USE OF CLUSTER AND DISCRIMINANT ANALYSIS IN IDENTIFYING CONFORMABLE-UNCONFORMABLE STRATIGRAPHIC RELATIONSHIPS

Neogene stratigraphic units in the Atlantic and Gulf Coastal Plains consist largely of sparsely fossiliferous, non-marine coarse clastics. While up-gradient exposures may often show erosional surfaces between various units, their down dip characteristics are frequently more controversial. Two classic examples of this include: (1) the Plio-Pleistocene Citronelle Sand versus underlying upper Miocene units in the Alabama-west Florida area and (2) the late Miocene-Pliocene Cohansey Sand versus the middle-late Miocene Kirkwood Formation in the New Jersey Coastal Plain.

Two multivariate statistical procedures can be used to provide help in clarifying down gradient relationships. Cluster analysis can operate on either a correlation or similarity matrix of sample data (sediment texture/mineralogy, etc.) and produce “groupings” of samples that share similar characteristics. The groupings are graphically displayed in the form of a dendrogram and have the advantage of preserving the identities of individual samples. Once samples have been assigned to groups, the powerful analytical procedure known as discriminant analysis can be used to determine: (1) which variables best separate samples belonging to different groups and (2) to generate an equation that permits samples from “floating sections” to be assigned to the most likely group.

Both of these procedures were successfully used to show that an unconformable relationship clearly exists between upper Miocene coarse clastic units and overlying Citronelle Sands in the northern Gulf Coast and to show that the Citronelle is considerably less thick than was previously believed. Similarly, statistical analysis of down-dip sediment properties of the Cohansey-Kirkwood sequence in New Jersey provided a strong argument that, while these units are separated by a (minor) unconformity up-dip, a conformable relationship between the two exists down gradient.