Quantifying Fault-Related Uncertainty in Full Waveform Inversion with Inverse Homogenization

in: 86th EAGE Annual Conference \& Exhibition, pages 1-5, European Association of Geoscientists \& Engineers

Abstract

It is well recognized that solving inverse problems for a single model is insufficient for quantitative analysis and decision making in subsurface applications. By providing a distribution of possible outcomes rather than a single estimate, stochastic inverse modeling approaches incorporate and quantify uncertainty on the model parameters of interest. Despite advantages in handling uncertainties, these approaches are not yet standard practice in seismic inversion, especially for large-scale full-waveform inversion (FWI) applications, due to the high computational cost of waveform modeling. In practice, FWI is solved with a local optimization approach providing, at best, a single representation of the real Earth complex structures. However, FWI based on frequency band-limited seismic data can only recover a smooth version of the true Earth, which is not suited for a proper geological interpretation below the decametric scale. To address this problem, we propose the use of the inverse homogenization, or downscaling inversion, to interpret the inverted large-scale model as a probability distribution over possible fine-scale models, thus reducing structural interpretation uncertainty. The methodology is validated on a synthetic two-dimensional test case, aimed at recovering the fault-related parameters and the distribution at the fine scale of the velocity within the model.

Download / Links

BibTeX Reference

@inproceedings{ruggiero:hal-05156984,
 abstract = {It is well recognized that solving inverse problems for a single model is insufficient for quantitative analysis and decision making in subsurface applications. By providing a distribution of possible outcomes rather than a single estimate, stochastic inverse modeling approaches incorporate and quantify uncertainty on the model parameters of interest. Despite advantages in handling uncertainties, these approaches are not yet standard practice in seismic inversion, especially for large-scale full-waveform inversion (FWI) applications, due to the high computational cost of waveform modeling. In practice, FWI is solved with a local optimization approach providing, at best, a single representation of the real Earth complex structures. However, FWI based on frequency band-limited seismic data can only recover a smooth version of the true Earth, which is not suited for a proper geological interpretation below the decametric scale. To address this problem, we propose the use of the inverse homogenization, or downscaling inversion, to interpret the inverted large-scale model as a probability distribution over possible fine-scale models, thus reducing structural interpretation uncertainty. The methodology is validated on a synthetic two-dimensional test case, aimed at recovering the fault-related parameters and the distribution at the fine scale of the velocity within the model.},
 address = {Toulouse, France},
 author = {Ruggiero, Giusi and Cupillard, Paul and Caumon, Guillaume},
 booktitle = {{86th EAGE Annual Conference \& Exhibition}},
 doi = {10.3997/2214-4609.2025101301},
 hal_id = {hal-05156984},
 hal_version = {v1},
 month = {June},
 pages = {1-5},
 publisher = {{European Association of Geoscientists \& Engineers}},
 title = {{Quantifying Fault-Related Uncertainty in Full Waveform Inversion with Inverse Homogenization}},
 url = {https://hal.univ-lorraine.fr/hal-05156984},
 volume = {2025},
 year = {2025}
}