Some observations on assessment of gaussianity for correlated profiles

Abstract

The classical Anderson-Darling test for Gaussianity cannot be applied on correlated data. It is possible to circumvent correlation by resampling the data at wider spacing. A simple resampling criterion is developed using digitally simulated standard Gaussian realizations following a single exponential correlation structure. The effects of resampling on Gaussianity assessment are assessed on actual normalized cone tip resistance profiles from highly homogeneous soil units. Copyright ASCE 2006.