Interpretation of EM Sounding Data for a Stratified Earth by Means of Non-Linear Regression

Abstract

Direct inversion methods for electromagnetic soundings have not been used widely for routine interpretation of field data, even though they provide more useful information to the interpreter than do indirect methods, because existing inversion routines require large amounts of computer time and are relatively difficult to use. In an attempt to make direct inverse methods accessible for routine use, a study was made of the various available methods and a new computer program was written. This program solves for conductivity on a layering specified by the user, and uses the Fast Hankel transform with Anderson’s (1979) filter coefficients, a quintic spline approximation, and improvements in the Marquardt-Fletcher nonlinear regression algorithm to reduce the computer time required for convergence. The routine was tested with numerically generated sounding data which includes some random noise, as well as with several sets of field data. The results indicate that the routine solves easily for simple structures with moderate (≤ 10:1) conductivity contrasts, while larger contrasts in the conductivity structure require some care in setting up the starting model. Resistive structures are not resolved well, but the depth to conductive features in the stratigraphy can be determined accurately. Results from the inversion of five sets of field data are given: they were found to compare well with geophysical information from well logs or with interpretations made by other means.