11:00 AM - 11:15 AM
[STT42-06] Accounting for strong effects of sub-wavelength heterogeneities in full waveform inversion based on wavefield gradient measurements such as rotational sensors or DAS
Keywords:Distributed Acoustic Sensing, Seismic Tomography, Full waveform inversion, Rotational wavefield, Inversion
Seismic full waveform inversion (FWI) uses mostly translational motions including displacement, velocity, and acceleration both for numerical tests and real applications. Most FWI schemes implicitly use the fact that translational motions are insensitive to sub-wavelength medium heterogeneities. This makes it possible to design model parametrization and image resolution based on the maximum frequency of the inverted filtered data.
Recent advances in seismic sensors such as distributed acoustic sensing (DAS) and rotational sensors enable us to observe wavefield gradients (strain and rotation). As opposed to translational motions, wavefield gradients have sensitivity to localized sub-wavelength structures. Nevertheless, this subwavelength sensitivity is limited to the vicinity of the receivers.
In this work, we investigate if this different nature of wavefields sensitivity to small-scale heterogeneities could make difference in the resolution, accuracy, and inversion strategy of full waveform inversion. We show that the local sensitivity of wavefield gradients is more a difficulty for the FWI process rather than improving its spatial resolution. We find that full waveform inversion using wavefield gradients can lead to incorrect tomographic images due to the effects of sub-wavelength structures, even when accurate images can be obtained using displacement data for the same configuration. This finding suggests a new strategy for the full waveform inversion using wavefield gradients is necessary.
We propose a new inversion method to mitigate the effects of sub-wavelength structures based on the two-scale homogenization theory which has been recently introduced to seismology. Here, a two-step inversion is adopted. The first step is the inversion of local receiver correctors that are needed to account for the effects of sub-wavelength structures and to correct simulated data and adjoint sources. The second step is the inversion of model parameters. We show that this two-step inversion using wavefield gradients can successfully mitigate the small-scale effects. In addition, we conducted a statistical analysis of inverted model parameters for the evaluation of our two-step inversion. The results of the statistical analysis confirmed that our new method estimates the solution with the same accuracy as the method using displacement data and has a similar resolution compared to one using displacement.
Our study suggests that the effects of sub-wavelength heterogeneities need to be taken into account when full waveform inversion is performed for the data of DAS or rotational sensor measurements.
Acknowledgement
This work was supported by JSPS KAKENHI Grant Number 21J21871 and JSPS Overseas Challenge Program for Young Researchers.
Recent advances in seismic sensors such as distributed acoustic sensing (DAS) and rotational sensors enable us to observe wavefield gradients (strain and rotation). As opposed to translational motions, wavefield gradients have sensitivity to localized sub-wavelength structures. Nevertheless, this subwavelength sensitivity is limited to the vicinity of the receivers.
In this work, we investigate if this different nature of wavefields sensitivity to small-scale heterogeneities could make difference in the resolution, accuracy, and inversion strategy of full waveform inversion. We show that the local sensitivity of wavefield gradients is more a difficulty for the FWI process rather than improving its spatial resolution. We find that full waveform inversion using wavefield gradients can lead to incorrect tomographic images due to the effects of sub-wavelength structures, even when accurate images can be obtained using displacement data for the same configuration. This finding suggests a new strategy for the full waveform inversion using wavefield gradients is necessary.
We propose a new inversion method to mitigate the effects of sub-wavelength structures based on the two-scale homogenization theory which has been recently introduced to seismology. Here, a two-step inversion is adopted. The first step is the inversion of local receiver correctors that are needed to account for the effects of sub-wavelength structures and to correct simulated data and adjoint sources. The second step is the inversion of model parameters. We show that this two-step inversion using wavefield gradients can successfully mitigate the small-scale effects. In addition, we conducted a statistical analysis of inverted model parameters for the evaluation of our two-step inversion. The results of the statistical analysis confirmed that our new method estimates the solution with the same accuracy as the method using displacement data and has a similar resolution compared to one using displacement.
Our study suggests that the effects of sub-wavelength heterogeneities need to be taken into account when full waveform inversion is performed for the data of DAS or rotational sensor measurements.
Acknowledgement
This work was supported by JSPS KAKENHI Grant Number 21J21871 and JSPS Overseas Challenge Program for Young Researchers.