MATERIALS & METHODS
Gas Chromatography

Hydrocarbon samples were analyzed on a Hewlett Packard 5890 instrument equipped with a split/splitless injector, J&W 30 meter DB-1 column and an FID detector: All gas flow rates were set to manufacturer specifications. Injections were made in split mode with a split ratio of 1:100. The column oven was programmed from -10 ^o to 350 ^o C at 10 ^o C/minute with 4 minute hold at 350^o C. The injector temperature is set at 350^o C and the detector temperature is 360^o C. Data was acquired and processed with an EZChrom Chromatography data system.

Gasoline Weathering Simulation

One of the weathering processes that can affect released gasoline is evaporation. To simulate evaporation under controlled conditions, three different grades of gasoline were obtained from a local retailer and allowed to evaporate to controlled volumes.

The gasoline components with the lowest boiling points tend to volatilize more rapidly than the components with higher boiling points. On the GC fingerprint the components that have the shortest retention times (left side of the GC fingerprint) are the most volatile and will tend to decrease in peak intensity preferentially as more volatilization takes place. This is clearly illustrated in Figure 6, where GC fingerprints from the same gasoline are shown under differing levels of volatilization.

Evaporation Procedures

The evaporation procedure that was used is described below:

• Samples of three grades of gasoline (87, 89, and 93 octane) were acquired at a local service station.
• Equal volumes of each grade of gasoline were placed in four (4) 40 milliliter vials, making a total of 12 vials.
• Three (3) vials of each sample were uncovered (total of 9) and allowed to volatilize at room temperature.
• The uncovered vials were closely monitored and capped when the gasoline was reduced to the desired volume resulting in one vial each at 75%, 50%, and 25% of original volume for each of the three grades of gasoline.

Algorithms Described
Correlation Coefficient

The correlation coefficient, denoted by r, measures the relationship between two data sets that are scaled to be independent of the unit of measure and is given by the formula:

Where x and y are values in each corresponding data set.

The value of the correlation coefficient is always between -1 and +1. A value of r equal to -1 indicates a perfect linear relationship between the sample values of x and y, with the value of y decreasing as the value of x increases. A value of r equal to +1 also indicates a perfect linear relationship between the sample values, but one in which the value of y increases as x increases. Larger values of y are associated with larger values of x; and smaller values of y are associated with smaller values of x. If there is no linear relationship between the sample values of x and y, then r will have a value near or equal to zero (Hayslett).

The correlation coefficient determines whether two data sets move together; that is, whether large values of one set are associated with large values of the other (positive correlation), whether small values of one set are associated with large values of the other (negative correlation), or whether the values in the two sets are unrelated.

In this study, 71 hydrocarbon chromatogram peaks, each representing a different hydrocarbon compound, were used in the analysis. Table 1, presents a list of the compounds. Integrated peak areas were measured and then tabulated for each of the five hydrocarbon samples in figures 1 through 5. The integrated peak is a measure of the intensity of the response of the flame ionization detector to each of the individual compounds measured in millivolt seconds. The library samples includes gasoline, kerosene, diesel, JP-8 jet fuel, and crude oil. Once the peak area data were collected and tabulated for these hydrocarbon samples, three additional hydrocarbon samples were also collected. The first sample collected had been identified as a kerosene from the provider, however, the GC fingerprint clearly illustrated a much broader range of hydrocarbons than the kerosene run earlier. The second sample was a laboratory standard mixture of gasoline and diesel fuel. The third sample was a regular grade of unleaded gasoline from a different service station. Figures 7, 8, and 9 present the GC fingerprints of these three respective samples.

Table 1. 71 Hydrocarbon Compounds Used For Analysis

1	iC4 = Isobutane		37	IP14 = C14 Isoprenoid
2	nC4 = Butane		38	1 M naph = 1 Methylnaphthalene
3	iC5 = Isopentane		39	nC13 = Tridecane
4	nC5 = Pentane		40	IP15 = Farnesane
5	2 M Pent = 2 Methylpentane		41	nC14 = Tetradecane
6	3 M Pent = 3 Methylpentane		42	IP16 = C16 Isoprenoid
7	nC6 = Hexane		43	nC15 Pentadecane
8	C6 Olefin = C6 Olefin		44	nC16 = Hexadecane
9	M Cycl Pent = Methyl cyclopentane		45	IP18 = Norpristane
10	2,4 DMP = 2,4 Dimethylpentane		46	nC17 = Heptadecane
11	Bnz = Benzene		47	Pristane = Pristane
12	Cyclo Hexane = Cyclohexane		48	nC18 = Octadecane
13	2 M Hex = 2 Methylhexane		49	Phytane = Phytane
14	3 M Hex = 3 Methylhexane		50	nC19 = Nonadecane
15	Isooctane = Isooctane or 2,2,4 Trimethylpentane		51	nC20 = Eicosane
16	nC7 = Heptane		52	nC21 = Heneicosane
17	MCHX = Methylcyclohexane		53	nC22 = Docosane
18	Tol = Toluene		54	nC23 = Tricosane
19	nC8 = Octane		55	nC24 = Tetracosane
20	EB = Ethylbenzene		56	nC25 = Pentacosane
21	m/p-xyl = m/p Xylene		57	nC26 = Hexacosane
22	o-xyl = o Xylene		58	nC27 = Heptacosane
23	nC9 = Nonane		59	nC28 = Octacosane
24	propylbenzene = n Propylbenzene		60	nC29 = Nonacosane
25	1M 3E benz = 1 Methyl 3 ethylbenzene		61	nC30 = Triacontane
26	1M 4E benz = 1 Methyl 3 ethylbenzene		62	nC31 = Hentriacontane
27	1,3,5 T M Benz = 1,3,5 Trimethylbenzene			63	nC32 = Dotriacontane
28	3,3,4 T M Hept = 3,3,4 Trimethylheptane			64	nC33 = Tritriacontane
29	1,2,4 T M Benz = 1,2,4 Trimethylbenzene			65	nC34 = Tetratriacontane
30	nC10 = Decane		66	nC35 = Pentatriacontane
31	1,2,3 T M Benz = 1,2,3 Trimethylbenzene		67	nC36 = Hexatriacontane
32	nC11 / 1,2,4,5 TeMB = Undecane and 1,2,4,5 Tetramethlybenzene			68	nC37 = Heptatriacontane
33	Naph = Naphthalene		69	nC38 = Octatriacontane
34	nC12 = Dodecane		70	nC39 = Nonatriacontane
35	IP13 = C13 Isoprenoid		71	nC40 = Tetracontane
36	2 M naph = 2 Methylnaphthalene

Once the peak areas for all of the 71 individual components were established for each of the samples, the correlation coefficient was calculated between the samples presented in Figures 1 through 5, and those presented in Figures 7, 8, and 9. Table 2 presents the results of these calculations.

Table 2. Results of Correlation Coefficient Determinations

	Gasoline I	Kerosene I	Diesel	JP-8 Jet Fuel	Crude Oil
	Figure 1	Figure 2	Figure 3	Figure 4	Figure 5
Kerosene II Figure 7	-0.156	0.732	0.882	0.932	0.379
Gas/Diesel Mixture Figure 8	0.638	0.333	0.528	0.505	0.440
Gasoline II Figure 9	0.894	-0.112	-0.213	-0.065	0.134

To evaluate the reproducibility of this process, the regular unleaded gasoline presented in Figure 9, was run on the GC five separate times, thus creating five GC fingerprints and five slightly differing numerical data sets. The data collected for all 71 compounds were then compared to each other, thus creating a total of twenty (20) correlation coefficient comparisons.

Gasoline Weathering

An algorithm was developed to model the volatilization process of gasoline released into the environment. This was accomplished by using experimental data obtained from the controlled evaporation of the three different grades of gasoline described earlier. GC fingerprint data were used to create a numerical function that describes the volatilization process. This numerical function can then be applied to fresh gasoline samples to predict what the product GC fingerprint would look like if weathered in an environmental release.

As discussed earlier, the gasoline components with the lowest boiling points tend to volatilize more rapidly than the rest of the components. The components with the highest boiling points (components at the right hand side of the GC fingerprints with retention times > 10 minutes) experience little volatilization under the weathering conditions described above.

With the above in mind, it was assumed that the actual volume of the naphthalene stayed relatively constant during the weathering simulation and can be used similar to an internal standard. By utilizing this, the GC data from each stage of the weathering process were normalized to the naphthalene peak. Once each GC fingerprint was normalized to naphthalene, each component was then evaluated as the total volume of product decreased. Once this process was completed for all components, then a matrix of volatilization multipliers was created. This matrix consists of a table of factors ranging from 0.0 to 1.0 describing the amount of volatilization of each of the 71 components at differing stages of evaporation of the total product.

To demonstrate how the matrix was created, Table 3 presents the integrated peak areas for the first 8 of the 71 components from the premium grade unleaded gasoline used in the experiment. Table 4, presents the same component integrated peak areas after they have been normalized to naphthalene. And Table 5, presents each component integrated peak area from table 4 normalized from 0 to 1. Table 5 represents the matrix of multipliers. Two other similar tables were also produced for the mid-grade and regular grades of gasoline. The entire matrix of multipliers for each of the grades of gasoline was not presented because of the size of the tables.

Table 3. Peak Areas For Components At Differing % Volatilization (Premium Grade Gasoline)

Sample Id	iC4	nC4	iC5	nC5	2 M Pent	3 M Pent	nC6	C6 Olefin
Gasoline % Vol.	Peak Area	Peak Area	Peak Area	Peak Area	Peak Area	Peak Area	Peak Area	Peak Area
Prem-0	5139	53846	129407	21998	34183	18112	14922	2865
Prem-25	0	737	23654	6196	20421	11421	10368	1940
Prem-50	0	0	0	0	1666	2305	2894	547
Prem-75	0	0	0	0	0	0	0	0

Table 4. Peak Areas For Components At Differing % Volatilization After Normalizing With Naphthalene (Premium Grade Gasoline)

Sample Id	iC4	nC4	iC5	nC5	2 M Pent	3 M Pent	nC6	C6 Olefin
Gasoline % Vol.	Peak Area	Peak Area	Peak Area	Peak Area	Peak Area	Peak Area	Peak Area	Peak Area
Prem-0	5139	53846	129407	21998	34183	18112	14922	2865
Prem-25	0	531	17047	4465	14717	8231	7472	1398
Prem-50	0	0	0	0	1165	1612	2024	383
Prem-75	0	0	0	0	0	0	0	0

Table 5. Matrix Of Multipliers For Individual Components Of Gasoline Under Differing Percentages Of Overall Product Volatilization (Premium Grade Gasoline)

Sample Id	iC4	nC4	iC5	nC5	2 M Pent	3 M Pent	nC6	C6 Olefin
Gasoline % Vol.	Peak Area	Peak Area	Peak Area	Peak Area	Peak Area	Peak Area	Peak Area	Peak Area
Prem-0	1	1	1	1	1	1	1	1
Prem-25	0	0.01	0.132	0.203	0.4305	0.4544	0.501	0.488
Prem-50	0	0	0	0	0.0341	0.089	0.136	0.1336
Prem-75	0	0	0	0	0	0	0	0

RESULTS & DISCUSSION
Reproducibility

The reproducibility of the gas chromatography analysis technique was evaluated by analyzing the gasoline sample presented in Figure 9 a total of 5 times. The 20 correlation coefficients calculated between each of the 5 analyses and the other four had a minimum of 0.99545, a maximum of 0.999989, an average of 0.997921 and a standard deviation of 0.001704. From this evaluation, truly alike GC chromatograms will probably have correlation coefficients of 0.99 or better. It is possible that other product types may have different reproducibilities since their data may include different peaks that come from a different part of the GC fingerprint.

Correlation

Table 2 presents the results of the correlation coefficients calculated when comparing the samples in Figures 1 through 5 to those in Figures 7 through 9. Prior to calculating the correlation coefficients it was expected that similar products, for example gasolines, would show higher correlations among themselves and less correlation when compared to other product types. The exact numbers, however, could not be anticipated nor how the correlation coefficients would vary between similar and different product types. Significant, logical, and reproducible differences and similarities in the correlation coefficient numbers are crucial for this process to be a useful tool. The correlation coefficient results must also make sense and compare favorably with visual inspection of the GC fingerprints. From this feasibility study, it appears that there are meaningful similarities and differences in correlation coefficient numbers calculated using GC fingerprint data. This study suggest that similar product types such as gasolines could be expected to have correlation coefficients of about 0.9 or better. Dissimilar product types have a much lower correlation coefficient of perhaps 0.6 or 0.5 or even less. The correlation coefficients shown here also make sense when compared to the visual evaluation of the GC fingerprints.

An unexpected and interesting result of the correlations was that the JP-8 jet fuel and the kerosene II sample had a high correlation coefficient. At first this seemed unusual, but it must be remembered that JP-8 and kerosene are often times from the same distillation range of the crude oil. Visual comparison of the two GC chromatograms confirms the rather high degree of similarity of the two products.

Gasoline Weathering

The matrix of multipliers created for the evaporation sequences for the three different grades of gasoline numerically models how the volatile components tend to evaporate from the sample. The matrix of multipliers can be used to numerically alter ("evaporate") the data from a fresh sample in an attempt to estimate the composition of the sample after a weathering process. Once the sample has been artificially altered, it can then be numerically compared to other controlled weathered samples.

This weathering algorithm can also be used in the inverse. For example, if one had a hydrocarbon sample from a site but did not know the extent of weathering that has already taken place, the sample could be correlated with the library of samples of known weathered gasolines. Once a library sample with the highest correlation has been determined, a matrix of multipliers of the sample with the highest correlation could be used to reconstruct the composition of the original gasoline sample when fresh. This matrix of multipliers needed to estimate the original gasoline composition would be constructed by simply using the inverse of the individual compounds within the matrix. that most closely correlated with the weathered sample. (For example, if Benzene's multiplier is 0.25, then to reconstruct the original amount of Benzene, one would multiply the peak area by 1/0.25 = 4.0.)

Future Work

The information and techniques presented in this paper represent some beginning examples of the types of analysis that are possible with GC fingerprint numerical data. Listed below are a number of additional numerical techniques and experiments that are under consideration for future work:

1. Investigate algorithms that numerically weight key compounds,
2. Develop algorithms that use the presence of unique compounds to indicate specific characteristics. For example olefins indicate the presence of catalytic cracked hydrocarbons.
3. Develop algorithms that further refine gasoline compound recognition,
4. Investigate additional controlled weathering experiments on other types of hydrocarbon products,
5. Investigate other weathering processes such as water washing, biodegradation, volatilization, and chemical speciation,
6. Expand the GC fingerprint library to include other petroleum products and hydrocarbon samples from the extraction of both soils and groundwater,
7. Investigate correlation experiments using isolated ranges of hydrocarbons; therefore, allowing mixtures of products (say gasoline and diesel fuel) to be evaluated separately and then numerically added together.