Abstract
SUMMARYInappropriate mathematical treatment of prediction errors associated with inaccurate forward modelling in an inversion scheme may result in significant unnatural short-wavelength components in the estimated slip distribution, which is a typical consequence of overfitting data. When geodetic data in observation stations following a non-uniform spatial distribution are used in a geodetic slip inversion, the spatial non-uniformity of the observation can possibly influence the distribution pattern of the short-wavelength components significantly, which may be confused with slip patterns that are geophysically meaningful. Such situations often occur when land and seafloor geodetic data are used in combination in slip inversions. To avoid overfitting, this study proposes a method that incorporates covariance components in the covariance matrix of the misfit vector, which originate from prediction errors. Because the proposed method retains the linearity of the inversion problem, widely known approaches that introduce prior constraints to a linear inversion problem are easily combined with the proposed method. This study demonstrates a combination of the newly introduced covariance components with a prior constraint on the smoothness of slip distribution, constructing a Bayesian model with unknown hyperparameters, which are objectively determined by minimizing Akaike’s Bayesian information criterion. In the synthetic tests, the proposed method estimated slip deficit rate (SDR) distributions that are closer to the true one, avoiding overfitting the geodetic data with spatial non-uniformity. By contrast, a conventional approach, which does not introduce covariance components, estimates unnaturally rough SDR distributions using the same synthetic data. The proposed method was applied to the estimation of SDR in the Nankai Trough subduction zone, using geodetic data of displacement rates provided by land GNSS stations and seafloor GNSS-Acoustic stations. This method estimates a reasonably smooth distribution of SDR, avoiding overfitting. The spatial distribution of residuals of the displacement rates suggests that the proposed method avoids overfitting some portions of the observed displacement rates that the forward model set for the analyses could not fundamentally explain.
Highlights
The development of the observation technique introduced seafloor geodesy such as GNSS (Global Navigation Satellite System)-Acoustic (GNSS-A) techniques and ocean bottom pressure (OBP) gauges to geodetic slip inversion, which was previously performed mainly based on GNSS data
We observe a significant change of residuals around Median Tectonic Line (MTL) and the Beppu–Shibabara graben (BSG), west of MTL; significant displacement rates towards the west are seen in the land station at the south of MTL and BSG, but are not seen at the north
I proposed a method to incorporate proper covariance components originated by modeling errors in geodetic slip inversion to avoid overfitting to geodetic data in biasedly distributed observation stations
Summary
The development of the observation technique introduced seafloor geodesy such as GNSS (Global Navigation Satellite System)-Acoustic (GNSS-A) techniques and ocean bottom pressure (OBP) gauges to geodetic slip inversion, which was previously performed mainly based on GNSS data. [6, 21] introduced covariance components that depend on model parameters to make estimations, focusing on the error in Green’s functions due to factors such as the uncertainty of elastic parameters and dip angles They formulated a nonlinear inversion based on a Bayesian sampling algorithm, while the geodetic slip inversion is commonly formulated as a linear problem. We propose an alternative approach that estimates spatial covariance between each geodetic data based on the Green’s function itself Because this method does not require model parameters to calculate the data covariance matrix, widely accepted linear inversion approaches such as [31] can be applied in the same way used by [32] for seismic waveform inversions.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.