Stable isotope analysis of human tissues has increasingly been used in forensic studies as a tool to predict the geographical origin of unknown human remains, and to assist with solving missing persons cases. The analysis of stable oxygen isotopes in human tissues such as tooth enamel, hair, and nails, provides insight into the geographical history of an individual as the values reflect those of local drinking water, which, in turn, are influenced by precipitation water values that vary geographically in a predictable pattern. Individuals can then be placed on a geographical map. Several linear regression models exist in the literature for converting stable isotope values of human tissues to drinking water values, however, with noticeable variations in their slopes and intercepts due to methodological differences. For example, some models were based on the isotope values of tap water and others on estimated precipitation water. Large offsets in the predicted drinking water values can be observed between the models, making it difficult to gain insight into the geographical origin of unknown individuals. There is a need to validate existing models to determine their predictability against known human samples. In this dissertation, human tissue samples with carefully collected biogeographical information were used to validate the existing linear regression models. Large offsets were found between predicted and actual values for human enamel oxygen, demonstrating the inadequacy of using linear regression models for predicting geographical origin. An alternative approach was then tested by taking a human sample-based approach, where individuals were geographically classified using human tissue values. Classification models were built directly from human tissue values and evaluated against the collected biogeographical information. Results demonstrated that the geographical assignment of individuals can be achieved by models such as the classification tree models. Although the human sample-based approach requires significant amount of time and resources for the collection of known human samples, it is the preferable approach as it will eliminate any errors associated with the use of linear regression models. The classification method will allow individuals to be identified as local or non-local, which can ultimately help narrow down missing persons searches.
Copyright is held by the author(s).
This thesis may be printed or downloaded for non-commercial research and scholarly purposes.
Supervisor or Senior Supervisor
Thesis advisor: Bell, Lynne
Member of collection