CLUSTER ANALYSIS: A NEW TOOL FOR DETERMINING SEDIMENT PROVENANCE

A recent study by Ackerson et al. (2015) showed that trace element abundances can be used to “fingerprint” the source of detrital quartz in modern sediments. They used scatter plots of titanium versus aluminum abundances to show that quartz grains derived from different types of granitoids and other rocks tend to cluster in different regions of such a plot. By comparing Ti-Al scatter plots of sands from a beach near the mouth of the Bega River (New South Wales, Australia) with analogous plots for rocks within and around the Bega catchment, Ackerson et al. were able to confirm the provenance of the sands and estimate the fractional contributions of different granitoid types to the sediment load.

In this paper we extend the work of Ackerson et al. to quantify the statistical significance of clustering in trace element plots. We use a well-studied statistical method, cluster analysis through the application of kernel density estimation. This method first separates the two abundance variables (Ti and Al) and convolves the data for each with a “kernel function” (a Gaussian in our study). The result is a smooth function for the probability density of each abundance variable. The two separate variables are then combined into a multivariate probability density which reveals how those densities relate to each other. When a new sample is collected; and the trace element abundances in quartz are measured, the Ti and Al content of the quartz grains can then be compared to the existing probability function to identify which populations they belong to. With this method, it will be possible to identify the chemical and geologic history behind quartz grains in a sediment sample, without direct knowledge of the source environment. The general technique of multivariate cluster analysis is applicable to trace-element analyses of all types of Earth materials, and perhaps even extraterrestrial samples.