Kernel Density–Based Algorithm for Despiking ADV Data

Abstract

Acoustic doppler velocimeter (ADV) data can be contaminated by spikes from various sources. Available despiking methods were found to encounter difficulties in despiking ADV data from a turbulent jet flow. An iteration-free despiking algorithm was developed for highly contaminated ADV data by applying a bivariate kernel density function and its gradient to separate the data cluster from the spike clusters. It is shown that the new method overcomes some of the deficiencies of the existing despiking methods.