HR: 1340h
AN: A43D-1557    [Abstracts]
TI: A novel application of median filters and Fourier transforms for eddy covariance data
AU: * Frank, J M
EM: jfrank@fs.fed.us
AF: U.S. Forest Service, RMRS 240 W. Prospect Rd., Fort Collins, CO 80526, United States
AU: Massman, W J
EM: wmassman@fs.fed.us
AF: U.S. Forest Service, RMRS 240 W. Prospect Rd., Fort Collins, CO 80526, United States
AB: Eddy covariance (EC) is the standard for measuring trace gas fluxes between the atmosphere and the plant- and soil-covered terrestrial surface. Because EC fluxes are based on concepts of transport by turbulent atmospheric motions they require lengthy, computer-driven, intensively sampled data streams. But as the data stream becomes more machine-driven, the more important it is to be able to objectively identify, and possibly correct for, outliers, such as data spikes, drifts, and discontinuities, any of which if undetected can significantly impact the EC flux estimates and reduce their reliability and value. This study describes a novel use of the median filter and the Fourier transform to identify spikes, drifts, and discontinuities in high-frequency eddy covariance data streams and in the associated, but less intensively sampled, ambient meteorological data. For either application, the median filter is used to separate the data stream into a time dependent mean, M(t_i), and a time dependent deviation from that mean, σ(t_i). In general, the advantage of the median filter is that it is not as influenced by bad data as the running mean filter or other linear filters. For the high-frequency EC data stream the median filter is combined with an often used turbulence despiker [Hojstrup 1993: 'A statistical data screening procedure'; Measurement Science and Technology 4, 153-157]. The resulting algorithm is considerably more robust and efficient. For the supporting ambient meteorological data the median filter cleanly removes the trends associated with sensor drift and data discontinuities. Data spikes are then removed by combining a second median filter with the Fourier transform, which are used in tandem to construct an ideal deviation time series [a σ(t_i) that would occur if the data were not contaminated by noise, spikes, etc]. The ideal σ(t_i) is then compared with the observed σ(t_i) to develop objective criteria for spike detection. The spike detection algorithms for both high- frequency EC data and the ambient meteorological data allow for the seasonal variability in the data stream and the associated noise detection thresholds.
DE: 0394 Instruments and techniques
DE: 0452 Instruments and techniques
SC: Atmospheric Sciences [A]
MN: 2007 Fall Meeting