Are Eddy Covariance Series Stationary?.

Spectral analysis via a discrete Fourier transform is used often to examine eddy covariance series for cycles (eddies) of interest.  Generally the analysis is performed on hourly or half-hourly data sets collected at 10 or 20 Hz.  Each original series is often assumed to be stationary.  Also automated filtering in frequency space is usually then utilized.  Through the examination of multiple data sets, this study examines the ‘stationary’ assumption.  The x vector wind speed (horizontal) and CO₂ flux series for multiple hourly and half-hourly 20 Hz records from the 2008 Bushland Evapotranspiration and Remote Sensing Experiment (BEAREX) are tested for trend via two nonparametric procedures - Kendall’s tau (τ) and Spearman’s rank correlation coefficient (r_s).  The analyses reveal that strong trend is almost always present in both variables for both time intervals.  Night time wind speeds tend to have higher absolute values for both τ and r_s compared to their corresponding day time values.  The coefficients for the CO₂ flux series generally exhibit a diurnal pattern with negative values during the day time hours and positive otherwise.  Eddy covariance series for horizontal wind speed and CO₂ flux from hourly and half-hourly periods are, thus, not generally stationary.  Some examples of removing the trend or dealing with the trend will be presented.