DBPapers
DOI: 10.5593/SGEM2014/B22/S9.074

THE USE OF LOCALLY WEIGHTED SCATTERPLOT SMOOTHING IN THE ANALYSES OF GPS TIME SERIES AUTOCORRELATIONS

J. Bogusz, M. Figurski, A. Klos, A. Araszkiewicz
Wednesday 1 October 2014 by Libadmin2014

References: 14th International Multidisciplinary Scientific GeoConference SGEM 2014, www.sgem.org, SGEM2014 Conference Proceedings, ISBN 978-619-7105-11-7 / ISSN 1314-2704, June 19-25, 2014, Book 2, Vol. 2, 591-598 pp

ABSTRACT
The GPS time series autocorrelations are the commonly known phenomena nowadays. Such autocorrelations can be identified both in the deterministic and stochastic part of the data. Their existence in the deterministic part of GPS time series has the form of trend (interpreted as station’s velocity) and seasonal components (annual and semiannual). The autocorrelation in the stochastic part of GPS time series was proven in the form of its power-law long-range dependencies with a numerous of methods. In this research, the use of locally weighted scatterplot smoothing (LOESS) was tested in the autocorrelation analyses. The LOESS method is described by the two parameters – the polynomial order and smoothing parameter. To analyse the LOESS function, we used the daily changes of topocentric coordinates in the ITRF2005 obtained within ‘repro1’ project from Polish EPN (EUREF Permanent Network) stations. The time series were obtained by the Centre of Applied Geomatics that cooperates at the Military University of Technology as one of the 16 independent EPN Local Analysis Centres (MUT LAC). It was investigated within the research, that the trend-related behaviour is modelled the best by both smoothing parameter and polynomial order equal to 1. The polynomial order equal to 2 with smoothing parameter close to 0.1 fits seasonal components quite well. Greater values of smoothing parameter flatten time series too much, while lower ones detect higher frequency changes. For all of the LOESS modelled curves, the autocorrelation function (ACF) was calculated and its values juxtaposed for different types of modelled phenomena.

Keywords: GPS, EPN, LOESS, time series autocorrelation