Legendre polynomial-based robust Fourier transformation and its use in reduction to the pole of magnetic data

Abstract

A new inversion based Fourier transformation technique named as Legendre-Polynomials Least-Squares Fourier Transformation (L-LSQ-FT) and Legendre-Polynomials Iteratively Reweighted Least-Squares Fourier Transformation (L-IRLS-FT) are presented. The introduced L-LSQ-FT algorithm establishes an overdetermined inverse problem from the Fourier transform. The spectrum was approximated by a series expansion limited to a finite number of terms, and the solution of inverse problem, which gives the values of series expansion coefficients, was obtained by the LSQ method. Practically, results from the least square method are responsive to data outliers, thus scattered large errors and the estimated model values may be far from reality. A definitely better option is attained by introducing Steiner’s Most Frequent Value method. By combining the IRLS algorithm with Cauchy-Steiner weights, the Fourier transformation process was robustified to give the L-IRLS-FT method. In both cases Legendre polynomials were applied as basis functions. Thus the approximation of the continuous Fourier spectra is given by a finite series of Legendre polynomials and their coefficients. The series expansion coefficients were obtained as a solution to an overdetermined non-linear inverse problem. The traditional DFT and the L-IRLS-FT were tested numerically using synthetic datasets as well as field magnetic data. The resulting images clearly show the reduced sensitivity of the newly developed L-IRLS-FT methods to outliers and scattered noise compared to the traditional DFT. Conclusively, the newly developed L-IRLS-FT can be considered to be a better alternative to the traditional DFT.