Improved Parker's Method for Topographic Models Using Chebyshev Series and Low Rank Approximation

Leyuan Wu,Qiang Lin

DOI: https://doi.org/10.1093/gji/ggx093

IF: 3.352

2017-01-01

Geophysical Journal International

Abstract:We present a new method to improve the convergence of the well-known Parker's formula for the modelling of gravity and magnetic fields caused by sources with complex topography. In the original Parker's formula, two approximations are made, which may cause considerable numerical errors and instabilities: (1) the approximation of the forward and inverse continuous Fourier transforms using their discrete counterparts, the forward and inverse Fast Fourier Transform (FFT) algorithms; (2) the approximation of the exponential function with its Taylor series expansion. In a previous paper of ours, we have made an effort addressing the first problem by applying the Gauss-FFT method instead of the standard FFT algorithm. The new Gauss-FFT based method shows improved numerical efficiency and agrees well with space-domain analytical or hybrid analytical-numerical algorithms. However, even under the simplifying assumption of a calculation surface being a level plane above all topographic sources, the method may still fail or become inaccurate under certain circumstances. When the peaks of the topography approach the observation surface too closely, the number of terms of the Taylor series expansion needed to reach a suitable precision becomes large and slows the calculation. We show in this paper that this problem is caused by the second approximation mentioned above, and it is due to the convergence property of the Taylor series expansion that the algorithm becomes inaccurate for certain topographic models with large amplitudes. Based on this observation, we present a modified Parker's method using low rank approximation of the exponential function in virtue of the Chebfun software system. In this way, the optimal rate of convergence is achieved. Some pre-computation is needed but will not cause significant computational overheads. Synthetic and real model tests show that the method now works well for almost any practical topographic model, provided that the assumption, that the entire topographic mass lies below the observation surface, is met.