Finite-difference two-dimensional correction filters were designed by least squares optimisation in the frequency domain. It is shown how these spatial convolution filters improve responses within a specified frequency range, with the constraint of a limited size. Examples are used to show the improved modelling results, with displays in the wavenumber and spatial domains. The superiority of this method is also shown by a direct comparison with the established Levander method, which uses a split step and fourth order accurate spatial derivatives.
View full article as PDF (0.50 Mb)