Machine Learning for Bayesian Experimental Design in the Subsurface

Thibaut, Robin

doi:10.22541/essoar.168057575.58936022/v1

Cited by 2 publications

(1 citation statement)

References 363 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Due to the expected non-linearity in the statistical relationship between seismic data and reservoir properties, [32] have used summary statistics extracted from unsupervised-and supervisedlearning approaches including discrete wavelet transform and a deep neural network combined with approximated Bayesian computation to derive a relationship. Similarly, [60] used a probabilistic Bayesian neural network to derive the relationship.…”

Section: Bel1dmentioning

confidence: 99%

Assessing and Improving the Robustness of Bayesian Evidential Learning in One Dimension for Inverting Time-Domain Electromagnetic Data: Introducing a New Threshold Procedure

Ahmed,

Aigner,

Michel

et al. 2024

Water

View full text Add to dashboard Cite

Understanding the subsurface is of prime importance for many geological and hydrogeological applications. Geophysical methods offer an economical alternative for investigating the subsurface compared to costly borehole investigations. However, geophysical results are commonly obtained through deterministic inversion of data whose solution is non-unique. Alternatively, stochastic inversions investigate the full uncertainty range of the obtained models, yet are computationally more expensive. In this research, we investigate the robustness of the recently introduced Bayesian evidential learning in one dimension (BEL1D) for the stochastic inversion of time-domain electromagnetic data (TDEM). First, we analyse the impact of the accuracy of the numerical forward solver on the posterior distribution, and derive a compromise between accuracy and computational time. We also introduce a threshold-rejection method based on the data misfit after the first iteration, circumventing the need for further BEL1D iterations. Moreover, we analyse the impact of the prior-model space on the results. We apply the new BEL1D with a threshold approach on field data collected in the Luy River catchment (Vietnam) to delineate saltwater intrusions. Our results show that the proper selection of time and space discretization is essential for limiting the computational cost while maintaining the accuracy of the posterior estimation. The selection of the prior distribution has a direct impact on fitting the observed data and is crucial for a realistic uncertainty quantification. The application of BEL1D for stochastic TDEM inversion is an efficient approach, as it allows us to estimate the uncertainty at a limited cost.

show abstract

Section: Bel1dmentioning

confidence: 99%

Assessing and Improving the Robustness of Bayesian Evidential Learning in One Dimension for Inverting Time-Domain Electromagnetic Data: Introducing a New Threshold Procedure

Ahmed,

Aigner,

Michel

et al. 2024

Water

View full text Add to dashboard Cite

show abstract

Assessing and improving the robustness of Bayesian evidential learning in one dimension for inverting TDEM data: introducing a new threshold procedure

Ahmed,

Aigner,

Michel

et al. 2024

Preprint

View full text Add to dashboard Cite

Understanding the subsurface is of prime importance for many geological and hydro-geological applications. Geophysical methods offer an economical alternative for in-vestigating the subsurface compared to costly borehole investigation methods, but geophysical results are commonly obtained through an inversion whose solution is non-unique. Deterministic inversions providing a unique solution are computationally efficient while stochastic inversions investigating the full uncertainty range are more expensive. In this research, we investigate the robustness of the recently introduced Bayesian evidential learning in one dimension (BEL1D) to stochastically invert time domain electromagnetic data (TDEM). In particular, we analyse the performance and accuracy of BEL1D when using the coarser discretization used for the computation of the forward solution using SimPEG. We demonstrate that it is possible to speed-up BEL1D by introducing a threshold rejection method on the data misfit to by-pass itera-tions. In addition, we discuss the impact of the prior model space on the results. Fi-nally, we apply the algorithm on field data collected in the Luy river catchment (Vi-etnam) to delineate saltwater intrusions. Our results show that the proper selection of timesteps and space discretization is essential to limit the computational cost while maintaining the accuracy of the posterior estimation. The selection of the prior distri-bution has a direct impact on fitting the observed data and is crucial to a realistic un-certainty quantification. The application of BEL1D for stochastic TDEM inversion is an efficient approach as it allows us to estimate the uncertainty at a limited cost.

show abstract

Machine Learning for Bayesian Experimental Design in the Subsurface

Abstract: To consult the statistician after an experiment is finished is often merely to ask him to conduct a post mortem examination. He can perhaps say what the experiment died of.

Cited by 2 publications

References 363 publications

Assessing and Improving the Robustness of Bayesian Evidential Learning in One Dimension for Inverting Time-Domain Electromagnetic Data: Introducing a New Threshold Procedure

Assessing and Improving the Robustness of Bayesian Evidential Learning in One Dimension for Inverting Time-Domain Electromagnetic Data: Introducing a New Threshold Procedure

Assessing and improving the robustness of Bayesian evidential learning in one dimension for inverting TDEM data: introducing a new threshold procedure

Contact Info

Product

Resources

About