Forensic geolocation of Norwegian soil samples using random forest analysis of microbiomes

Abstract

Soil is a substrate easily picked up during everyday activities due to its ubiquity and adhesiveness. A method to determine the origin of soil precisely would therefore be useful in forensic casework as it can provide insight into a person`s or object`s previous whereabouts. To evaluate the potential of microbiomes for soil provenance determination, we collected soil samples from 15 locations in and around Oslo, Norway. Samples were collected multiple times from the same locations to assess changes in the microbiome over time. Additionally, a mock soil stain on clothing sample was collected at each site to validate this sample type. The microbial composition of samples was determined via amplicon sequencing of the V4 region of the 16S rRNA bacterial gene. Our results showed that the microbiomes were significantly impacted by location, with both soil and soil stain samples from the same site mainly exhibiting greater similarity than those from different sites. Notably, seasonal variations affected microbiome composition, leading to significant changes in some locations. Machine learning was employed to associate samples with their geographic origins, achieving classification accuracies above 85 % for both soil stain samples and soil samples collected within one week of each other. However, accuracies were lower for samples collected across different seasons, between 55 % and 64 %, indicating that temporal variation can limit the reliability of soil microbiome analysis when there is a delay between sample collection times.