Time reverse modeling of low-frequency tremor signals
Published at
ETH Zurich
Date of publication
15 July 2009
Abstract
This thesis investigates time reverse modeling as a numerical technique for detecting and locating sources of low-frequency (< 10 Hz) tremor signals in a complex subsurface. The idea of locating the spatial origin of signals with time reverse modeling has been adapted from existing studies by other authors in physics, medicine and seismology. This thesis develops time reverse modeling further as an application to sources of continuous tremor signals for which no identification of first arrival time or single event is possible.
The software used for numerical modeling was developed during this thesis. A computer code of full elastic wave propagation in 1D, 2D and 3D based on an explicit finite-difference algorithm has been programmed. This code has been constructed for application on a normal PC (personal computer). It is able to calculate forward and reverse modeling. Reverse modeling can be performed with synthetic signals from forward simulations as well as with field signals acquired by seismometers during field measurements. Prerequisites for the field signals used for reverse modeling are that they are complete in time, synchronous for multiple seismometers and measured with three-component seismometers. An external code for preparation of field data as input signals for numerical reverse modeling is added to the wave propagation code. The wave propagation code is further linked to a parallel code by Erik H. Saenger which allows the performance of highly resolved 2D or demanding 3D simulations distributed on multiple processors. A major topic of time reverse modeling research is the imaging condition as a tool to analyze and present the results. Analyzing individual time steps during reverse modeling is neither convenient nor representative for continuous tremor signals. The imaging condition has rather to cope with the time axis as an additional parameter together with space. An approach is to collapse the time axis in imaging the maximum absolute particle velocity per grid point throughout the entire time of reverse modeling. The highest values of this imaging condition correspond either to a high amplitude of one wave front or to the positive interference of signals reversely emitted by multiple seismometers. The latter is a possible indication of a seismic source as might be created by the simultaneous return of signals at their common source location. This imaging condition comprises both body waves (P-waves and S-waves) into one image. The independent behavior of the P-waves and the S-waves is imaged with their individual maximum energy densities as additional imaging conditions. Advantages and drawbacks of the individual imaging conditions are specifically addressed and discussed.
The time reverse modeling algorithm is applied to synthetic and field data for 2D and 3D geometry. Knowledge and experience gained from 2D applications are advanced and described in this thesis. Results from preliminary studies on 3D applications are summarized in the Appendix.
Synthetic studies with single or multiple sources in the subsurface are carried out to learn about the general focusing behavior of reversely propagated signals and the spatial detectability of seismic sources. In these studies, specific questions are addressed, for example, to signals from different types of sources, to different S/N-ratios (signal to noise ratios) and to a varying Poisson's ratio for the individual P-wave and S-wave velocity models. The synthetic studies reveal the general applicability of time reverse modeling for detecting and locating sources of continuous low-frequency seismic signals. Furthermore, they deliver essential information how to deal with time reverse modeling and how to interpret results, as they point out the limitations and some artifacts of this technique.
Field signals were acquired by measurements distributed over different surveys above two well known hydrocarbon reservoirs in Voitsdorf, Austria. A hypothesis states that hydrocarbon reservoirs may be locations of seismic sources which emit low-frequency tremor-like signals. This hypothesis is surveyed by reverse modeling of recorded low-frequency signals without specifically questioning the mechanical background of possible signal emission. The studies indicate, for the specific area in Voitsdorf, possible presence of sources of low-frequency tremor signals in the vicinity of the hydrocarbon reservoirs. However, some open questions require further investigation before a conclusive evaluation of the hypothesis can be made.
Integral aspects of the application of time reverse modeling on field data are, in addition to modeling, pre-processing of the input signals and post-processing of the output image. Pre-processing consists of collecting synchronous seismograms, band-pass filtering of the data for the low-frequency range and resampling of the time axis to the time sample required for numerical modeling. The input signals, furthermore, have to be revised for possible external disturbances or internal distortion by the seismometers themselves. For example, a short-time strong amplitude which is very much likely not related to a source in the subsurface has to be removed from the seismogram as it may remarkably lower the quality of the result produced by reverse modeling. Post-processing mainly covers adjustment of the color scale of the image to visualize possible source locations. Post-processing also aims to improve multiple results for the same area in statistically balancing randomly distributed side effects, e.g. by stacking of the images.
Although the development and the application of time reverse modeling have been performed and presented in a general way, the examples and the findings are transferable to closely related situations in Earth sciences, engineering or other scientific fields dealing with signal recording and processing. Examples of possible continuative applications might be the localization of volcanic tremors or the tracking of acoustic emissions caused by strain energy release in construction.
The software used for numerical modeling was developed during this thesis. A computer code of full elastic wave propagation in 1D, 2D and 3D based on an explicit finite-difference algorithm has been programmed. This code has been constructed for application on a normal PC (personal computer). It is able to calculate forward and reverse modeling. Reverse modeling can be performed with synthetic signals from forward simulations as well as with field signals acquired by seismometers during field measurements. Prerequisites for the field signals used for reverse modeling are that they are complete in time, synchronous for multiple seismometers and measured with three-component seismometers. An external code for preparation of field data as input signals for numerical reverse modeling is added to the wave propagation code. The wave propagation code is further linked to a parallel code by Erik H. Saenger which allows the performance of highly resolved 2D or demanding 3D simulations distributed on multiple processors. A major topic of time reverse modeling research is the imaging condition as a tool to analyze and present the results. Analyzing individual time steps during reverse modeling is neither convenient nor representative for continuous tremor signals. The imaging condition has rather to cope with the time axis as an additional parameter together with space. An approach is to collapse the time axis in imaging the maximum absolute particle velocity per grid point throughout the entire time of reverse modeling. The highest values of this imaging condition correspond either to a high amplitude of one wave front or to the positive interference of signals reversely emitted by multiple seismometers. The latter is a possible indication of a seismic source as might be created by the simultaneous return of signals at their common source location. This imaging condition comprises both body waves (P-waves and S-waves) into one image. The independent behavior of the P-waves and the S-waves is imaged with their individual maximum energy densities as additional imaging conditions. Advantages and drawbacks of the individual imaging conditions are specifically addressed and discussed.
The time reverse modeling algorithm is applied to synthetic and field data for 2D and 3D geometry. Knowledge and experience gained from 2D applications are advanced and described in this thesis. Results from preliminary studies on 3D applications are summarized in the Appendix.
Synthetic studies with single or multiple sources in the subsurface are carried out to learn about the general focusing behavior of reversely propagated signals and the spatial detectability of seismic sources. In these studies, specific questions are addressed, for example, to signals from different types of sources, to different S/N-ratios (signal to noise ratios) and to a varying Poisson's ratio for the individual P-wave and S-wave velocity models. The synthetic studies reveal the general applicability of time reverse modeling for detecting and locating sources of continuous low-frequency seismic signals. Furthermore, they deliver essential information how to deal with time reverse modeling and how to interpret results, as they point out the limitations and some artifacts of this technique.
Field signals were acquired by measurements distributed over different surveys above two well known hydrocarbon reservoirs in Voitsdorf, Austria. A hypothesis states that hydrocarbon reservoirs may be locations of seismic sources which emit low-frequency tremor-like signals. This hypothesis is surveyed by reverse modeling of recorded low-frequency signals without specifically questioning the mechanical background of possible signal emission. The studies indicate, for the specific area in Voitsdorf, possible presence of sources of low-frequency tremor signals in the vicinity of the hydrocarbon reservoirs. However, some open questions require further investigation before a conclusive evaluation of the hypothesis can be made.
Integral aspects of the application of time reverse modeling on field data are, in addition to modeling, pre-processing of the input signals and post-processing of the output image. Pre-processing consists of collecting synchronous seismograms, band-pass filtering of the data for the low-frequency range and resampling of the time axis to the time sample required for numerical modeling. The input signals, furthermore, have to be revised for possible external disturbances or internal distortion by the seismometers themselves. For example, a short-time strong amplitude which is very much likely not related to a source in the subsurface has to be removed from the seismogram as it may remarkably lower the quality of the result produced by reverse modeling. Post-processing mainly covers adjustment of the color scale of the image to visualize possible source locations. Post-processing also aims to improve multiple results for the same area in statistically balancing randomly distributed side effects, e.g. by stacking of the images.
Although the development and the application of time reverse modeling have been performed and presented in a general way, the examples and the findings are transferable to closely related situations in Earth sciences, engineering or other scientific fields dealing with signal recording and processing. Examples of possible continuative applications might be the localization of volcanic tremors or the tracking of acoustic emissions caused by strain energy release in construction.
