, attested around 40%. Indeed, in testing a predictive model's effectiveness, it is just as important to investigate those areas where the model predicts the absence of sites, as a guarantee of impartiality in the final data collecting (Banning 2002).

The survey has thus been carried out in a systematic way, trying to cover all of the land units across the sample area by walking in regular transects, with field walkers some 3-5m from each other. That allowed a nearly total coverage of the land, so as not to miss any possible archaeological finds, however small. Final results show us that, in total, eight sites were identified, with a clear bias towards high risk areas, in which six sites were located. Sites were almost all dated to the Roman period, with only one exception, dated to the late Palaeolithic. Even though we cannot properly evaluate these results under a rigorous statistical profile, they clearly show that the majority of the discovered sites fall into the highest level class of archaeological sensitivity, while only two have been identified in medium/low risk areas.


© Internet Archaeology/Author(s)
University of York legal statements | Terms and Conditions | File last updated: Thu Dec 1 2011