Are the Plates Still Rigid? - A Case Study of GPS Time-Series Noise in Australia

GPS has been widely applied to monitor and quantify plate tectonics. Previous studies have demonstrated that within the uncertainties of GPS velocities, plates appear to be rigid. Now that we have longer and less noisy time-series and a better understanding of time-series noise models, a new study into velocity uncertainties and their evolution over time is warranted. We determined from a proper assessment of time-series noise whether velocity uncertainties are still consistent with Australia being a geodetically rigid plate. We analyzed time-series of nine continuous GPS sites in Australia to determine the evolution of horizontal velocities and their uncertainties between 1998 and 2009.5. We used the CATS software to estimate the best power-law (plus white noise) model to explain the noise in time-series of 1) fixed 2.5 year intervals and 2) increasing time span. We also estimated velocity uncertainties under the assumption that the noise is characterized by flicker plus white noise. We ran similar analyses for time-series in an ITRF2005 reference frame and for time-series that have continental scale common-mode noise removed. Even though the power law for 2.5-year time-spans ranges from -0.2 to -1.5, these models do not fit the noise better than a flicker noise model. In fact, we found there to be a trade-off between the power law index and amplitude, which is because 2.5 years is still too short to reliably obtain power law noise parameters. For increasing time-spans (as well as more recent times), the index approaches -1 (particularly in North direction), and over time Maximum Likelihood Estimates (MLE) go up. This confirms that flicker plus white noise is the appropriate noise model. Over time, the amplitude of the flicker noise decreases, and consequently the velocity uncertainty goes down. We performed similar analyses on filtered time-series. In the filtering we removed for every time-series a common-mode error that was estimated each day from a weighted stack of position residuals from all other time-series. We found that the amplitudes of the power law noise for the filtered time-series are typically 1.5-2.5 times smaller than the unfiltered cases. There can be a large, but unsystematic, change of the index. These results suggest that filtering the time-series makes them less noisy, but that it could also alter the intrinsic noise model. For each time-period, we estimated a rigid body rotation from the horizontal velocities. For the older time series, velocity residuals are greater than 1-sigma velocity uncertainties, but with increasing time span, the residuals fall outside the uncertainties. This effect is more extreme when using results from the filtered time-series. The residual velocities do not indicate a coherent continent-scale intraplate deformation, but could reflect regional tectonic strain rate. More local GPS data is need to verify that. Alternatively, the observed velocities may indicate monument instability. If true, local site effects could proof to be the limiting factor in using GPS in refining plate motion and detecting small tectonic signals.