Skip to main content
ARS Home » Northeast Area » Beltsville, Maryland (BARC) » Beltsville Agricultural Research Center » Hydrology and Remote Sensing Laboratory » Research » Publications at this Location » Publication #319568
Title: Assessment of de-spiking methods for turbulence data in micrometeorology

Author
item STARKENBURG, D. - Collaborator
item METZGER, S. - Collaborator
item FOCHESATTO, G.J. - Collaborator
item Alfieri, Joseph
item GENS, R. - Collaborator
item PRAKASH, A. - Collaborator
item CRISTOBAL, J. - Collaborator

Submitted to: Journal of Geophysical Research Atmospheres
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/14/2016
Publication Date: 9/1/2016
Citation: Starkenburg, D., Metzger, S., Fochesatto, G., Alfieri, J.G., Gens, R., Prakash, A., Cristobal, J. 2016. Assessment of de-spiking methods for turbulence data in micrometeorology. Journal of Geophysical Research Atmospheres. 33(9):2001-2013. https://doi.org/10.1175/JTECH-D-15-0154.1.
DOI: https://doi.org/10.1175/JTECH-D-15-0154.1

Interpretive Summary: Data describing evapotranspiration (ET) is critical for a broad range of scientific and practical applications. The eddy covariance (EC) method, one of the most widely used techniques for measuring ET, is susceptible to errors due to data spikes, i.e. noise in the raw measurements that are not due to physical processes. As a result, de-spiking is a standard step of the data post-processing. The objective of this study was to assess four alternate methods for conducting the de-spiking, each of which is based on differing theoretical and mathematical considerations. The analysis using both synthetic and observational data indicates the ability of these approaches to identify spikes without incorrectly flagging valid data varies significantly. Overall, the median filter approach proved to be the most reliable method for identifying and removing data spikes. Based on these results, it is recommended when post-processing EC data.

Technical Abstract: The direct computation of the turbulent fluxes of heat, momentum and greenhouse gases requires fast-response measurements. This is achieved by coupling data streams from sonic anemometers and infrared gas analyzers in ground-based and airborne platforms. Several steps are required to attain turbulent flux calculations from raw data. One step involves the detection and removal of sudden, short-lived variations that do not represent physical processes, and which contaminate the data (“spikes”). Our objective is to assess the performance of several noteworthy de-spiking methodologies in order to provide a benchmark assessment and a recommendation of which is most applicable to high frequency micrometeorological data in terms of efficiency and simplicity. The performance of three algorithm types (phase-space, wavelet-based, and median filter) are compared to one another and also to a statistical, window-based methodology widely used in micrometeorology. These four algorithms are first applied to a synthetic signal (a clean reference version, then one with spikes), in order to assess general performance. After rejecting the phase-space method for this analysis due to its sensitive thresholding requirements and lack of performance under low frequency turbulent motion, the remaining three algorithms are applied to a time series of real CO2 measurements taken from a micrometeorological tower which contains extreme systematic spikes every hour owing to instrumental interference, as well as several smaller, random spike points. We find that the median filter and wavelet threshold methods are most reliable, and their performance by far exceeds window-based methodologies based on the median or arithmetic mean operator (-34% and -71% reduced RSMD, respectively). Overall, the median filter is most easily automatable for a variety of micrometeorological data types, including data with missing points and low frequency coherent turbulence.