![Hier klicken, um den Treffer aus der Auswahl zu entfernen](images/unchecked.gif) |
Titel |
Eddy-covariance data with low signal-to-noise: time-lag determination, uncertainties and limits of detection |
VerfasserIn |
Ben Langford, Joe Acton, Christof Ammann, Amy Valach, Eiko Nemitz |
Konferenz |
EGU General Assembly 2015
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 17 (2015) |
Datensatznummer |
250107284
|
Publikation (Nr.) |
EGU/EGU2015-6977.pdf |
|
|
|
Zusammenfassung |
In addition to systematic errors, eddy-covariance flux measurements are subject to two main
random errors. These are associated (a) with the geostatistical representation of
turbulence through a single measurement and (b) the instrument noise. While the
former error is usually the main component for flux measurements of CO2 and water
vapour, trace gases and aerosols are often measured with sensors providing limit
signal-to-noise ratios (SNR). Examples include particle counters, mass spectrometry methods
(Proton Transfer Reaction Mass Spectrometry for VOCs, PTR-MS; Aerosol Mass
Spectrometry, AMS), optical spectrometers for CH4 and N2O, as well as some fast ozone
sensors.
The analysis of flux data from noisy instruments that deploy inlet lines is further
complicated by the fact that preferably the time-lag is determined by maximisation of the
cross correlation between vertical wind component and concentration. If this approach is
applied to noisy data, random errors nevertheless induce a systematic bias towards larger
values of emission or deposition. This results in a poorly shaped frequency distribution in
the derived fluxes, with hardly any fluxes near zero, and an average that differs
significantly from the true average. While this problem has been noted regularly in
the literature, a systematic assessment of the effect does not appear to have been
made.
We here examine the consequences using example data from a range of instruments and
by performing numerical experiments on temperature data, that is degraded to mimic the
behaviour if different instruments. This study explores the effect using three different
methods to determine the time-lag. It provides a novel approach to assessing the random error
due to random instrument noise separately and recommends an optimised strategy for data
processing and the calculation and reporting of errors when using instrumentation with low
SNR.
Finally we demonstrate how flux data, even if associated with a large relative error, is
nevertheless useful, if each flux value is reported together with its uncertainty. When
averaging the fluxes for the calculation of long-term budgets, average diurnal cycles or
according to a driving force, long time-series still provide statistically significant results. |
|
|
|
|
|