Adaptive parameterization of electric conductivity in inversion of electromagnetic data

Abstract

We describe a methodology developed for 3-D parameter identification, with focus on large-scale applications such as monitoring subsea oil production and geothermal systems. The methodology is designed to handle challenges related to low parameter sensitivity, nonuniqueness of the inverse solutions and costly numerical calculations. A reduced, composite parameter representation is chosen to meet these challenges. Our contributions to the methodology involves choosing a reduced representation with radial basis functions, to maintain a low number of parameters. We also propose the use of a 1. order selection measure in the refinement process to reduce the computational costs. The performance of the proposed changes in the methodology is illustrated in a series of examples for estimating the change in electric conductivity from time-lapse electromagnetic observations. The results show some limitetions regarding the accuracy of the 1. order selection measure. For the investigated numerical examples, radial basis functions, together with the described methodology, effectively provide an estimated of the electric conductivity field using electromagnetic measurements.