| Summary: | This thesis concerns over the various aspects of environmental forensics investigation of the oil spilled. The uncontrolled contamination either intentional or accidental as resulted from the industrialization (including improper handling and maintenance) and energy demand in Malaysia may cause the chronic spillages of oil into navigational waterways, soils, rivers, seas and drains. The oil spillages has the potential of a long¬term impact to the environment, including the aquatic life especially for fishes. The objectives of this study are to come up with the environmental forensics, investigate both acute accidental of oil spilled and intentional discharge of oil spilled from the operational discharges from the historical perspective and management. Two analytical chemistry approaches, namely gas chromatography flame ionization detector (GC-FID) and gas chromatography mass spectrometry (GC-MS) were used to characterize, identity and quantify the oil hydrocarbon products spillage within the environment. The statistical analyses (conventional Chemomeaic and advaneed Chemometric) were applied to validate the results obtained from the analytical analyses. The primary data of oil spilled was collected from the 47 sampling points in year 2016, and the secondary data was obtained from DoE covering the period of 2013 to 2014. This enables the development of accurate result interpretation of the hydrocarbon oil compounds from the oil spill fingerprinting data according to brands and origin (include the frequency, magnitude and consequences). The analytical analyses integrated with several Chemometrics techniques was important for establishing oil spill database in the future. Significantly, the pattern-matching was also developed based on the secondary data to identity the brand of oil spilled and the specific oil source. The application of conventional (CA, DA and PCA) and advanced Chemometrics techniques (MLP-ANN, Kernel-RBF-SVM and DMAlC Six Sigma) had provided a complete and accurate evidence-gathering techniques in interpreting the data analysis of the complex mixtures of oil spilled. Finally, the development of MLP-ANN, Kernel-RBF-SVM prediction models were carried out based on the results obtained by the analytical analyses. Hierarchical agglomerative cluster analysis (HACA), discriminant analysis (DA), principal component analysis (PC A), factor analysis (FA), MLP-ANN, DMAIC Six Sigma and Kernel-RBF-SVM were applied to validate the oil types from the analytical analyses based on the physico-chemical characteristics. The results showed that for the primary data there was three oi I type clusiers were formed based on hydrocarbon-oil parameter analysis using HACA. These clusters were designated as diesel, lube oil blended with diesel-based products and mixture of many kinds of oils, respectively. The most significant oil spilled variables that contribute to the clustering of oil classification were also re-confirmed and validated using DA. Diesel, lube oil, bunker C, waste oil (WO) and mixture oil (MO) were successfully discriminated using DA based on hydrocarbon concentrations. PCA was used to ascertain and qualitatively describes the oil spilled contribute to the environmental pollution. Five principal components (PCs) were obtained with 93.09% of the total variance for the oil type classification. Th contaminant sources of oil spilled for VFI were TSH, TPH and US EPA Priorit PAHs origins (intentional discharge of diesel based product and marine-engineproduct based like lubricant oil into the seawater). For the YF2, the alkylated PAH pollutants from the illegal or unintentional discharge of waste oil and petroleum product related from the boating activities was identified as the main source of pollution. For secondary data, HACA has formed four clusters. DA has yielded four significant oil types from the oil spilled data, namely, diesel, MOLFO, HFO and WOo Four principal components (PCs) were obtained with 85.34 % of total variance. Based on the cross-validation techniques onto primary and secondary oil spilled data by MLP-ANN analysis, the predicted clustered of oil spilled variables, input-output relationships was established. Five best networks were retained for both primary and secondary data, however, only one architecture performed 100% classification performance for the primary data, while four architectures had achieved of 81.25% classification performance for the secondary data. Another cross-validation technique was, a Kemel-RBF-SYM application was based on invisible-classifier functions developed for predicting the accurate multi-classification of the oil spilled via the hyperplane constructed. Through the analyses, only two oil types were accurately classified for the primary data, namely diesel and mixture oil (MO). While, 100% accuracy achievable for classification performance, namely diesel, mixture oil lubricant and fuel oil (MOLFO), hydrocarbon fuel oil (I-1FO) and waste oil (WO). From this study, we can conclude that the integration of the analytical analyses and various Chemometrics methods on the oil spill fingerprinting dataset, it was able to reveal the meaningful information on a complete and accurate evidence-gathering techniques for database establishment. Besides, this study was able to show the importance of an accurate result interpretation of the hydrocarbon oil compounds from the oil spill fingerprinting data by brands and origin. The application of the several Chemometrics techniques has offered great advantages to validate the results obtained from GC-FID and GC-MS. The MLP-ANN, Kemel-RBF-SYM and DMAIC Six Sigma models have yielded the useful models that can be employed as decision tools for Government of Malaysia in the event of oil spills, in terms of frequency, magnitudes and consequences for more stringent policies and oil spill closed¬surveillance programs to put in place.
|
|---|