Automated Transient Identification and Characterization in GNSS Time Series Abstract


  • The scientific uses of GNSS time series can be placed in one of two categories: the analysis of long-term trends as applied to the study of secular crustal motion, and the modeling of short-term signals to gain insight into transient geophysical phenomena. Even in the former case, signal transients must be identified and modeled in order to obtain the best possible estimates of long-term parameters of motion. We use heuristic algorithms to locate and estimate step discontinuities and changes in station velocities. The basic model for both algorithms is a base function consisting of a linear term and annual and semiannual harmonic constituents. For step detection, a web-enabled script first flags all equipment changes that might produce a step, and step amplitudes are estimated at each of these locations by default. To find other step discontinuities, a Heaviside function is fit to the time series. at each prospective epoch The weighted sum-of-squares of residuals is used to pick the epoch at which the step is fixed. This process is repeated for successive steps. The amplitude of the Heaviside function is the criterion used to decide whether to accept a step or to end the search. Other a priori information, such as catalogs of earthquakes that might produce significant offsets, is used to aid the search for steps. Since this method generally results in many false positives, several criteria are used to minimize these and to distinguish between equipment-induced steps and steps due to geophysical causes. These criteria include step magnitude, the F-statistic for each successive step, and a measure of the steepness of the step based on an arctangent fit. The velocity change recognition algorithm is very similar to the step identification algorithm but is much more expensive computationally. The first velocity change is found by postulating a rate change at every possible epoch and fitting lines for the segments on both sides of each trial epoch. The epoch which results in the lowest chi-squared value is chosen. This location is not held fixed when looking for subsequent changes; instead, all possible change epoch combinations are tested. The F-statistic is used to determine the number of rate changes which are justified by the data. In order to identify the clusters in space in time characteristic of geophysical phenomena, we produce animated maps of steps and velocity changes to help distinguish these events from local, possibly human-induced transients and from noise that is mapped into steps. Because some identified transients may be artifacts of processing, we run our algorithms on data produced by other groups such as PBO to minimize these effects.

publication date

  • 2012

presented at event