Discussion of ‘ ‘ Despiking Acoustic Doppler Velocimeter Data ’ ’ by Derek

The authors are commended for their novel approach to the ficult problem of removing spikes from velocity time series r corded by acoustic Doppler velocimeters ~ADVs!. Their paper has prompted some further investigations into modifications may make the method even more practicable and robust. To briefly review, spikes in ADV time series can be caused many factors, including high turbulence intensities, aerated fl that have undesirable acoustic properties, and phase differ ambiguities that occur when velocities exceed the upper limit ADV probe velocity ranges. Although spikes can be reduced eliminated in many cases by adjustment of probe operationa rameters, there are some situations in which spikes canno entirely avoided. A method for spike detection such as that p posed by the authors is thus necessary. The phase-space thresholding method set forth by the au is based on the observation that good ADV data are tig clumped within an ellipsoid in phase space @three-dimensiona ~3D! plots of velocity, u, and approximations of the first an second derivatives, Du andDu, respectively#. Spikes are thos points plotting outside of elliptical projections of the ellipso onto the three principal phase-space planes, u–Du, Du–Du, andu–Du. The first two ellipses are symmetric about both ax while the third is rotated due to correlation between u andDu; positive velocities are associated with negative second deriva and vice versa as the velocity tends to move back toward average value when an extreme is reached. The authors siz ellipsoid and the projected ellipses using the standard deviat of the u, Du, and Du time series and a Universal Thresho parameter related to the length of the time series. The Unive Threshold requires no tuning, a fundamental advantage of algorithm. The method is iterative, with spikes identified and placed, and the algorithm repeated one or more times. Iteratio necessary because the standard deviations, and thus the e boundaries, are sensitive to the spikes and change as spike identified and replaced. The authors point out that spike repl ment is an arbitrary procedure with many strategies availa Spike replacement is necessary to use the iterative phasethresholding method, since derivative and second derivative mates are needed in subsequent iterations.