Identification of Sources for Illegal Oil Spills by Using GC-MS (Gas Chromatography and Mass-Spectrometry) Databases and Multivariate Statistics
MetadataShow full item record
- Institutt for kjemi 
Preparing a defensible oil spill fingerprinting is always challenging. Presently available well established method for this purpose is a univariate method by comparison of diagnostic ratio using repeatability limit as suggested by European Committee for Standardization (CEN)-method. The consistency result of this method, however, tends to depend on the skill of the analyst who performs the analysis. The shortcoming of CEN-method was then demonstrated by the application of the method to the "MS Server" and "MV Full City" oil spill cases. The author s analysis result exposed some discrepancieswhen was compared with the one performed by SINTEF. Therefore, this thesis focuses on the effort of pursuing alternative or at least complimentary methods to ease the shortcoming of the CEN-method. The main investigation was then emphasized on the possibility of employing the multivariate analyses, i.e. principal component analysis (PCA), cluster analysis, and partial least square discriminant analysis (PLS-DA). The performance of those multivariate analyses were examined by applying to the case studies of "MS Server" and "MV Full City" oil spills. Later on, it was found that PCA failed to classify the samples properly according to the match or non-match with the reference samples. The power of the PCA was revealed when the method was combined with thecluster analysis. The PCA combined with the cluster analysis demonstrated to be faster and undoubtedly more objective (in term of the analyst skill and expertise) as compared to the CEN-method. PLS-DA also showed the same benefits. Moreover, the PLS-DA gives more similar result to the CEN-method applied by SINTEF (irrespective of the difference gaps of the analysts skill, i.e. the SINTEF researchers v.s. the author) as compared to the PCA combined with the cluster analysis. However, the main drawback of the PLS-DA is therequirement of quite large number of sample to obtain a good result. At last, we could see that there is possibility of applying several multivariate analyses, i.e PCA combined with the cluster analysis, and PLS-DA, for the alternatives or complementary of a well established univariate analysis of oil spill fingerprinting (CEN-method). In order to develop the alternative method(s) properly, further research is needed, especially the one which employs more controllable samples.