SCEC Award Number 13057 View PDF
Proposal Category Individual Proposal (Integration and Theory)
Proposal Title Network Based Estimates of Noise in Geodetic Time Series
Name Organization
Paul Segall Stanford University
Other Participants Ksenia Dmitrieva
SCEC Priorities 1, 1, 5 SCEC Groups Geodesy, SDOT, Transient Detection
Report Due Date 03/15/2014 Date Report Submitted N/A
Project Abstract
Understanding errors in GPS positions is important for a number of SCEC activities. Current approaches focus on individual time series, ignoring spatial
information. We have developed a Network Noise Estimator (NNE) that exploits the known spatial coherence of signal (plate motion and GIA) in
stable plate interiors. Synthetic tests show that NNE estimates weak long-period noise more accurately than traditional time series methods.
NNE estimates using 20 stations from the stable mid-continent with a mix of (high quality) IGS and (poorer quality CORS) stations, leads to velocity
uncertainties over three times higher than the median estimate using traditional single-station methods.
Intellectual Merit We have developed a Network Noise Estimator (NNE) for quantifying noise parameters in GPS time series.
Our underlying assumption is that tectonic signals must be spatially coherent, whereas at least some
noise sources are spatially incoherent. From a crustal deformation prospective, two stations with separation smaller than characteristic lengths of
tectonic processes should see similar signals. Any incoherent motion, such as caused by local processes (e.g., landsliding) would be considered noise from
a tectonic prospective. Some GPS noise processes, such as monument instability and multipath, are spatially incoherent, whereas others, such as
orbit errors and uncompensated tropospheric delays will be spatially correlated. It is advantageous to focus on areas, such as
stable plate interiors where the signal is well understood.
The network approach has two advantages over standard (single station) maximum likelihood methods. First, by detrending the time series, sMLE treats any trend in the data as signal. In contrast, the NNE
only removes trends due to known processes, such as plate rotation and GIA. Time dependent noise can introduce apparent trends in time series which are removed
by detrending, potentially biasing noise estimation and thus site velocity uncertainty.
Secondly, the network method analyzes multiple stations simultaneously, rather than individual time series. This allows the network method
to resolve weak random walk in the presence of flicker noise that is not revealed by sMLE.
Broader Impacts Understanding errors in GPS time series data will prove useful for other types of studies. In particular, understanding strain-rate uncertainties in
intraplate earthquake environments, such as the New Madrid Seismic Zone, are crucial to physical interpretations. Knowledge of errors in vertical
GPS time series are important for understanding other geophysical processes including sea-level rise and post-glacial rebound.
Exemplary Figure Figure 3. Noise parameters from Network estimator NNE (green diamonds) and standard MLE estimates (box plots, red bar is median, box is 75% and 25%, whiskers are 95% and 5%, and red plus symbols are outliers. Units are: mm for white noise, mm/yr0.25 for flicker noise and mm0.5 for random walk.