Investigating the Effect of Stochastic Model in the Time Series Analysis: Noise Models Concerned


Creative Commons License

Duman H. , Şanlı D. U.

INTERNATIONAL SYMPOSIUM ON GIS APPLICATIONS IN GEOGRAPHY AND GEOSCIENCES, Çanakkale, Türkiye, 18 - 21 Ekim 2017, ss.104

  • Basıldığı Şehir: Çanakkale
  • Basıldığı Ülke: Türkiye
  • Sayfa Sayıları: ss.104

Özet

GPS time series consisting of thousands of daily coordinates have been used frequently to determine geophysical phenomena such as plate motion. In this study, it is aimed whether there is any statistical difference between velocity estimations using both maximum likelihood estimation (MLE) with noise models and least square estimation (LSE) without noise models as the number of coordinates in the time series increases. Furthermore, it is generally suggested that at least 2.5 years data span ought to be adopted in the literature. For this purpose, the time series of continuous GPS stations (GOLD, JPLM, PIN1, PIN2 and VNDP) operated over two decades, whose data have been archived and published by the JPL, were used. Initially, the first 2.5 years data span of each time series was extracted and analyzed via Hector v1.6 using MLE by introducing white noise and power-law noise combination. The data were also analyzed using LSE and applying harmonic analysis by including annual and semi-annual terms and not introducing any noise model. Following that, the time series were analyzed over longer periods by extending the 2.5 year data 6-months at each step until all the data belonging to a station were analyzed. At each analysis step, hypothesis testing was applied to test the significance of the difference in estimated velocities. Approximately 96% of the differences were found to be statistically not significant.