Modeling of the Acoustic Wave Equation by Staggered‐Grid Finite‐Difference Schemes with High‐Order Temporal and Spatial Accuracy

Zhiming Ren,Zhenchun Li,Yang Liu,Mrinal K. Sen

DOI: https://doi.org/10.1785/0120170068

2017-01-01

Abstract:Finite difference (FD) is widely used for modeling seismic-wave propagation in theoretical and applied seismology. Traditional FD methods with the cross stencil exhibit spatial general-order and temporal second-order accuracy, causing strong temporal dispersion and poor stability. We propose two novel time-space domain staggered-grid FD (SFD) schemes, in which the spatial derivatives are approximated through a weighted summation of function values from the axial and some off-axial nodes. We derive the dispersion relation of acoustic wave equation and estimate FD coefficients by the Taylor series expansion (TE). The TE-based SFD scheme with the new stencil can obtain (2N) th-order accuracy in both time and space domains along all the propagation directions for N <= 4. In contrast, it reaches (2N) th-order accuracy along four propagation directions and eighth-order accuracy along other propagation directions for N > 4. Furthermore, we investigate an SFD scheme using a combination of TE and least squares (LS). FD coefficients at the off-axial nodes are computed by TE, and then FD coefficients at the axial nodes are optimized by LS. We compared our schemes with the conventional, the temporal fourth-and sixth-order, and several optimal SFD schemes. Dispersion analysis, stability analysis, and 2D and 3D numerical examples show that our TE-based SFD scheme has smaller temporal dispersion and better stability than other schemes. Besides, our TE+LS-based SFD scheme can further suppress spatial dispersion with slightly degraded stability. We also analyze the efficiency of different schemes. Our new SFD schemes allow for shorter operator lengths and larger time steps owing to higher accuracy and less stringent stability conditions. Numerical results reveal that the computational cost of our TE+LS-based scheme is 49.1% and 85.7% of that of the temporal fourth-order and optimal SFD schemes, respectively, under the same accuracy demand.