2011 年 67 巻 4 号 p. I_217-I_222
Our main instruments for collecting current flow velocity data are acoustic Doppler sensors. An inherent issue with application of this type of instrument is that the observed data are contaminated by spurious records. Detecting and replacing becomes important in turbulence studies. In the present paper we introduce two effective approaches for detecting multipoint spikes effectively by applying wavelet decomposition. Moreover, by taking advantage of time series modeling a reliable method for replacing multipoint spikes is presented. This method is able of predicting future points while keeping the trend and high-frequency fluctuations. Applying the introduced detecting- replacing outliers algorithms to different samples and comparing the results with other approaches confirms reliability and accuracy of them.