![Hier klicken, um den Treffer aus der Auswahl zu entfernen](images/unchecked.gif) |
Titel |
Testing stationarity hypothesis and its influence on low frequency turbulent flux variability |
VerfasserIn |
Luis Martins, Otávio Acevedo, Gulherme Welter, Franciano Puhales, Felipe Costa |
Konferenz |
EGU General Assembly 2011
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 13 (2011) |
Datensatznummer |
250046231
|
|
|
|
Zusammenfassung |
Eddy covariance is the most used method for quantifying turbulent fluxes and the
associated transport of scalar and vectorial quantities. However, only if the time
series considered are stationary, in the sense that their statistical moments are not a
function of time, such fluxes can be represented by a covariance. Furthermore, the
turbulent flow in the atmospheric boundary layer is subject to processes having
different temporal and spatial scales, so that micrometeorological observational
time series are very often non-stationary. Spectral analysis is a commonly used
tool in micrometeorology. An example of such approach is the multiresolution
decomposition, which consistently shows that there is a temporal scale beyond
which the cospectra presents an extremely enhanced variability, often referred as a
cospectral gap (Howell and Mahrt, 1997; Vickers and Mahrt, 2006). Such gap is usually
regarded as the limit between turbulent and mesoscale (or submeso) motions. The
purpose of the present study is to verify whether the large flux variability on scales
larger than the cospectral gap is caused by the fact that the stationarity hypothesis is
violated on such scales, as a consequence of mesoacale events and daily cycles. A
secondary goal is to compare different stationarity tests existent in the literature
and verify which are better suited to analyse turbulent atmospheric time series.
The first step of the analysis was to consider null hypothesis tests, which permit
rejecting or not a given imposed hypothesis about two populations based on samples
taken from them. Using t-student’s test, f-test (Andreas et al., 2008), Tukey’s test,
run test (Dias et al., 2004) and median test, it was possible to verify the temporal
independence between some statistical properties, such as means, variances, medians and
distributions. To analyze the importance of trends, an empirical mode decomposition
(EMD) was used. Such decomposition consists on decomposing the series on an
adaptive base, resulting in a set of intrinsic mode functions (Huang et al., 1998,
1999). Xu et al. (2004) showed that the EMD residual naturally represents the mean
temporal variation, which is somehow related to non-stationarity. Assuming that the
decomposition residual represents a mean trend of the turbulent variables, it allows the
determination of how the trends affect both the turbulent fluxes and the time series
stationarity. The data used for the analysis were obtained at a micrometeorological
experiment carried out at Candiota, in southern Brazil, in 2007. Twelve daytime periods
from 1000 to 1700 LST were considered, consisting of 10 Hz series of the Wind
components and temperature, at 8 m above the surface. The results indicate that
t-student’s test was the one that better suited our purpose, because it explicitly tests the
mean temporal variability, being therefore sensible to temporal trends. The averaged
stationarity temporal scale found with this test coincides with the multiresolution
cospectral gap, supporting the hypothesis that low frequency variabiliyy is caused by
non-stationarity. Besides, such test was consistently sensible to low-frequency trend, in the
sense that the average stationarity tendency increased as low-frequency modes were
removed. Such result was not achieved with other stationarity tests considered. |
|
|
|
|
|