Presentation: Polynomial Equivalent Layer

Slides for the oral presentation "Polynomial Equivalent Layer" presented at the SEG International Exposition and Eighty-Second Annual Meeting in Las Vegas, Nevada.

ABSTRACT

We have developed a new cost-effective method for processing large-potential-field data sets via the equivalent-layer technique. In this approach, the equivalent layer is divided into a regular grid of equivalent-source windows. Inside each window, the physical-property distribution is described by a bivariate polynomial. Hence, the physical-property distribution within the equivalent layer is assumed to be a piecewise polynomial function defined on a set of equivalent-source windows.We perform any linear transformation of a large set of data as follows.
First, we estimate the polynomial coefficients of all equivalentsource windows by using a linear regularized inversion. Second, we transform the estimated polynomial coefficients of all windows
into the physical-property distribution within the whole equivalent layer. Finally, we premultiply this distribution by the matrix of Green’s functions associated with the desired transformation to obtain the transformed data. The regularized inversion deals with a linear system of equations with dimensions based on the total number of polynomial coefficients
within all equivalent-source windows. This contrasts with the classical approach of directly estimating the physical-property distribution within the equivalent layer, which leads to a system based on the number of data. Because the number of data is much larger than the number of polynomial coefficients, the proposed polynomial representation of the physical-property distribution within an equivalent layer drastically reduces the number of parameters to be estimated. By comparing the total number of floating-point operations required to estimate an equivalent layer via our method with the classical approach, both formulated with Cholesky’s decomposition, we can verify that the computation time required for building the linear system and for solving the linear inverse problem can be reduced by as many as three and four orders of magnitude, respectively. Applications to synthetic and real data show that our method performs the standard linear transformations of potential-field data accurately.