Ohio State nav bar

Seminar: Nancy McMillan

Photo of Nancy McMillan
January 25, 2018
All Day
209 W. Eighteenth Ave. (EA), Room 170

Title

Using Probabilistic Modeling of GIS Data for Forensic Search Efforts

Speaker

Nancy McMillan, Battelle

Abstract

Examinations of soil traces associated with forensic evidence can be used to narrow potential source area(s) by characterizing features of the trace soil assemblage which limit possible regions of origin.  Comparison of the soil observations to increasingly available digital map data enables the integration of multiple geospatial data sets to aid in forensic geographic attribution.  Properties of a forensic soil sample may be used to assign likelihoods of the soil being derived from distinct areas based on mapped features, on both categorical maps such as a geological feature maps, and numerical data, such as distance from a point source.  Seldom, or perhaps never, do digital map data precisely represent the characteristics observable in a forensic soil trace.  Despite the limitations of digital map data, probabilistic inferences can be made from relative likelihoods of regions as potential source areas. 

In this presentation, the probabilistic model for geographic attribution will be discussed and a tool developed to automate the model will be demonstrated.  The model and tool will be applied to a case from 2003/2004 in which forensic soil samples derived from digging tools were characterized to narrow the search area of a clandestine grave; the suspect traveled approximately 5000 km before arrest, making the geographic attribution derived from soil examination useful for narrowing the possible search areas.  In the issued report on this case, geographic attribution of the soil evidence was performed without the aid of digitals maps.  We use our geographic attribution model to infer the grave location at three spatial scales aimed at demonstrating the usefulness of the probabilistic processes and products to aid a forensic search effort.  The resulting geographic attribution models greatly narrowed the recommended search area, as did the original reports issued in 2004, but as the output is digital, it can be easily over-laid on infrastructure maps to aid search efforts.  The region of the actual grave site was correctly identified in both the original report and in the digital models described in this paper.