An algorithm for finding spurious points in turbulent signals Free

An algorithm is presented for finding spurious points in time series consisting of data from a turbulent process. The usual statistical criterion of eliminating points a certain number of standard deviations (σ) from the mean is shown to work well provided that (1) running means and standard deviations of about 100 points are used, (2) the interval for testing is moved by one point at a time, and (3) spurious points are eliminated from the statistics as the run progresses. This procedure allows the use of a criterion that points be 4 or 5σ from the running mean, and it preserves near discontinuities and other desired features in the time series. An automated procedure such as this is essential for the processing of certain types of data, such as those obtained from some spacecraft instruments.