Hybrid dispersion relation preserving finite difference approach to infrasound propagation modeling on staggered grids

Infrasound is a key component of the geophysical monitoring network that detects large-scale artificial and natural events such as nuclear tests, earthquakes, volcanic eruptions, and fireballs. In this context, full waveform propagation modeling at regional and local distances enables more sophisticated identification of source properties. However, infrasound propagation modeling often suffers from heavy computation due to the dense grid nodes to avoid the numerical dispersion. This study proposes a hybrid dispersion relation preserving (DRP) finite difference to efficiently implement modeling on staggered grid. DRP offers a trade-off between numerical dispersion and accuracy by optimizing finite difference coefficients through least squares fitting within a given wavenumber range. We modify the previous staggered grid DRP schemes to ensure that dispersion error is evenly distributed within the designated wavenumber domain. Then, this is applied to the collocated grid as well, so that advection terms in infrasound governing equations can be handled accordingly. We establish the relationship between the cutoff wavenumber in DRP and minimum points per wavelength (PPW) for modeling, so this relationship suggests minimum PPW required for each finite difference order. Numerical simulations demonstrate that the proposed hybrid DRP outperforms the traditional finite difference method of the same order, particularly in suppressing numerical dispersion. Our modeling is 2D modeling in Cartesian coordinates and is associated with a line source. Therefore, attenuation by geometrical spreading is smaller than that of point source observations. To address this, a line-source to point-source transformation filter is applied to compensate for the attenuation difference, allowing for a direct comparison with observed infrasound signals. The processed synthetic signal shows good agreement with acoustic explosion models, such as Kinney and Graham (1985) model. Lastly, 155 mm artillery acoustic signals are experimentally acquired at distances of 200 m, 1, 3, 5, 10 km, and the source time function was estimated from the recording at 200 m. The synthetic results show a good match with observations from 1 to 10 km, proving that proposed modeling is capable of identifying source properties.