Hide/Show Apps

Transient signal detection in continuous GPS coordinate time series using empirical mode decomposition and principal component analysis

Özdemir, Soner
Continuous Global Positioning System (GPS) coordinate time series might be exposed to tectonic and non-tectonic transient signals as well as the persistent signals such as secular rates and seasonal motions. Transient signal detection becomes challenging when the targeted signal is weak and buried in the noise. Incoherency of the transient signal in space and large number of sites in the GPS network make the detection even more complicated. We propose a new approach based on Empirical Mode Decomposition (EMD) and Principal Component Analysis (PCA) in this study to detect tectonically-driven transient signals. Thanks to the adaptive nature of EMD, we did not make any a priori assumptions about the type of the colored noise present in the time series, and suppressed the white noise by means of the filtering properties of the EMD method. We utilized PCA to recognize the coherent but localized transient signals in space. The main geographic area of interest is Turkey. We focused on analyzing the data collected in CORS-TR network, which is a real-time kinematic type permanent network in Turkey, and making the time series employable in tectonic monitoring. We demonstrated the existence of Common Mode Error (CME) at CORS-TR sites and reduced it for further investigations. We proved the capability of our method by successfully detecting the slow slip events in Cascadia, the transient inflation at Akutan volcano, Alaska, and the postseismic deformation following the October 23, 2011 Van earthquake, Turkey.