STATISTICAL EVALUATION OF GRAIN SIZE FACIES MODEL FOR BARRIER ISLAND SYSTEMS, FIRE ISLAND, NEW YORK

Recently refined methodology for measuring coarse-grained sands using laser diffraction has reduced analytical measurement uncertainty to under 2%. This improved resolution in up-core grain size measurements allows improved differentiation between transgression / regression sedimentary facies of active coastal margin deposits. In this study we applied a refined methodology to allow grain size to be used as a proxy of barrier island depositional environment. The methods were applied to five sediment cores taken in a shoreline normal transect across Kismet, Fire Island, New York and sampled at centimeter scale intervals. Modern sediments from the well-developed barrier island facies were used as an experimental control and marker for the statistical analyses.

This study strengthens existing models by introducing a new method of recognizing clusters in the data through the use of an unsupervised k-means clustering algorithm. The algorithm can efficaciously be applied to the data as an unbiased way to recognize clusters in the statistically analyzed grain size data as well as in the grain size data plotted with depth.

The study confirms previous research that showed statistically analyzing grain size data can be used as a method for facies modeling. The combined use of clustered skewness and unsupervised k-means increased facies cluster separation allowing improved facies identification. The cluster models allow us to link five barrier island facies (beaches, dunes, backbarrier or aeolian flats, lagoonal, and inlets) with modern surface sediments controls. This proxy method can be used to develop facies secession models and with expanded spatial extent the ability to address the evolution of barrier island systems through 3D subsurface modeling.