Sandstone Reservoir Prediction in the Beikang Basin Using a Fusion Workflow of Tomographic Velocity Inversion and Multiple Seismic Integration

Abstract

The Beikang Basin is a large petroliferous sedimentary basin in the southern South China Sea, and is regarded as an important frontier area for deepwater hydrocarbon exploration in the Nansha region. Previous exploration by foreign oil companies has demonstrated its significant hydrocarbon potential. However, no wells have yet been drilled by China in the Beikang Basin. In this underexplored, well-free setting, sandstone reservoir prediction is hindered by several key challenges, including low velocity-picking accuracy, difficulty in constructing reliable low-frequency models, and limited inversion accuracy. These limitations collectively reduce the reliability of reservoir identification and thus increase exploration risk. To address the challenge of accurately predicting sandstone reservoirs in underexplored areas without well control, this study proposes a reservoir prediction workflow that integrates horizon-controlled gridded tomographic velocity modeling with seismic multiple integration. The workflow includes (1) pre-stack gather optimization, (2) tomographic velocity inversion to construct a precise velocity field, (3) multiple integration to supplement missing medium- to low-frequency information, and (4) pre-stack amplitude-versus-offset (AVO) simultaneous inversion to derive P-wave and S-wave impedances. Combined with petrophysical analysis from adjacent areas to establish a lithology interpretation framework, this integrated approach enables quantitative prediction of sandstone reservoirs in the Beikang Basin. Application to the study area demonstrates that the predicted sandstone thickness distribution is consistent with the regional geological conditions and is supported by drilling results from surrounding areas. The proposed method, therefore, provides a reliable technical approach for reservoir evaluation in the Beikang Basin, and has practical value for sandstone reservoir prediction in other underexplored frontier basins with limited or no well control.

1. Introduction

Since the 1990s, the Guangzhou Marine Geological Survey (GMGS) has conducted oil and gas resource investigations in the Beikang Basin. Nearly 30 years of research have demonstrated that the basin possesses favorable petroleum geological conditions and significant hydrocarbon potential [1,2,3,4,5]. In addition, foreign oil companies have drilled several exploration wells in the basin. Some encountered oil and gas shows (e.g., well Paus-1), some discovered hydrocarbon accumulations of commercial interest (e.g., well Talang-1), and others encountered high-quality sandstone reservoirs (e.g., well Bako-1) [6,7,8,9]. However, the drilling success rate in the Beikang Basin overall remains low, primarily because sandstone reservoirs have not been accurately identified. For instance, wells Paus-1, North-1 and Lala-1 encountered siltstone reservoirs, well Kekek-1 encountered mud-rich reservoirs, and wells Lanjak-1 and Rajawali-1 encountered tight sandstone reservoirs, none of which have commercial development value [7,8,9]. Therefore, reservoir characterization has become a key bottleneck restricting exploration breakthroughs in the Beikang Basin, and accurate reservoir prediction in areas lacking well control remains an urgent challenge.

Acoustic impedance inversion is an important tool for detailed reservoir characterization, as well as for both qualitative and quantitative reservoir evaluation [10,11,12,13]. Previous exploration activities have demonstrated that the results of acoustic impedance inversion are influenced by multiple factors, including the dominant frequency, bandwidth of the seismic data, and the quality of the low-frequency model [10,11,14]. Seismic data commonly lack low-frequency components below 10 Hz [10,13], and the absence of these components can significantly reduce the accuracy of reservoir prediction [10,15]. Therefore, compensation for missing low-frequency information is critical for improving inversion reliability. In recent years, considerable progress has been made in establishing low-frequency models in well-free areas using seismic velocity fields [15,16,17]. For example, high-density velocity analysis [18] and pre-stack depth migration [19] have been employed to improve velocity field accuracy and establish velocity models with broader bandwidths that better reflect actual stratigraphic variations [16,19,20,21,22]. Broadband seismic data have also been obtained through techniques such as deghosting and inverse Q-filtering [19], while high-precision gridded tomographic velocity fields have been acquired via pre-stack depth migration [23,24]. In addition, some studies have integrated seismic facies and low-frequency relative variations derived from trace integration into initial models based on interval velocity, and then combined these with complex frequency domain inversion to generate more reliable low-frequency models [15,25,26,27,28,29,30]. Alternatively, low-frequency models can be constructed from velocities obtained by stereotomographic inversion of 3D seismic data volumes [10]. Practical applications have further demonstrated that low-frequency compensation methods based on multiple integration can effectively mitigate the issue of weak low-frequency seismic energy [13].

The original seismic data lack low-frequency information in the 0–10 Hz range, which is a critical parameter for quantitative reservoir prediction, lithology identification, and physical property inversion. Conventional methods address reservoir prediction challenges and improve reliability by utilizing the low-frequency information contained in well logs. However, as logging information could not be collected for the Beikang Basin, conventional methods are insufficient to fill the seismic bandwidth gap and achieve absolute acoustic impedance inversion. Based on previously proposed methods for compensating low-frequency information in seismic inversion [10,13,14], this paper proposes an integrated technique combining tomographic velocity inversion and multiple integration to effectively compensate for the missing 0~10 Hz low-frequency information in the original seismic data, thereby restoring the absolute acoustic impedance background of the strata. Combined with pre-stack AVO simultaneous inversion techniques [31,32,33,34,35], this transforms the “relative variations” of conventional seismic data into “absolute values” incorporating the low-frequency model, enabling sandstone-mudstone lithology classification, reservoir identification, and distribution characterization. Practical applications have demonstrated that the reservoir prediction results of this technique show high consistency with actual drilling information, providing technical support for subsequent reservoir prediction in the Beikang Basin and other low-exploration areas in the southern South China Sea.

2. Geological Background

The Beikang Basin is located on the Nansha Block in the southwestern South China Sea, and is a Cenozoic sedimentary basin with favorable petroleum potential [36,37,38]. The basin is bounded to the north by fault zones and low uplifts separating it from the Nanweixi and Nanweidong basins, to the southwest by the Tingjia Fault adjacent to the Zengmu Basin, and to the east by the northwestern boundary fault of the Nansha Trough. It covers an area of approximately 63,000 km², and comprises two geomorphic units including the continental slope and the deepwater basin [39,40,41] (Figure 1). The basement consists of pre-Cenozoic metamorphic rocks and felsic to mafic igneous rocks. The stratigraphic succession has been well developed since the Eocene, with a maximum sedimentary thickness exceeding 12,000 m [39,40]. During the tectonic evolution of the Nansha Block, including its rifting from the South China continental margin, subsequent drifting, and eventual collision with the Zengmu Block, two major regional geological boundaries were formed, namely T₃ and T₄ (Figure 2). T₃ represents the boundary between the Early and Middle Miocene and is the most prominent angular unconformity in the basin. It documents the Nansha Movement resulting from the collision between the Nansha Block and the Borneo Block [25,26,27]. T₄ marks the boundary between the Early and Late Oligocene, and is the largest regional paraconformity in the basin. It corresponds to the South China Sea Movement caused by the opening of the new South China Sea and the closure of the proto-South China Sea [41,42,43,44,45].

During the T₃–T₄ interval (Late Oligocene to Early Miocene), global sea level was generally characterized by lowstand regression. Large rivers from Borneo, located to the south of the basin, supplied abundant clastic material, allowing sediments to spread across much of the Beikang Basin and resulting in the development of multiple superposed deltaic to deepwater fan systems [36,46,47,48]. During this period, the Beikang Basin was mainly deposited in littoral to shallow marine environments with relatively strong hydrodynamic conditions. As a result of marine reworking and winnowing, most sandstone reservoirs developed favorable porosity and permeability, making them favorable targets for hydrocarbon accumulation. Consequently, the sandstone reservoirs developed during this interval constitute the primary exploration targets in the Beikang Basin [48,49] (Figure 2).

Figure 2. Comprehensive histogram of sequence stratigraphy and tectono-sedimentary evolution of the Beikang Basin [49].

Figure 2. Comprehensive histogram of sequence stratigraphy and tectono-sedimentary evolution of the Beikang Basin [49].

3. Data and Methods

3.1. Seismic Data

The 2D seismic data used in this study were acquired in the Beikang Basin by a survey vessel operated by the Guangzhou Marine Geological Survey (GMGS). The acquisition parameters included 60 receiver channels, 15-fold coverage, 26 m receiver spacing, 53 m shot spacing, a 2 ms sampling interval, a 7 s record length, and a 1500 in³ air-gun source.

3.2. Pre-Stack Gather Optimization

The quality of pre-stack seismic gathers has a significant impact on the reliability of AVO inversion [31,32]. The seismic data used in this study were acquired in the 1990s and, due to the limitations of acquisition technology at that time, are characterized by relatively poor data quality and a low signal-to-noise ratio (S/N). Therefore, preprocessing of the gathers was required to improve the S/N ratio while preserving true-amplitude information. In this study, pre-stack gather conditioning was carried out mainly using F-X domain linear noise attenuation [50,51] and Radon transform-based processing [52,53].

3.3. Horizon-Controlled Grid Tomographic Velocity Modeling and Inversion

Tomographic velocity inversion utilizes interpreted structural constraints to perform tomographic ray tracing and automatic horizon picking, thereby generating the equations required for 2D tomographic inversion. Through iterative updating, strong reflection events on each depth-migrated CRP (Common Reflection Point) gather are progressively flattened, resulting in more accurate velocity modeling and improved depth-migration imaging [54,55,56,57].

In this study, the tomographic velocity inversion is primarily based on the integral relationship between travel time and velocity (1):

$$T={\int }_{path}{\displaystyle \frac{ds}{v\left(s\right)}}$$

(1)

where “path” represents the actual propagation path of the seismic wave in the subsurface, and “v(s)” is the medium velocity at any point along that path. The velocity model is updated iteratively by continuously comparing “theoretical travel time” and “observed travel time” until the residuals are minimized. The main steps are as follows:

• Structural Model Building: To establish the initial depth-velocity model, horizon calibration was first performed based on the time-migrated seismic section. Time-domain horizon calibration is a fundamental step in pre-stack depth migration because it provides the foundation for depth-domain modeling and constrains both structural geometry and lateral velocity variation. On the time-migrated sections, continuous, high-energy reflectors that effectively define the structural framework of the study area were selected for tracking and interpretation. A total of 13 horizons were interpreted, and these were used to construct a detailed time-domain structural model (Figure 3).
• Construction and Optimization of the Initial Velocity Model: The accuracy of tomographic velocity inversion depends dominantly on the quality of the interval velocity field. Estimation of this field, therefore, becomes a critical step for both tomographic inversion and pre-stack depth migration. In this study, the interval velocity field was determined primarily through the establishment of an initial velocity model followed by iterative updating and optimization of the model [57].
• Horizon-controlled Grid Tomographic Velocity Inversion: Gridded tomographic inversion is a mature global optimization method that utilizes travel-time residuals to update the velocity field. Under the constraints of geological interpretation and seismic reflection characteristics, the inversion becomes more stable, and reduces the non-uniqueness commonly associated with least-squares solutions, ultimately yielding a more accurate velocity model.

3.4. Petrophysical Analysis

Drilling data from the Beikang Basin and adjacent basins were compiled. Analysis indicates that Well-X possesses a complete logging suite, including P-wave velocity, S-wave velocity, density, neutron porosity, bulk modulus, and shear modulus curves. From these data, fundamental elastic parameters such as P-wave impedance, S-wave impedance, bulk modulus, shear modulus, and the V_p/V_s ratio can be further derived. Cross-plot analysis was employed to optimize sensitive parameters for sandstone and mudstone discrimination in the study area [58,59]. Well-X is located in the Zengmu Basin, which is adjacent to the Beikang Basin. The two basins exhibit certain similarities in tectono-sedimentary evolution prior to the Middle Miocene (T₃ horizon). Therefore, the petrophysical data from this well are applicable to the Beikang Basin to a certain extent. However, as logging data are available from only a single well, there are inherent limitations.

3.5. Low-Frequency Model Construction Based on Multiple Seismic Integration

Multiple seismic integration refers to the repeated summation of seismic trace amplitudes. Starting from an initial time t₀, seismic amplitudes are continuously accumulated and recorded at time t, producing a new data volume. The number of cumulative summations performed over the interval from t₀~t defines the order of integration [59,60,61,62,63,64]. Through amplitude accumulation, multiple seismic integration enhances medium- to low-frequency information:

$$s_m\left(t\right)={\displaystyle \underset{m}{\int }{\displaystyle \int s\left(t\right)}}dt$$

(2)

where “S(t)” represents the seismic record, “m” denotes the order of integration, and “S_m(t)” is the integrated seismic response.

$$s_m\left[\left(N-k\right)\Delta t\right]={\displaystyle \sum _{n=1}^{N-k}{s}_{m-1}}\left(n\Delta t\right)\hspace{1em}\hspace{1em}\hspace{1em}k=\left(N-1\right),\left(N-2\right),\cdots 2,1,0$$

(3)

where “N” is the total number of samples, “n” is the sample index at the current time position, “Δt” is the sampling interval, and “k” is a positive integer. In Equation (3), “S₀(nΔt)” is defined as the original seismic trace or wavelet.

In the frequency domain, multiple integration is defined as follows:

$${\overline{I}}^{\left(-n\right)}\left(\omega \right)={\left({\displaystyle \frac{1}{j\omega }}\right)}^{n}S\left(\omega \right)$$

(4)

Taking the integration of a Ricker wavelet as an example, the amplitude spectrum “I(ω)” of the integrated trace is related to the amplitude spectrum “S(ω)” of the original signal. By extension,

$$\left|I(\omega )\right|={\displaystyle \frac{1}{\omega }}\left|S(\omega )\right|$$

(5)

The amplitude spectrum of the targeted low-frequency model to be reconstructed can then be expressed as follows:

$$\left|M\left(\omega \right)\right|=\left({\displaystyle \sum _k{a}_k\frac{1}{{\omega }^k}}\right)\left|S\left(\omega \right)\right|$$

(6)

The key step is to determine the reconstruction coefficient “a_k”.

The relative impedance derived from well logs is expressed as follows:

$$L\left(\omega \right)={\displaystyle {\displaystyle \sum _n}{L}_n\left(\omega \right)}={\displaystyle {\displaystyle \sum _n}{g}_n\left(I^{(-n)}\left(\omega \right)\right)}$$

(7)

To minimize the difference between “I(ω)” and “S(ω)”, the reconstruction coefficient is solved as follows:

$$\min \left\{{\displaystyle \sum _n\left[\left|L_n\left(\omega \right)-a_n{\left(\frac{1}{j\omega }\right)}^{n}S\left(\omega \right)\right|\right]}\right\}$$

(8)

Two key points are critical for reducing reconstruction error, including (a) the sum of the band-limited impedances must equal the impedance of the target frequency band; and (b) the order of multiple integration selected must show the highest similarity to the frequency band represented by the corresponding band-limited impedance.

3.6. Pre-Stack AVO Simultaneous Inversion

Pre-stack simultaneous inversion makes effective use of AVO information contained in pre-stack seismic data. Through simultaneous inversion of multiple common-angle partial stack data volumes, elastic parameters such as P-wave impedance, S-wave impedance, density, V_p/V_s ratio, and Poisson’s ratio can be obtained, providing more diagnostic parameters or parameter combinations for lithology identification. The workflow for pre-stack AVO simultaneous inversion includes low-frequency model construction, seismic wavelet estimation, well-to-seismic calibration, and simultaneous inversion. In this study, the low-frequency model was derived from horizon-controlled gridded tomographic velocity modeling combined with multiple seismic integration, the seismic wavelet was extracted during well-to-seismic calibration, and the inversion was implemented using a sparse spike inversion algorithm [65,66,67,68].

4. Results

4.1. Pre-Stack Gather Optimization Results

Analysis of the pre-stack gathers revealed two main types of data-quality issues: (1) vertically blank, strip-like artifacts and upward-curving linear noise along reflection events, and (2) distortion of events at mid to far offsets caused by multiple contamination including event discontinuity, polarity reversal, amplitude weakening, and blurred reflections. Application of F-X domain linear noise attenuation and Radon transform-based processing effectively suppressed linear noise and multiples present in the original gathers [49,50,51,52,53,69], while preserving valid seismic reflections. As a result, the reflection events became significantly clearer (Figure 4), providing higher-quality input data for the subsequent horizon-controlled gridded tomographic velocity inversion and pre-stack AVO inversion.

4.2. Tomographic Velocity Modeling

Through structural model construction, establishment and optimization of the initial velocity model, and horizon-controlled gridded tomographic velocity inversion [59], a more accurate velocity field was obtained. The updated velocity field shows a spatially consistent distribution in both magnitude and spatial pattern, and its geometry is consistent with the structural framework of the study area (Figure 5). The refined tomographic velocity inversion not only reduced horizon-based residual move-out spectra to near zero and flattened the depth-migrated CRP gathers, but also minimized the vertical residual move-out spectra. Furthermore, the migrated seismic images obtained with the optimized velocity field show increased focusing and event continuity relative to those from the initial model.

4.3. Establishment of the Lithology Interpretation Template

P-wave impedance was cross-plotted against S-wave impedance, density, V_p/V_s ratio, and neutron porosity, with shale content as the color scale in the attribute cross-plots.

Sandstone is characterized by relatively high P-wave and S-wave impedances. P-wave impedance or S-wave impedance alone cannot reliably distinguish sandstone from mudstone. However, in the cross-plot of P-wave impedance vs. S-wave impedance, the two lithologies follow different background trends. Using the blue boundary shown in the figure, sandstone and mudstone can be effectively distinguished (Figure 6).

Sandstone is also characterized by relatively high P-wave impedance and low density. In the cross-plot of P-wave impedance versus density, sandstone and mudstone can be discriminated either by the blue dividing line shown in the figure, or by a density threshold of less than 2.58 g/cm³ (Figure 7).

As drilling data from the Beikang Basin could not be collected, a method of establishing virtual wells within the basin was adopted. The deployment of virtual wells follows the principle of uniform distribution across different structural positions with a minimum of three wells; for key targets, the number may be further increased. The extraction of virtual wells follows the principle of selecting locations at structural highs and areas of moderate geological variation, with a minimum of three wells. The established low-frequency model data are then implanted into the virtual well data traces to obtain the acoustic impedance curves for the virtual wells.

A comparison of P-wave impedance between the adjacent Well-X and the pseudo-well in the Beikang Basin was conducted. Statistical analysis of the 3600~3900 m interval showed that the background impedance of the Well-X log data is approximately 9400 (g/cm³)·(m/s), whereas that from the pseudo-well in the Beikang Basin is approximately 9000 (g/cm³)·(m/s). Based on this difference, the lithology interpretation template derived from the adjacent area was adjusted to establish a template suitable for the study area (Figure 8).

4.4. Low-Frequency Model Construction via Multiple Seismic Integration

The original seismic data have a bandwidth of 9~48 Hz and lack low-frequency information. First-order integration produced a data volume with a bandwidth of 8~40 Hz [13,60]; third-order integration produced a bandwidth of 7~18 Hz; and fifth-order integration produced a bandwidth of 3~10 Hz. These results show that the dominant frequency decreases progressively with increasing order of integration (Figure 9), indicating that multiple seismic integration can be used to compensate for the frequency content missing below 9 Hz in the original seismic data.

By integrating the fifth-order integration volume with the seismic tomographic velocity spectrum, compensation of the 0~10 Hz band can be achieved. This effectively supplements the missing low-frequency content of the original seismic data, providing a valid low-frequency model for inversion in the absence of well control.

In this study, multiple tests were conducted to determine the optimal integration window and integration order. The core control parameters can be divided into: integral multiplicity, sliding time window, seismic wavelet, reservoir geological parameters, and sampling and frequency parameters. Together, they determine the frequency reduction magnitude of the integrated seismic data, reservoir identification accuracy, and lateral stability.

Integral Multiplicity: Higher multiplicity results in stronger frequency reduction and richer low-frequency information; single integration preserves 14~48 Hz, triple integration reduces to 7~14 Hz, and quintuple integration reduces to 3~7 Hz.

Sliding Time Window: A larger window enhances low frequencies and increases smoothness; a smaller window preserves high frequencies and retains richer details.

Seismic Wavelet: The conventional dominant frequency ranges from 20 to 40 Hz, which directly affects the frequency band division after integration; the lower the dominant frequency, the higher the proportion of low-frequency components after integration. The zero-phase wavelet is most commonly used, ensuring symmetric waveforms after integration and clear reservoir interfaces.

Reservoir Geological Parameters: The greater the reservoir thickness, the higher the multiplicity required to enhance the low-frequency response; the wave impedance contrast between the reservoir and surrounding rock controls the magnitude of reflection coefficients, thereby affecting the strength of amplitude anomalies after integration.

The integration window was selected to avoid spectral distortion while rapidly compensating for the missing frequency band. The integration order was chosen to achieve effective matching between the low frequency seismic content and the higher frequency velocity information. Based on these tests, an integration window of 100 ms and an integration order of 5 were selected for the study area (Figure 10).

Figure 11 compares the spectra of tomographic velocity model, the fifth-order integrated seismic data, and the original seismic data. The results show that fifth-order seismic integration effectively fills the spectral gap between the velocity model and the original seismic data.

Figure 12 shows the original seismic section, the integrated seismic sections, and the velocity sections for first-, third-, and fifth-order integration. The integrated seismic sections preserve the original structural geometry, and show good agreement with the velocity field, while higher orders of integration correspond to progressively lower frequencies.

4.5. Pre-Stack AVO Simultaneous Inversion

Figure 13 shows the results of pre-stack AVO simultaneous inversion. The inverted P-wave impedance, S-wave impedance, and V_p/V_s ratio all exhibit strong correspondence with the seismic reflections. Among these parameters, the cross-plot of P-wave impedance versus S-wave impedance provides the basis for quantitative lithological interpretation [37,38,39].

4.6. Lithology Prediction

Using the lithology interpretation template established from the petrophysical analysis, quantitative lithological prediction was performed. Specifically, sandstone cutoff values were defined by combining the P-wave and S-wave impedance volumes, and these thresholds were then used to generate lithology interpretation sections distinguishing sandstone from mudstone. The lithology interpretation section along Line DD’ (Figure 14) shows that sandstone bodies are mainly distributed within the T₃–T₄ sequence. Laterally, the sandstone bodies exhibit good continuity and broad lateral extent; vertically, they occur as stacked and interbedded units with substantial cumulative thickness.

5. Discussion

5.1. Feasibility of the Multiple Seismic Integration Method

Given that no wells drilled by China are currently available in the Beikang Basin and that foreign drilling data are difficult to obtain, the key challenge in using seismic inversion to predict sandstone reservoirs is the construction of a reliable broadband low-frequency model. Conventional approaches generally rely on abundant well data. In underexplored areas such as the Beikang Basin where well control is absent or extremely limited, low-frequency models established using conventional methods commonly suffer from inadequate low-frequency constraints and limited inversion accuracy, thereby reducing the reliability of sandstone reservoir identification. The integrated workflow proposed in this study, which combines tomographic velocity inversion with multiple seismic integration, can effectively supplement low-frequency information in the 3~10 Hz range and thus, alleviates the problem of missing low-frequency components in seismic inversion without well control.

Figure 15 compares the modeling and inversion results obtained using the integrated tomographic velocity inversion and multiple seismic integration workflow with those obtained using conventional tomographic velocity inversion alone. The comparison shows that incorporation of multiple seismic integration improves the characterization of sandstone bodies within the target interval and more clearly resolves their superposed geometries. This improvement is expressed mainly in two aspects: first, slope-confined sand bodies are delineated more clearly and appear as drape-like or strip-like features attached to the slope; second, the vertical stacking relationships among basin floor sand bodies are displayed more distinctly.

5.2. Geological Implications

Two representative seismic sections, FF’ and GG’, were selected from the Beikang Basin. For the T₃–T₄ interval, which constitutes the main sandstone-bearing target sequence, sandstone reservoir prediction was conducted using the integrated tomographic velocity inversion and multiple seismic integration workflow. On this basis, a plan-view sandstone thickness map for the T₃–T₄ sequence was compiled (Figure 16). Combination of the plan-view and sectional results shows that areas with greater predicted sandstone thickness in map view, represented by warmer colors, correspond to deep sag zones on the seismic sections, where the cumulative vertical thickness of sandstone is also greatest. This spatial relationship is consistent with the expected geological pattern of sandstone deposition. Currently, only 2D seismic data are available in the Beikang Basin. Consequently, these data are insufficient for accurately characterizing the three-dimensional geometry of reservoirs, and the planar thickness maps of sand bodies can only be qualitatively predicted at this stage. Further detailed delineation and quantitative characterization of the planar distribution and thickness of sand bodies will be conducted as additional data become available.

Comparison of the predicted sandstone thickness map with the sedimentary facies map (Figure 17) further reveals that areas with more strongly developed sandstone on the thickness map generally correspond to delta plain or delta front facies on the sedimentary facies map. These regions are recognized as the main zones of sandstone reservoir development in the Beikang Basin, consistent with the regional geological framework. Furthermore, the drilling information shown in Figure 17 indicates that the predicted sandstone distribution is supported by existing well data [41,46]. Therefore, these results suggest that the integrated tomographic velocity inversion and multiple seismic integration workflow can be extended to sandstone reservoir prediction over a broader area of the Beikang Basin.

6. Conclusions

The Beikang Basin developed numerous sandstone reservoirs during the T₃–T₄ interval (Late Oligocene to Early Miocene), which represent important exploration targets in the basin. The tomographic velocity inversion and multiple integration fusion techniques established in this study effectively enhance low-frequency signal quality, enabling low-frequency information compensation in the 3~10 Hz range. Combined with pre-stack AVO simultaneous inversion technology, this approach achieves quantitative prediction of sandstone reservoirs in the Beikang Basin.

The sandstone reservoir thickness maps predicted by the technical method proposed in this study are consistent with actual geological conditions, and the prediction results have been validated by drilling data. This indicates that the method provides a reliable technical approach for sandstone reservoir evaluation in the Beikang Basin.

The proposed methodology is based on 2D seismic data from underexplored areas. Constrained by data limitations, sandstone reservoir identification remains at the qualitative prediction level. Nevertheless, this approach holds practical value for preliminary hydrocarbon surveys in underexplored areas and can facilitate subsequent fine-scale characterization of sandstone reservoirs.