Automating Procedures in Initiating a Full Waveform Inversion Process of GPR data
Abstract
Full waveform inversion of GPR data offers the promise of improved resolution of subsurface heterogeneities. For surface GPR data collected in the commonly used constant-offset mode, however, computational expense limits us to methods that find local rather than global minimum solutions to the inverse problem. As a result it is critical to have a good starting model, a good initial estimate of the source wavelet, and optimally selection of inversion parameters. We propose methods to assist in automatically setting up the initial model of subsurface structure and the source wavelet for the full-waveform inversion (FWI) process using grid search and other algorithms. The goal is to reduce the need for subjective user model and parameter selections.
We use a Sparse Blind Deconvolution (SBD) algorithm that estimates both the source wavelet and the sparse representation of the subsurface reflectivity model using an alternating fashion. However, this algorithm requires two critical Split-Bregman parameters that require user's adjustment and frequent tests. We propose to implement a grid search algorithm that is capable of finding the optimized values for the Split-Bregman parameters that later can provide the optimized source wavelet and the sparsest reflectivity model. Second, we focus our efforts on full-waveform inversion of GPR profiles dominated by diffraction hyperbolas from objects such as pipes, rebar, or tree roots. We test various methods described in the literature for preliminary identification of the locations of the target objects producing the diffraction hyperbolas. Once the preliminary location of the target object is established, ray-based analysis can be used to get improved estimates of soil velocity, and target depth and size. These estimates in turn are used as the starting model for the full-waveform inversion. The overall goal is to help automate the development of the starting model for the full-waveform inversion.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2018
- Bibcode:
- 2018AGUFMNS31B0749J
- Keywords:
-
- 0545 Modeling;
- COMPUTATIONAL GEOPHYSICSDE: 3260 Inverse theory;
- MATHEMATICAL GEOPHYSICSDE: 3270 Time series analysis;
- MATHEMATICAL GEOPHYSICSDE: 7270 Tomography;
- SEISMOLOGY