 |
| Titel |
Analysing the impact of data smoothing procedures on temporal correlations using examples of GPS residual time series |
| VerfasserIn |
X. Luo, M. Mayer, B. Heck |
| Konferenz |
EGU General Assembly 2009
|
| Medientyp |
Artikel
|
| Sprache |
Englisch
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 11 (2009) |
| Datensatznummer |
250023279
|
|
|
|
|
|
| Zusammenfassung |
| The understanding of dynamical processes in the complex Earth system requires not only a
large amount of observation data in form of time series but also appropriate analysis
procedures to decompose the time series into deterministic signals and stochastic noise. The
signals can be extracted by data smoothing to create an approximating function which
attempts to capture important patterns in the data while the noise component can be fitted by
means of stationary probabilistic models, e.g. autoregressive (integrated) moving average
(AR(I)MA) processes. Obviously, the data smoothing algorithm applied in signal
processing affects the selection of an appropriate stationary time series model for the
noise.
In this paper several data smoothing methods (finite moving average, robust weighted
local regression, exponential smoothing, finite impulse response filter) are analysed and their
influences on temporal correlations in the obtained noise sequences are investigated by means
of various hypothesis tests based on empirical autocorrelation function and empirical spectral
density respectively as well as non-parametric tests. The analysed database consists of 285
GPS double difference residual time series resulting from 1-Hz data processing and 191
simulated time series using first order autoregressive AR(1) resp. moving average MA(1)
processes. Each time series has the same length of 3600 values and the GPS data are
processed applying an improved observation weighting model based on signal-to-noise ratio
measures using the Bernese GPS software version 5.0. The presented test results are
largely consistent and show significant mitigation in temporal correlations after
exponential smoothing. Additionally, a higher local regression degree results in stronger
temporal correlations in the noise sequences. The presented data smoothing methods
and hypothesis tests can be applied analogously to other similar research works. |
|
|
|
|
|
|