A new method to process borehole strainmeter data; least squares with correlated data
Abstract
The newly installed Plate Boundary Observatory (PBO) strainmeters record signals from tectonic activity, Earth tides, and atmospheric pressure. Some of the tectonic signals have amplitudes close to those of tides and pressure loading. If incorrect assumptions are made regarding the background noise in the data, then adjusting these strain data will produce incorrect results that can obscure or contaminate any underlying tectonic signal. The use of simplifying assumptions that data are uncorrelated can lead to such incorrect results and, for example, pressure loading will not be completely removed from the raw data. Instead, any algorithm used to process strainmeter data must incorporate the strong temporal correlations that are inherent with these data. For instance, techniques based on auto-regressive methods or Kalman filters can successfully remove the pressure load and the Earth tides. The technique described here is adapted from error analysis of geodetic time-series of ground displacements. The technique uses least squares but employs data covariance that describes the temporal correlation of strainmeter data. There are several advantages to this method since many parameters are estimated simultaneously. These parameters include: 1) functional terms that describe the underlying error model, 2) the tidal terms, 3) the pressure loading term(s) 4) amplitudes of offsets, either those from earthquakes or from the instrument, 5) rate and changes in rate, and 6) the amplitudes and time constants of either logarithmic or exponential curves that can characterize postseismic deformation or diffusion of fluids near the strainmeter. With the proper error model, realistic estimates of the standard errors of the various parameters are obtained; this is especially critical in determining the statistical significance of a postulated, tectonic strain signal. Because the algorithm uses a maximum likelihood method, it is cpu-intensive. However, obtaining the error model by fitting a power-law relation to the power spectrum of the adjusted data greatly reduces the computations. The algorithm described here also provides a method of tracking the various adjustments required to process strainmeter data. In addition, the algorithm provides several plots to assist with identifying either tectonic signals or other signals that may need to be removed before any geophysical signal can be identified.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2008
- Bibcode:
- 2008AGUFM.G21B0689L
- Keywords:
-
- 1211 Non-tectonic deformation;
- 1294 Instruments and techniques;
- 3270 Time series analysis (1872;
- 4277;
- 4475)