Assessment of Despiking Methods for Turbulence Data in Micrometeorology

Abstract

AbstractThe computation of turbulent fluxes of heat, momentum, and greenhouse gases requires measurements taken at high sampling frequencies. An important step in this process involves the detection and removal of sudden, short-lived variations that do not represent physical processes and that contaminate the data (i.e., spikes). The objective of this study is to assess the performance of several noteworthy despiking methodologies in order to provide a benchmark assessment and to provide a recommendation that is most applicable to high-frequency micrometeorological data in terms of efficiency and simplicity. The performance of a statistical time window–based algorithm widely used in micrometeorology is compared to three other methodologies (phase space, wavelet based, and median filter). These algorithms are first applied to a synthetic signal (a clean reference version and then one with spikes) in order to assess general performance. Afterward, testing is done on a time series of actual CO2 concentrations that contains extreme systematic spikes every hour owing to instrument interference, as well as several smaller random spike points. The study finds that the median filter and wavelet threshold methods are most reliable, and that their performance by far exceeds statistical time window–based methodologies that use the median or arithmetic mean operator (−34% and −71% reduced root-mean-square deviation, respectively). Overall, the median filter is recommended, as it is most easily automatable for a variety of micrometeorological data types, including data with missing points and low-frequency coherent turbulence.