Uncertainty quantification for regularized inversion of electromagnetic geophysical data - Part II: application in 1-D and 2-D problems

This paper is Part II of a two-part series on a mathematical and computational framework for computing a meaningful uncertainty quantification (UQ) for regularized inversions of electromagnetic data. In Part I, we explained the theory behind a sampling algorithm, which we call RTO-TKO, and in Part II, we showcase RTO-TKO in practice. We individually and jointly invert seafloor magnetotelluric (MT) and surface-towed controlled source electromagnetic field data, collected for imaging offshore freshened groundwater beneath the U.S. Atlantic margin. We also invert seafloor MT data collected for subsalt imaging to produce 2-D resistivity models and uncertainty estimates that characterize the salt body geometry and surrounding sediments. We compare the UQ of the RTO-TKO with results from trans-dimensional sampling, and explain the differences arising from different underlying (prior) assumptions of the two algorithms. We also discuss the practical implications of these findings. Most importantly, however, the 2-D case study unambiguously demonstrates the computational advantages of RTO-TKO and its ability to make use of massive parallelism.

Automated summary

This paper is about a mathematical and computational framework for computing uncertainty quantification in regularized inversions of electromagnetic data. It showcases the practical application of the RTO-TKO sampling algorithm and demonstrates its computational advantages and ability to utilize massive parallelism through case studies of seafloor magnetotelluric (MT) and surface-towed controlled source electromagnetic data inversion. The paper also compares the uncertainty quantification of RTO-TKO with results from trans-dimensional sampling and discusses the practical implications of the findings.