DISCRIMINATION BETWEEN COASTAL SUBENVIRONMENTS USING TEXTURAL CHARACTERISTICS

The objective of this study was to discriminate between modern beach s ubenvironments based on textural characteristics obtained using the gr aphical (percentile) method, the moment method, and the log-hyperbolic distribution (LHD). A total of 126 surface sedimentation units were s ampled at the nodes of a 21 x 6 rectangular grid (1000 m(2)) on a carb onate sand beach, Oahu, Hawaii. Sampling was conducted at low energy c onditions from the lower foreshore to the backshore. Non-parametric di scriminant analysis was used as an objective tool in defining distinct subenvironments. Confidence bands around the canonical variates deriv ed from the graphic mean, sorting, skewness and kurtosis indicated fou r separate subenvironments (lower foreshore, mid-foreshore, upper fore shore and backshore). Three distinct subenvironments were identified u sing the mean, sorting (standard deviation) and skewness measures deri ved by the method of moments. A similar subenvironment distinction was obtained using five statistics of the LHD (gamma, gamma; nu nu; delta , delta; tau, tau; and xi, xi). No significant difference was noted in textural characteristics between the upper foreshore and backshore zo nes, and these zones were grouped into one subenvironment. These resul ts indicate that different process scenarios would be needed to explai n different subenvironment partitioning based simply on the approach a dopted. Discriminant analysis indicated that fewer subenvironment samp les were misclassified and separation distances between subenvironment s in bivariate canonical plots were greater for the standard moment me asures compared with the statistics derived from fitting the computati onally intensive LHD model. Examination oi. the mass frequency gain si ze distributions indicated that the LHD was generally the most appropr iate model. These observations were confirmed by the hyperbolic shape triangle which indicated that the LHD rather than the more commonly us ed log-normal distribution was generally optimal in describing sedimen ts. These results support the use of the LHD statistical measures in s ubenvironment discrimination over the graphic-inclusive measures.