1.   Introduction

Figure  2.  Legacy seismic section from a NAZ acquisition

下载: 全尺寸图片 幻灯片

The loess plateau, situated in central-northern China, presents distinctive geomorphic attributes marked by massive loess accumulation (50−300 m thickness), highly dissected landscapes, and abrupt slope variations. As illustrated in Fig. 1, elevation fluctuations within a representative survey area range dramatically between 845 m and 1322 m. The compounded effects of extreme surface irregularity and unconsolidated loess strata substantially amplify seismic exploration complexities. Critical technical barriers extend beyond field acquisition to encompass reliable subsurface velocity modeling and high-fidelity imaging^[4]. Hydrocarbon resource assessments indicate that 95% of Ordos Basin’s reserves are concentrated beneath this loess-dominated terrain^[5]. predominantly hosted in low-porosity, tight clastic reservoirs^[6].

Figure  1.  Elevation map of typical loess plateau

下载: 全尺寸图片 幻灯片

3D seismic survey launched in the early 2000 s employed acquisition geometries featuring restricted offsets, limited fold of coverage, narrow azimuth, and extended receiver arrays. While these configurations partially mitigated operational uncertainties, resultant time-domain images proved inadequate for reservoir delineation owing to compromised signal integrity and spatial resolution.

Two decades of technological evolution in seismic instrumentation, computational infrastructure, and advanced processing workflows have catalyzed transformative innovations in loess exploration methodologies^[7]. This study comprehensively reviews cutting-edge acquisition systems, evaluates their data processing implications, develops objective-driven seismic enhancement protocols, validates outcomes through imaging case studies, and concludes with strategic recommendations for future exploration.

2.   Fully nodal seismometer seismic survey with high fold of coverage

In order to achieve thin reservoir prediction and deep small-scale karst reservoir detection, high-quality 3D seismic acquisition technology targeted at complex and thick loess plateau area were put forward and carried out 10 years ago^[8].

Based on comprehensive analysis of both geological exploration requirements and geophysical effects, an optimized acquisition scheme was proposed and implemented. The new scheme is characterized by small shot array, appropriate bin size, full azimuth, and high fold of coverage. Detailed acquisition parameters are listed in Table 1 below.

Table  1.  Acquisition Parameters

Acquisition Geometry	40L5S228T760F
Fold of Coverage	760
Number of Traces	9120
Receiver Line Spacing	200 m
Shot Line Spacing	120 m
Trace Spacing	40 m
Shot Point Spacing	80 m
CMP Grid	20 m (Crossline) × 40 m (Inline)
Min/Max Offset	28. 28 m/ 6037. 55 m

下载: 导出CSV 

| 显示表格

The adoption of key technologies, including vibroseis-dynamite joint broadband shooting and fully nodal seismometer applications, makes it possible to generate seismic data with high shot/trace density, high fold of coverage, and high aspect ratio. On the other hand, the dramatically irregular source locations caused by complex topography lead to strong spatial data inconsistency in both amplitude and wavelet, along with extremely poor signal-to-noise ratio (SNR), as shown in Fig. 3. These data features add to the signal processing difficulty, especially in statics, noise attenuation, surface consistent processing and imaging.

Figure  3.  Raw shots of the acquired data: (left)-dynamite, (right) vibroseis acquisition

下载: 全尺寸图片 幻灯片

For the acquired seismic data, processing difficulties caused by regional geological conditions remain. In the meanwhile, the extremely low SNR and data inconsistency give rise to additional challenges. After analyzing the new data and reviewing previous research^[9]. the key processing issues are summarized in Fig. 4, including first break picking and static correction on low SNR data, vertical and spatial inconsistency issues, and high-precision true surface modeling.

Figure  4.  Summarized challenges of data processing acquisition

下载: 全尺寸图片 幻灯片

3.   Data processing technologies for low SNR seismic data

To address these prominent issues of the new data, a complete set of characteristic high-precision static correction and imaging technologies targeted at seismic data in the thick loess plateau area was formed and applied. The key processing strategies and technologies are listed in Fig. 5.

Figure  5.  Key processing technologies and strategies acquisition

下载: 全尺寸图片 幻灯片

3.1   Signal processing leveraging AI first break picking and improved technologies

(1) Statics Strategy

The refraction statics calculation itself is not difficult, but the crux of statics lies in appropriate first break picking. Precise static correction can only be achieved when accurate first break information is available. The statics issue in this study is properly solved due to the following two factors.

● Iterative AI first break picking based on seed shots

We found it difficult for common commercial first arrival picking algorithms to determine first breaks accurately owing to the distorted first arrival and its weak energy. An iterative AI first break picking based on seed shots was innovated according. The sparse seed shots were selected, and their first breaks were manually picked first. AI first break picking then runs, with a first round statics generated afterwards. Another round of AI first break picking was implemented on the data with the first round statics applied. This process was iterated several times until satisfying statics was derived.

● Updated idea on statics solving strategy

Although the proposed first break picking strategy is effectively implemented, the picking quality is still poor, especially for some vibroseis shots on far offsets. The current picking accuracy is significantly less than the statistical accuracy of conventional dynamite onshore seismic data, which is generally greater than 80%. Considering the specific characteristics of the current data: small shots areal, fully nodal seismometer acquisition, high fold of coverage, and vibroseis-dynamite joint shooting, we propose a new threshold of good quality picks as 60% for this data type. This threshold was obtained after detailed tests. We believe the sufficient data volume would compensate for some of the deficiency of the picking accuracy to a certain degree. The updated idea helps free us from the dilemma of statics issue for extremely low SNR data. Fig. 6 shows the statics correction effects for a typical dynamite shot on the tableland and a vibroseis shot in the ravine.

Figure  6.  Static correction result of different shot gathers acquisition

下载: 全尺寸图片 幻灯片

(2) Improved signal processing ideas and techniques

To improve the SNR of the data, a set of improved signal processing workflows and technologies are designed.

● Innovative approach for target amplitude recovery

Apart from the complex and thick loess layer that result to strong signal absorption, the highly developed Mesozoic coal seams could generate even more severe energy shielding effects on transmission energy, which brings uncertainty to the description of Paleozoic reservoirs, especially in amplitude fidelity. In response to the thickness change and strong shielding effect of the buried coal seams, a surface consistency compensation approach was innovated. The expected outputs of target amplitude and frequency were constructed first. Amplitude compensation factors to compensate for the shielding effect caused by overlying strata were designed^[10-11] and applied to achieve true amplitude recovery under coal seams. The RMS amplitude attribute of the target layer indicates a significant improvement in amplitude preservation after the shielding compensation factor was applied, as shown in Fig. 7.

Figure  7.  RMS attribute QC before (a) and after (b) amplitude recovery factor applied

下载: 全尺寸图片 幻灯片

● 5D data regularization to achieve uniform fold and azimuth preservation

The strong irregular distribution of shot points has a negative impact on subsequent offset vector tile (OVT) domain processing, which is of great importance for azimuth and offset preservation. 5D regularization^[12] was applied to achieve regularization of surface source/detector locations, uniform coverage and azimuth preservation. Significant improvements can be observed in Fig. 8.

Figure  8.  Fold of Coverage map and offset/azimuth preservation display after 5D regularization

下载: 全尺寸图片 幻灯片

3.2   ‘True surface’ depth modeling technology

The inherent drawback of time migration, which is the inability to handle strong lateral velocity variations, prevents its successful application in the loess plateau region. Common issues of previous time imaging results include inability to depict deep structures, strong lateral energy inconsistency and well- seismic mismatch. All these problems add to the uncertainty on reservoir characterization of target clastic rock. Only through depth modeling and imaging can we mitigate the uncertainties. However, the complex topography of the loess plateau and the complex near surface with strong uneven spatial and vertical distribution pose additional challenges to depth imaging. Previous depth modeling in similar areas usually starts from the bottom of weathered layer, but the resultant subsurface geological trend cannot be correctly identified, and the depths of seismic events on imaging sections do not match well with well markers.

‘True surface’ depth migration solution was formed after comprehensive studies and tests, and the following describes the key procedures:

1) Defining ‘true surface’ /migration surface based on the geometry layout and geology settings;

2) Calculating high-frequency statics which can’t be properly solved by tomography;

3) Applying high-frequency statics to first beak time;

4) Deriving near-surface velocity model through diving wave tomography;

5) Building initial depth model from ‘true surface’.

Through the above processing steps, an initial velocity model considering both the surface and the near-surface issues was obtained, and it’s ready for wide azimuth (WAZ) tomography iterations. The detailed workflow is illustrated in Fig.9.

Figure  9.  Depth modeling workflows for Loess plateau area

下载: 全尺寸图片 幻灯片

3.3   Results

Implementing updated static correction methodologies and novel signal conditioning workflows enabled effective preparation of OVT-domain seismic data for depth migration. Subsequent refinement of near-surface velocity modeling facilitated the establishment of an detailed “true-surface” velocity model, serving as the foundation for wide-azimuth tomographic inversions. Multi-cycle velocity model iterations yielded both enhanced velocity fields and superior imaging outputs. Comparative analysis in Fig. 10 between conventional time imaging and the updated PSDM sections demonstrates substantial signal-to-noise ratio improvements, enhanced structural coherency, and optimized event continuity. Critical horizon interpretation reliability has been notably elevated through this approach.

Figure  10.  ‘True surface’ PSDM imaging section (b) compared with legacy PSTM result (a)

下载: 全尺寸图片 幻灯片

Through high-precision velocity modeling and imaging, the seismic-to-well tie accuracy at target horizons was optimized to less than 3‰ (Table 2. ).

Table  2.  Seismic-to-well mis-tie analysis

WELL	Well Depth/m	Imaging depth/m	Mis-Tie/m	Mis-tie (‰)
W1	−2861.7	−2857.6	4.1	1.43
W2	−2817.9	−2813.2	4.7	1.68
W3	−2787.9	−2783.1	4.8	1.73
W4	−2763	−2759.9	3.1	1.13
W5	−2614.5	−2610.5	4	1.53
W6	−2605.1	−2599.8	5.3	2.03
W7	−2586.7	−2581.9	4.8	1.45

下载: 导出CSV 

| 显示表格

The seismic processing workflow strictly adhered to data fidelity principles, ensuring that AVO responses in seismic gathers align with geological interpretations and well-logging observations, Fig. 11. shows the AVO response of the seismic gather is consistent with that of the well logging forward modeling

Figure  11.  AVO consistency analysis of Seismic gather and forward modelling gather (Well W3/W5)

下载: 全尺寸图片 幻灯片

The high-fidelity imaging and pre-stack gather deliverables enabled enhanced reservoir characterization and prospect delineation, the structural interpretation workflow precisely reconstructed paleogeomorphic configurations prior to sediment deposition. High-resolution sequence stratigraphic slicing techniques delineated favorable fluvial depositional architectures by quantifying three-dimensional geometric variations. Confronted with the challenges of thin distributary channel reservoirs and ambiguous P-impedance contrasts in target zones, the team implemented a novel pre-stack elastic inversion framework guided by seismic facies analysis. This approach, empowered by azimuthally-rich pre-stack gathers, achieved robust predictions of reservoir geometry and gas-bearing sweet spots, with post-drilling validations confirming 85% accuracy in sand thickness estimation and 75% reliability in gas saturation delineation.

Five exploratory well locations were prioritized, with three drilling trajectories being dynamically optimized through real-time geological feedback. Synergistic advancements in subsurface interpretation protocols and technical workflows have substantially boosted operator confidence in unlocking the Ordos Basin’s hydrocarbon potential.

4.   Conclusions and discussion

The entire set of processing sequences formed in this study has great reference value and is worth popularizing for the seismic exploration of the loess plateau. It also provides guidelines for the processing of seismic data acquired through nodal seismometers in other complex onshore areas. Some practical solutions have been mentioned in the paper to alleviate data inconsistency issues, but further studies are required for better effects. More advanced and practical near-surface modeling approaches are highly expected in the loess plateau region, such as full waveform inversion for land.

Seismic exploration in loess plateau areas presents extreme technical challenges and heavily relies on high-quality seismic data. Currently, numerous domestic researchers are engaged in processing loess plateau seismic data through approaches such as: AI-optimized first-break picking^[13]. High-precision near-surface Q estimation^[14]. Q model construction and migration^[15] etc. These efforts aim to enhance seismic data resolution and broaden frequency bandwidth. However, fidelity preservation processing of seismic data and consistency processing of target layers in loess plateau areas prove more critical for predicting tight thin reservoirs and hydrocarbon identification. Meanwhile, recent achievements in joint surface-borehole exploration^[16] will contribute to advancing geological understanding of loess plateau exploration, so there is still a long way to go to fully resolve the exploration of the loess plateau. Nevertheless, the following studying areas also require urgent implementation: paleogeomorphic environment analysis during hydrocarbon accumulation in Ordos Basin, identification of favorable source rocks, refined prediction methods for hydrocarbon-bearing sweet spots, Optimization of unconventional hydrocarbon development schemes.

Acknowledgement: The authors would like to thank Sinopec North China Petroleum Bureau for initiating and supporting the Loess Plateau prospecting projects.