学术文献库

Modelling radar wave propagation in frequency domain is appealing in full waveform inversion because it allows decreasing the non-linearity of the problem, decreasing the dimension of the data space, better description of attenuation, and handling efficiently multiple sources. Besides, performing 2.5D modelling is interesting when physical properties can be assumed invariant in one horizontal dimension because it allows reducing drastically computation requirements compared to the 3D case. In 2.5D, finite-difference methods can be used to propagate the wave in two directions in space and a spatial Fourier transform is performed in the third direction to get a full three dimensional solution. With a simple central finite-difference implementation, second order accuracy in space is obtained and up to twenty grid points per wavelength are necessary to accurately simulate electromagnetic waves. Such a large number of grid points will impact on the storage requirement associated with frequency domain modelling. We propose a high accuracy algorithm to solve the frequency domain electromagnetic wave equation by finite-differences in 2.5D. The algorithm relies on a nine-point stencil to build weighted-averaging numerical operators. The weights are chosen to minimize numerical dispersion and anisotropy, which allows relaxing the requirements on grid cell size and thus decreases computational costs by a factor of about 3.6 compared to the central finite-difference method. This new algorithm reduces the numerical error without increasing the numerical bandwidth of the matrix system to solve, and can be easily transposed to 3D frequency domain modelling.