Despiking velocity time-series-Optimisation through the combination of spike detection and replacementmethods

The experimental investigation of turbulence has been greatly aided by the development of instruments capable of measuring and recording instantaneous velocity measurements at high frequencies, such as Laser Doppler and Acoustic Doppler Velocimeters. As a consequence of the techniques and algorithms used by some of these instruments, the introduction of 'spikes' of invalid data into the recorded velocity time-series is inevitable, with resulting errors in the calculated turbulence characteristics. These spikes should, therefore, be removed and replaced with statistically valid values if power spectra and parameters such as turbulence intensity are to be examined. A number of existing spike detection and replacement methods are discussed and combinations of these have been applied to data from a variety of sources-artificially corrupted test data, data from laboratory experiments in a flume and data from field experiments in a natural river. Earlier recommendations, based on studies using a smaller number of despiking methods and data sources, are improved; contrary to previous findings, Phase. -Space Thresholding is shown to accurately reconstruct the power spectrum when used with the appropriate replacement method.