Abstract
A simple diagnostic testing procedure is described that can help non-expert practitioners to search for unmarked graves at historic burial sites using ground-penetrating radar. The methodology is illustrated with data acquired at a historic cemetery in Texas, USA. The acquired radargrams are classified, from which unmarked-grave decisions are made under strict, moderate, and lax thresholds. Since there is no gold standard, or means to check a decision, an alternative strategy for decision-making is adopted based on the following key assumption: the distribution of radar signatures generated by the marked graves at the site is the same as the distribution generated by the unmarked graves. About half of the marked graves generated no discernible radar signature, so this proportion of the unmarked graves is likely to be missed. About one-third of the marked graves generated the tell-tale signature of a deep-seated hyperbola so this proportion of the unmarked graves is likely to be found. The remaining signals are complex and ambiguous. The uncertainty is a result of the wide variety of radar signatures that are expressed by burials and other subsurface objects at the site. The diagnostic testing procedure allows a non-expert practitioner to develop acuity in recognizing unmarked-grave signatures and hone a decision-making capability that leads to improved stakeholder trust. A machine learning algorithm could be developed wherein the training set comprises the radar signatures of the marked graves.