Clever Data Trick Doubles the Sharpness of Deep Seismic Images in South China

Beneath the rolling hills and steep mountains of southern China lies one of the most geologically consequential suture zones on Earth, a belt of crushed and folded crust where two ancient continental blocks collided nearly a billion years ago. Reading that buried history, however, requires seismic images of extraordinary clarity, and for years the pictures coming back from the Jiangnan Orogenic Belt were frustratingly blurry. Now a team of Chinese geoscientists has shown that a relatively simple computational fix, applied at exactly the right stage of processing, can transform those murmurings into some of the sharpest deep crustal images ever produced in the region, with direct implications for finding copper, tungsten, and gold.

The study, published in the journal Solid Earth, was led by Hui Zhang and Jiayong Yan of the Chinese Academy of Geological Sciences, working from the State Key Laboratory of Deep Earth and Mineral Exploration. The team revisited a 200-kilometer-long deep seismic reflection profile acquired in 2019 along the Wuning to Ji’an section of the central Jiangnan Orogenic Belt, a crooked survey line that snakes from Jiugong Mountain in the north, across the Xiushui-Wuning Basin and the Jiuling Upheaval, and down through the Pingle Depression toward the Wugongshan Upheaval. Their central finding is deceptively simple: before the data are migrated into a subsurface image, the irregularly spaced explosive shots used to generate the seismic waves should be computationally repositioned and interpolated onto a dense, regular grid at 100-meter intervals.

Why does irregularity matter so much? Deep seismic reflection profiling works by detonating charges in boreholes and recording the echoes that return from rock layers as far as 30 to 100 kilometers beneath the surface. In an ideal world, shots and receivers would sit at perfectly uniform intervals along a straight line, allowing every reflected wave to be summed coherently. In the real world of the Jiangnan Belt, the northern section climbs through Jiuling Mountain, where the relative elevation difference reaches 1,000 meters and slopes commonly exceed 45 degrees, making it nearly impossible to position drilling rigs exactly where the survey design demands. The southern section, by contrast, crosses densely populated plains and hills studded with towns, villages, and factories, forcing further deviations from the plan.

The consequences of those deviations ripple through the entire processing chain. When shots and receivers sit at irregular positions, the common midpoint points where reflected waves converge scatter across a two-dimensional plane rather than lining up neatly along the profile. Uneven offset distributions, large gaps between traces, and spatial aliasing, in which steeply dipping wave energy masquerades as false signal, all follow. When such data are migrated, the process that repositions reflections to their true subsurface locations, the irregularities generate migration artifacts, phantom structures that can be mistaken for real faults or rock bodies. Previous contractor-processed versions of the Jiangnan data showed exactly these problems: low signal-to-noise ratios overall, and no clear reflections from the major faults that were the survey’s principal targets.

Zhang and colleagues attacked the problem with an anti-aliasing version of a technique called Matching Pursuit Fourier Interpolation, or MPFI. The method converts irregular seismic data from the space-time domain into the frequency-wavenumber domain using Fourier transforms, then iteratively identifies the strongest coherent spectral components and rebuilds the missing data from them. The crucial innovation is the use of priors: the energy of the low-frequency part of the signal, typically the 15 to 35 hertz band in this study, is analyzed at various dips, and those energy curves are used as weighting factors for the higher frequencies up to 100 hertz. Because genuine reflections and aliasing noise behave differently across frequency, the priors allow the algorithm to distinguish real signal from aliased energy that would otherwise be wrongly selected, a failure mode that has long plagued conventional frequency-domain interpolation of sparse land data.

The team tested four regularization strategies on the Jiangnan dataset. The first simply filled gaps in the offset domain while leaving shot positions irregular, raising the fold, the number of traces summed at each point, from 80 to roughly 100 and producing moderate improvement. The second fully regularized all data in the offset domain, restoring common midpoint locations to their designed positions. The third repositioned irregular shots onto their pre-designed 200-meter grid in the shot domain. The fourth went further, regularizing the shots and then interpolating entirely new shot gathers to halve the spacing to 100 meters, doubling the fold to about 160. Quantitative analysis of power spectra, signal-to-noise estimates, and variance attributes showed a clear hierarchy, with the shot-domain methods consistently outperforming the offset-domain approaches.

The winning strategy, regularizing and infilling shot gathers at 100-meter intervals, raised the signal-to-noise ratio from approximately 1.5 to 2.5 in the shallow profiles, covering time windows of 0.5 to 6.5 seconds, and from roughly 2.0 to 3.5 in the deep sections between 5.5 and 11.5 seconds. Importantly, the effective frequency bandwidth remained essentially unchanged, meaning the gains came from genuine noise suppression and better coherence rather than artificial sharpening. The authors attribute the superiority of shot-domain processing to its physical logic: while offset-domain regularization merely shifts common midpoint positions and drags the underlying shot and receiver coordinates along with them, shot-domain regularization moves the actual shot points directly to their designed acquisition coordinates, a more faithful reconstruction of what the survey was meant to record.

The payoff became most dramatic in the pre-stack time migration images. Compared with the legacy stack profiles produced independently by two contractors, one processed in the field and one in-house, the newly processed sections revealed far clearer images of the region’s two dominant structures: the Yifeng-Jingdezhen Fault, a nearly vertical break extending westward into Hunan, and the Pingxiang-Guangfeng Fault, a crust-scale, northward-dipping fault that cuts through the crustal detachment surface down to the Moho and offsets it. The latter, identified as the southern boundary of the Qin-Hang belt and the northern boundary of the South China Block, is thought to act as a critical conduit for magma rising from depth. Major faults and their secondary fractures, the authors note, provided pathways for deep-source magmas and ore-forming fluids, and their reactivation promoted mineralization, which is precisely why sharper images of them matter for exploration.

The method does have limits, and the team is candid about them. Improvements were slightly weaker in the deepest sections, where signal attenuation with depth weakens the effective reflections that regularization depends on. The technique also performs best on crooked profiles whose lateral deviation stays within a quarter of the Fresnel radius of the target reflectors, a geophysical yardstick that ranges from roughly 100 to 700 meters for shallow targets and expands to 700 to 1,750 meters for deep crustal reflectors at 15 to 30 kilometers. Beyond that threshold, regularization risks distorting the wavefield and erasing small-scale geological features. Amplitude analysis at the survey line’s two major corners, with curvature angles of 19 and 4 degrees, confirmed that relative amplitude trends with offset were preserved, keeping amplitude-based interpretations trustworthy.

For a field that has long focused its innovation on velocity analysis and migration algorithms, the message of this study is that the humble step of data regularization, performed in the right domain before migration, can matter just as much. The work was supported by China’s National Science and Technology Major Project for Deep Earth Probe and Mineral Resources Exploration, the National Natural Science Foundation of China, and the China Geological Survey’s Deep Geological Survey Project. As demand grows for reliable images of the deep crust, both to reconstruct ancient supercontinents like Rodinia and to guide the hunt for buried metal deposits, the Jiangnan results suggest that the sharpest views of Earth’s interior may come not from new hardware, but from treating the data we already have with considerably more mathematical care.

Subject of Research: Shot-domain regularization of deep seismic reflection data in the Jiangnan Orogenic Belt, South China

Article Title: Enhancing 2D deep seismic reflection imaging using shot domain regularization: a case study from the Jiangnan Orogenic Belt, South China

Article References: Zhang, H., Yan, J., Liu, Z., Han, J., Wang, H., & Liu, J. (2026). Enhancing 2D deep seismic reflection imaging using shot domain regularization: a case study from the Jiangnan Orogenic Belt, South China. Solid Earth, 17(4), 689-709. https://doi.org/10.5194/se-17-689-2026

Image Credits: AI Generated

DOI: 10.5194/se-17-689-2026

Keywords: deep seismic reflection, seismic data regularization, Matching Pursuit Fourier Interpolation, Jiangnan Orogenic Belt, South China, signal-to-noise ratio, pre-stack time migration, mineral exploration, crustal structure, shot domain interpolation, tectonics, geophysics