Simplified Kernel Density–Based Algorithm for Despiking ADV Data

Acoustic Doppler velocimeters (ADVs) are often used to measure turbulent velocities in laboratory and field settings. However, in many applications of ADVs, erroneous velocity spikes that distort turbulence statistics and spectral distributions are recorded during data collection. In this study, we present a new, efficient, and robust despiking method using a univariate kernel density estimation (KDE) of the measured velocity to identify spikes within the time series. The new technique builds upon previous bivariate and univariate KDE methods and can be applied to both clean and spike-contaminated time series. The new method was developed and compared with other KDE methods using ADV data from the field and available direct numerical simulation (DNS) results from a turbulent boundary layer. The methods were evaluated using visual assessment of the time series, comparison of the power spectra, and changes in the statistical moments of the velocity distributions. The analysis showed that the new method provides improvement over past methods when applied to a wide range of time series. In particular, the new method is computationally efficient and more capable than previous methods of not overfiltering clean data while retaining its ability to filter out spikes.