Answers are grounded in this paper and cite their sources.
Summary Internal multiple elimination (IME) is a challenging task in seismic exploration. If internal multiples (IMs) are not eliminated, they may be misinterpreted as primaries, leading to imaging artifacts. Among the few available methods, Marchenko multiple elimination (MME) is the most appealing, as it is data-driven and requires no model information. However, the computational cost is considerable due to the extensive calculations of multidimensional convolution and correlation, which include forward and inverse fast Fourier transforms among shot gathers. Moreover, its application to field data remains limited, indicating that more attention should be paid to practical issues. Starting from the PyMarchenko MME, a GPU-accelerated IME technique, GPU-MME, to alleviate the computational burden of MME is developed. GPU-MME faithfully replicates the PyMarchenko MME results and delivers a significant acceleration—up to nearly two orders of magnitude faster—due to the GPU’s efficiency in handling the key computations involved in MME. GPU-MME is applied to a series of synthetic datasets and an open field dataset to demonstrate its effectiveness and practical value. The comparison results are either provided or reproduced from existing work. GPU-MME is competitive with a state-of-the-art MME method. The impact of key parameters in GPU-MME and how to determine their optimal values are investigated. When accurate surface-free and wavelet-free reflection response is unavailable, an estimation of multiple and source (EMS) technique can successfully eliminate surface-related multiples (SRMs) and provide the surface-free, wavelet-free input for GPU-MME. However, additional scaling of reflection response and adaptive adjustment of the initially estimated IMs may be necessary to ensure effective IME. The scaling and adjustment procedures are discussed in detail, as they are critical when an exact reflection response is not available. For synthetic data, the input can be modeled perfectly accurate, so a discussion for these cases is unnecessary. The EMS and GPU-MME techniques can work together in theory and in practice, and it is hoped that they will be widely adopted by industry for the effective estimation of both SRMs and IMs.
No takes yet. Share an insight, caveat, or question.
Shao et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: