Oil spill forensics - Identification of possible sources for oil spills found along the coastline of mid-Norway - An experimental study combining GC-MS analysis and multivariate statistics.

Abstract

This thesis aims to characterize 112 weathered oil samples collected on shore, at 18 islands, along the coastline of Mid-Norway during a time period of 2011-2015. Emphasis have been made on characterizing samples by three different multivariate methods; Principal component analysis (PCA), Partial least square-discriminant analysis (PLS-DA) and Hierarchal cluster analysis (HCA), however univariate methods have been applied as a starting point.

In multivariate data analysis, diagnostic ratios were calculated between biomarkers and PAH components and applied to look for interesting structures in the plot. The classification from univariate methods, were combined to identify the position of different oil types in the plots.

PCA, PLS-DA and HCA demonstrated their ability to categorize weathered samples, and identified samples that could not be identified by the traditional univariate method. The multivariate techniques were able to classify samples without some of the typical identifying biomarkers that are used in univariate oil spill forensics and indicates that multivariate techniques could be a promising method for identifying heavily weathered samples that often have inconclusive or missing measurements for typically used biomarkers and diagnostic ratios.

Selected samples were imported into an international oil spill database to identify matches to external samples from other projects and laboratories. Six samples in this study were a probable match to oil samples collected at the Shetland islands.