A Case Study of the Performance of Different Detrending Methods in Turbulent-Flux Estimation

Abstract

The performance of different detrending methods in removing the low-frequency contribution to the calculation of turbulent fluxes is investigated. The detrending methods are applied to the calculation of turbulent fluxes of different scalars (temperature, ultrafine particle number concentration, carbon dioxide and water vapour concentration), collected at two different measurement sites: one urban and one suburban. We test and compare the performance of filtering methodologies frequently used in real-time and automated procedures (mean removal, linear detrending, running mean, autoregressive filter) with the results obtained from a reference method, which is a spectral filter based on the Fourier decomposition of the time series. In general, the largest differences are found in the comparison between the reference and the mean-removal procedures. The linear detrending and running-mean procedures produce comparable results, and turbulent-flux estimations in better agreement with the reference procedure than those obtained with the mean-removal procedure. The best agreement between the running mean and the spectral filter is achieved with a time window of 15 min at both sites. For all the variables studied, average fluxes calculated using the autoregressive filter are increasingly overestimated for a time constant \(\tau \) compared with that obtained using the spectral filter. The minimization of the difference between the two detrending methods is achieved with a time constant of 120 s, with similar behaviour observed at both sites.