Discriminant function analysis of laboratory measurements resulted in an effective method for distinguishing among crude oils of different origin and formation. It will be shown that an established statistical technique on two data sets, infrared absorbance ratios and system chemical compositional data of six Nigerian crudes has provided a useful technique for classification. Multivariate analysis techniques were used to evaluate two categories of data sets generated from multianalytical methods for derivation of compositional analyses and mass fragmentography and infrared spectrophotometric analysis of passive indices inherent in six Nigerian crude oils, tar balls and the five foreign export crudes to the United States.
Oil exploration and production activities are carried out dominantly in the Niger Delta region and the choice of samples are representative of the crude oils produced in the country.
The major export crude oils in Nigeria are Bonny Light, Bonny Medium and Forcados Blend. They are therefore the most pertinent crude oils for spill source identification system since they are the most transported crudes in our regional waters. The other crude samples, Delta Field, Elelenwo and Umuechem Field crude oils are representative crudes from the Western and Eastern sections of the delta. They are complementary samples in this study and their fate are evaluated along with the export crude oils. Tar balls are included to give an indication of the level of weathering obtained in the test facility used here. Tar balls are tarry lumps of crude oil which had caked and hardened by weathering processes of advection, emulsification and sedimentation. Tar balls are the result of extreme weathering conditions over long period of time.
Compound Indices Matrix for Discriminant Analysis
A total of 19 compound indices were selected for further evaluation with the complete set of weathering test data. This data provided the basis for developing the final template function for the six different Nigerian crude oils.
Additional compound indices of five export crudes to the United States were also evaluated. These data provided the basis for comparative analysis of the system application on two different sets of compound based compositional data. It also provided the basis for comparison between the final templates for the six Nigerian crudes and the five export crudes to the United States.
Whole Crude Characterization.
Discriminant Test on Compound Type Indices.
The statgraphics graphic system performed a multiple discriminant test on individual type compound indices. The regression analysis of some of the variables showed that the distribution of the data sets were multivariate normal.
Evaluation of the discriminant functions of the 570 observations, the canonical correlation and the wilks’ lambda statistics are computed for group means of the individual compound indices, the group standard deviations and the classification function coefficients for the group of six groups within groups covariance matrix, within groups correlation matrix for the 19 variable and the derived functions for the standardized and unstandardized function coefficients and group centroids. The preliminary step involved data standardization. Since the data have been standardized, all variables have approximately the same mean and standard deviation. Therefore the magnitude of the coefficients represent a measure of discriminatory power of each index and variable tested.
In the plot of the first two canonical variables, five groups are distinctly classified. Group 5 (Elelenwo) and Group 6 (Umuechem) overlapped, suggesting that both crude samples have common alluvia origin. An interesting observation is the classification of Group 3 (Bonny Light). It is classified on the lower left corner of the plot of canonical variates along with Delta field and Forcados blend crude oils which are other light crude oils. Other combinations of canonical variates could also be plotted.
Successive separation of the different crude oils was however, achieved by rotatory combination of canonical variates for whole crude characterization. The group classification achieved by the discriminant test is novel being the first reported for any set of Nigerian crude oils. It is a very wholesome classification even though the data are statistically divergent.
Discriminant function analysis for the five export crude oils involved evaluation of 650 observations of chemical fingerprint indices. The same analysis steps are involved as has been described earlier. In a plot of the first two canonical variates, all five groups are distinctly classified. Thai shows that the chemical fingerprints are quite discriminatory. Secondly that the discriminant function actually performed the discrimination test on the compound indices and that the crudes are from different origin and formation.
Discriminant Test on Ratios of Infrared Absorbance
Data obtained from infrared spectra analysis were similarly subjected to multiple discriminant test to distinguish amongst the crude oil samples. The day sets were multivariate normal on account of regression analysis of some selected ratios of infrared absorbances.
The preliminary step for the evaluation of the discriminant functions of 360 observations, involved data standardization. Since the data have been standardized, all variables have approximately the same mean and standard deviation. Therefore, the magnitude of the coefficients represent a measure of discriminatory power of each index or variable tested.
In the plot of the first two canonical variates, Group 1 (Delta field crude oil) and Group 2 ( Forcados blend crude oil) are well differentiated from the other four groups. Group 3 (Bonny Light) and Group 4 (Bonny medium) overlapped considerably. Similarly, Group 5 (Elenlenwo) and Group 6 (Umuechem) showed considerable overlapping while Groups 3 and 4 slightly overlap with Groups 5 and 6. Other combinations of the canonical variables showed poorer classification. Ratios of infrared absorbance classification template provide a means for comparing the performance of the discriminatory test on laboratory experimental analytical results.
Plots of other combinations of canonical variates also differentiate between these two groups of crude. These plots are for the 1st and 5th, 2nd and 3rd, 2nd and 5th, 3rd and 4th, 3rd and 5th and the 4th and the 5th. It should be recalled that only a plot of the 1st and second canonical variates discriminated with ratios of infrared absorbance.
Comparative Analysis of System Application.
Whole crude characterization is important in the oil spill identification system, oil source correlation, palaeontological studies and oil pricing. Most identification activities were directed toward components that are intrinsic to the oil. Identification by this means is called “passive tagging”. Passive tagging establishes the broad characteristics of the oil and is often called characterization.
The earliest methods of characterization involved structural identification of the stable carbon to carbon skeleton of the fossil fuel because the organic molecules are highly resistant to weathering and therefore faithfully reflect the composition and distribution of the corresponding compounds in the original oil. X-ray crystallography is used for structural identification of these molecules.
The triterpanes fingerprints find application in the spills identification system. Five different structural forms of triterprenoids which are abundantly distributed in the plant world. The molecules have successfully been used as biological markers and in palaetological correlations by consideration of their stereochemistry, structural configuration and optical activity. For crude oil of vastly different origin and hence different structural forms, they have also been used to clearly distinguish between them.
The limitation of this method include the use of X-ray crystallography for identification of structural forms, difficulties in isolating and presenting the samples in crystalline forms and the subjectivity of visual matching of structural forms when distinguishing among crude oil types. Also, this method does not allow for the inclusion of mathematical treatment and statistical discrimination to ascertain reliability of the identification system. Consequently, with advances in instrumentation, other identification methods evolved. These include infrared spectral analysis, system chemical analysis and correlation of stable isotope ratios for final purposes of characterization and identification.
Infrared spectral analysis has featured prominently in the development of procedures for identification systems. The successful application of infrared techniques has made significant contributions to environmental improvement in establishing responsibility for water quality standards. Infrared spectrophotometry when combined with data treatment, data transformation and discriminant analysis through computer assistance resulted in a more precise and accurate method of distinguishing between crude oils and their products. Infrared spectra in this region give indication of the characterization factors, such as aromaticity and paraffinicity. The ratios of intensity of the infrared absorption at one frequency to that at another frequency are considered to find a most discriminating and characterization property for the crude samples and derivatives.
The shortfall in infrared techniques, however, is that measurement in the near infrared region is difficult and requires high resolution spectrophotometers. On the other hand, infrared spectroscopy yields information on the absolute levels of bulk hydrocarbon at the selected absorbance frequencies and thus limits its use. Another limitation is that since passive tagging establishes broad range of intrinsic properties of the crude sample, infrared spectra do not yield quantitative and measurable properties of the crude and is therefore not wholesome in providing a data base for characterization.
System chemical analysis together with GC/HPLC have been used to develop the discriminate template. Each measured parameter and/or ratio of measured parameters are employed to distinguish among the samples. Statistical discriminate analysis was applied on selected fingerprint indices for discrimination. The classification achieved in this test is distinct. The classification reflects both geological and geochemical properties of the crude samples. All the intrinsic properties of the crude have been utilized in generating discriminant template. Thus, it is not only useful for characterization but also distinctly differentiates between crudes. The dual purpose of characterization and identification have thus been established.
Infrared spectra photometric data utilized here provided a means of comparison between the compositional data set on the one hand and the infrared data set on the other. The plot of canonical variates of the ratio of infrared absorbance also classified the different test samples. The extent of this classification is reduced by the use of infrared absorbances. The less distinctive nature of the classification model could be attributed to the constraint posed by instrumentation since the infrared absorbance at the near infrared region was not measured here in the analytical step. At the near infrared region, the measured absorbances give stronger indication of the characterization factors of the crude samples and if incorporated into the discriminant test would probably enhance the discriminatory capabilities of these fingerprints. However, the same obvious pattern was observed in the classification model as in the compound type module. The crude samples which were classified tended to be distinguished on the basis of their geological and geochemical intrinsic properties. Whereas, the data measures the intensity of the infrared light, the classification pattern is suggestive of a likely correlation of the between infrared light intensity and crude intrinsic properties.
The results thus presented show that certain specific chemical compounds e.g groups or ratios, which are present in crude oil samples are retained even after they have undergone extensive weathering. These chemical fingerprints indices include weight concentration ratios, the high molecular weight paraffins and polynuclear aromatics.
The compound indices selected for discriminant function analysis to distinguish among the six different Nigerian crude oils are intrinsic properties of these crude oils. Therefore, derivation of detailed compositional data of both source and spilled samples is required for characterization and identification. A priori knowledge of the suspect source is thus a prerequisite in the oil spill identification system. The methodology developed in the characterization system should therefore be generally applicable and should always select the best group of tags for characterization and identification.
Infrared spectra in this region give an indication of the characterization factors, such as aromaticity and paraffinicity. Measurements of infrared spectra in this region required high resolution infrared spectrophotometer. The measurement of absorbance in the near infrared would generate more absorbance ratios with strong indication of characterization factors for the discriminant test and would probably enhance the discriminatory capabilities of these fingerprints. Additionally, the composition of individual naphthenes would have probably enhance the quality of the characterization and identification process. This analysis will require the use of a gas chromatography/mass spectrometry which was not used here.
The stepwise variable selection technique was limited by program 64 kbytes RAM space available for data handling. In the multiple regression analysis step, the backward procedure for variable selection involving all the variables could not be executed at the same time because of this limitation. Consequently, the evaluation of component effect of other variables on each selected variable was therefore not achieved.