COLLINS, STEVEN L., The University of North Carolina, Geology Department, Chapel Hill, NC

ABSTRACT: Compositional Data Analysis of Eastern Kentucky Coals

About 100 variables on 500 coal channel samples from 33 coal beds in eastern Kentucky were statistically analyzed. Data were supplied by the Kentucky Geological Survey and resulted from a joint program with the United States Geological Survey. After deletion of anomalous samples and variables with too many missing values, 402 samples and 84 variables remained.

Sixty variables, including ultimate analysis, sulfur forms, major ash elements, and major, minor, and trace elements in coal and ash, were included in one composition. Eighteen other variables were combined in compositions having two to eight parts. Centered log ratios were calculated to avoid spurious results caused by the constant-sum constraint on compositional data.

Principal components analysis was performed on centered log ratios of the 78 compositional variables plus six noncompositional variables, yielding 17 principal components, which had eigen values greater than one, and which explained 78% of the variance. The first principal component was a contrast of organic matter with mineral matter, and explained 23% of the variance. The second principal component contrasted iron, pyritic sulfur, arsenic, and ash with organic constituents, vitrinite reflectance, and ash fusion temperatures; it explained 9% of the variance. Other principal components were interpreted as representing rank, carbonate minerals, detrital minerals, and petrographic composition.

This study shows that interpretable results can be obtained when centered logratios of variables from different compositions and noncompositional variables are included in one principal components analysis.

AAPG Search and Discovery Article #90987©1993 AAPG Annual Convention, New Orleans, Louisiana, April 25-28, 1993.