The capability to estimate the geographical origin of soil evidence can provide valuable forensic information to assist investigators and (re)direct police resources by triaging areas of low and high interest. Historically, this provenancing capability has relied heavily on the local knowledge and opinions of experts, geologists, and soil scientists, often with limited scientific objectivity underpinning their source predictions. While this has led to valuable information in some cases, the reliance on subjective approaches raises issues of reliability, foundational validity, determination of uncertainties, and risks due to cognitive bias. Alternatively, fit-for-purpose regional, national, and continental scale soil surveys could support quantitative and probabilistic methods to objectively evaluate the provenance of soil evidence. However, the lack of established best practices and standardised methods creates challenges for the implementation of this approach. To work towards fulfilling this knowledge gap, this body of research aims to provide a systematic investigation and development of empirical provenancing models from a forensic science perspective. To provide a suitable reference dataset, a high-density geochemical topsoil survey (1 site per 1 km2; 0 – 5 cm sampling depth) was carried out over Canberra, Australia. A total of 823 samples were collected from 686 sites, covering 660 km2 of urban, suburban, and semi-rural areas. Following dry sieving, the <75 μm fraction, chosen for its capacity to transfer and persist on individuals and objects, was analysed using X-ray Fluorescence (XRF) and Inductively Coupled Plasma – Mass Spectrometry (ICP-MS) to determine total element content, and by X-ray Powder Diffraction (XRPD) to determine mineralogy. Thirteen “blind” simulated evidentiary samples were also similarly collected, prepared, and analysed for use in the validation of developed provenance models. Two different and complementary approaches were tested to determine, or at least reduce, the provenance area of an evidentiary sample based on a spatially continuous (interpolated) map. For the first “investigative” approach, analytical similarities between measured values of the evidentiary sample and the respective values of each cell/pixel in a reference map was attributed from the overlap between their respective Cauchy probability distributions. This generated a map over the survey area with provenance “match probabilities” ranging from 1x10-6 to 1.00, inferring areas as low and high interest respectively. This approach was further complemented by a successive “evaluative” Bayesian approach, where the analytical similarity of each cell/pixel was weighed against the probability of observing the cell/pixel value in the total dataset using a Cauchy kernel density-based likelihood ratio approach. Utilising these approaches to assign provenance similarity and likelihood, an evaluation of mapping techniques and selection of elements was performed to obtain the most conservative and reliable method to transform geochemical point data to a spatially continuous map. Approaches including a traditional Natural Neighbour interpolation on a coarse 500 m x 500 m grid, a method to upscale from a fine 25 m x 25 m Natural Neighbour interpolation to a coarse 500 m x 500 m grid, and a grid-agnostic alternative based on Voronoi polygons were compared for their ability to account for errors and objectively generate reproducible predictions, as assessed using the thirteen “blind” samples. The latter method based on Voronoi polygons was the most objective, reproducible, and least susceptible to cognitive bias and arbitrary operator decisions. While also being more efficient to compute, this approach regularly yielded at least equal, or often more accurate provenance predictions than that of the grid-based methods. Based on the probability of observing discrete population percentiles in the total dataset, a spatial specificity concept was developed to estimate the effectiveness of individual elements/variables as provenance indicators prior to considering evidentiary material. Of the 55 elements measured (XRF and ICP-MS), eight major oxides (XRF) and seven trace elements (ICP-MS) were found to be the most useful at predicting provenance as they tended to have high spatial variability with low measurement uncertainty. The inclusion of additional low-correlated discriminatory elements, and mineralogy (XRPD) provided complementary information, that further identified areas as low interest that otherwise may not be differentiated using one technique alone. Realistic casework samples are often influenced by transfer, persistence, and recovery dynamics that may alter the natural composition of soil evidence from that of the source; thus, leading to risks of wrongful exclusions and misleading provenance outcomes. Research was undertaken to identify the scientific principles, material properties, and contact interactions that influence the continuity of the source signal during soil transfer, persistence, and recovery events. This led to the recognition of materials science and tribology as being among the foundational scientific disciplines underpinning trace evidence. A materialagnostic framework and “way of thinking” was proposed for the interpretation of trace evidence and/or simulation of activities involving trace evidence, including soil materials. While provenancing workflows can achieve accurate predictions and correct area exclusions, the research emphasised that empirical models are likely to wrongfully exclude geographical areas that were poorly represented by the reference data. Equally misleading predictions may occur when the questioned sample (e.g., deep subsoil) does not reflect the sample medium (e.g., surface topsoil) included in the selected reference data. While this research has focused on provenancing soil material, the scientific basis and reasoning guiding the development of the proposed provenancing models are, for the most part, fundamentally material-agnostic. Thus, this research and established models are likely broadly applicable, with minor adaptation, for different types of forensic provenancing, such as animal, human, plant, and food materials.
| Date of Award | 2023 |
|---|
| Original language | English |
|---|
| Supervisor | Jurian HOOGEWERFF (Supervisor), James Robertson (Supervisor) & Ken Mcqueen (Supervisor) |
|---|