Seismic Identification and Characterization of Deep Strike-Slip Faults in the Tarim Craton Basin

Abstract

Through hydrocarbon explorations, deep carbonate reservoirs within a craton were determined to be influenced by deep strike-slip faults, which exhibit small displacements and are challenging to identify. Previous research has established a correlation between seismic attributes and deep geological information, wherein large-scale faults can cause abrupt waveform discontinuities. However, due to the inherent limitations of seismic datasets, such as low signal-to-noise ratios and resolutions, accurately characterizing complex strike-slip faults remains difficult, resulting in increased uncertainties in fault characterization and reservoir prediction. In this study, we integrate advanced techniques such as principal component analysis and structure-oriented filtering with a fault-centric imaging approach to refine the resolution of seismic data from the Tarim craton. Our detailed evaluation encompassed 12 distinct seismic attributes, culminating in the creation of a sophisticated model for identifying strike-slip faults. This model incorporates select seismic attributes and leverages fusion algorithms like K-means, ellipsoid growth, and wavelet transformations. Through the technical approach introduced in this study, we have achieved multi-scale characterization of complex strike-slip faults with throws of less than 10 m. This workflow has the potential to be extended to other complex reservoirs governed by strike-slip faults in cratonic basins, thus offering valuable insights for hydrocarbon exploration and reservoir characterization in similar geological settings.

1. Introduction

Strike-slip faults in cratonic basins, such as ruptures, are the products of pre-existing structures responding to stress concentrations within the craton. These faults are notably situated a significant distance away from active plate boundaries [1,2,3]. A distinguishing feature between intra-cratonic strike-slip faults and other large-scale strike-slip faults, such as those found at plate boundaries or embedded collision zones, lies in their slip distances, with the former typically ranging from hundreds to thousands of metres and the latter commonly extending to hundreds of kilometres [4]. Characterized by weak activity and high steepness, these strike-slip faults are difficult to identify with two-dimensional seismic data and are often overlooked [5,6]. Recent oil and gas discoveries related to strike-slip faults in deep and ultra-deep strata within cratonic basins have underscored their substantial role in controlling reservoir development and hydrocarbon accumulation, giving rise to the “near-source and fault-controlled” model [7]. Nevertheless, accurately characterizing these complex strike-slip faults using conventional seismic interpretation methods is difficult because of the limitations of low signal-to-noise ratios and resolutions in seismic datasets, leading to numerous controversies surrounding their underlying mechanisms [8]. Consequently, improving the seismic resolutions and extracting the precise characterizations of these deep and intricate craton basin faults pose notable challenges [9].

Seismic attribute fusion has been at the forefront of innovations aimed at refining the accuracy and reliability of fault identification in subsurface exploration. Utilizing multiple attributes provides a comprehensive perspective, enhancing the clarity of fault systems within seismic datasets. In the Bakken Formation of the Williston Basin, North Dakota, principal component analysis of seismic attributes has been pivotal in discerning fault patterns, exemplifying the power of methodological integration for fault detection [10]. Additionally, merging multiple seismic attributes with advanced techniques like ant tracking and neural networks has been demonstrated to be particularly effective in fault detection, as highlighted in a study conducted in an Iranian oilfield [11]. The potential of multi-seismic-attribute analysis techniques in unveiling intricate fault configurations has also been evidenced in the southern Indus Basin, Pakistan, shedding light on previously obscured subsurface mysteries [12]. Furthermore, when analyzing structural complexities in compression-driven terrains, 3D seismic attributes play an indispensable role, as showcased in a study from the western Sichuan Basin [13]. The amalgamation of these methodologies underscores the rapidly evolving landscape of seismic exploration and the indispensable role of seismic attribute fusion in modern geophysics.

At present, pinpointing attributes for deep strike-slip faults within cratons is still in its preliminary phase [9]. Notably, Qiu et al. (2019) made use of coherent volumes to detect strike-slip faults in the Tarim Basin, yielding significant outcomes in constructing a static model for hydrocarbon complex accumulation [14]. Similarly, Albesher et al. (2020) employed ant tracking attributes to map the Algeciras strike-slip fault in the Greater Llanos Basin, revealing the presence of WNW–ESE and NW–SE Riedel-type shear faults [15]. However, the characterization of strongly heterogeneous paleokarst fracture-vuggy carbonate reservoirs controlled by strike-slip faults at depths of 7000–8000 m proves challenging solely through seismic attributes [16,17]. A potential solution involves combining geophysical data, geological prior information, signal processing, and image fusion technology to establish a comprehensive and accurate identification method for deep strike-slip faults within cratons.

By considering the complex geological–geophysical mapping relationship of deep strike-slip faults, the structural styles of strike-slip faults and their controlling effects on oil and gas are summarized in this paper through outcrop and physical simulation experiments. To improve the quality of three-dimensional (3D) seismic data, principal components, structure-oriented filtering, and fault-focused imaging were used to control the quality of the seismic data. Based on mathematical theories, the seismic attributes were divided into four categories: energy, curvature, correlation, and gradient. In this study, the amplitude gradient attribute and ant track attribute were selected. The new seismic attribute was generated by the image fusion of grid generations, ellipsoid expansions, and wavelet transforms, which can be used to identify the exact location and inside details of faults. The strike-slip fault structure model was established by profile interpretation and outcrop calibration. The fault characterization results can provide a research example for the geological structure and oil and gas enrichment theory of strike-slip faults in cratonic basins.

2. Geological Structure of Strike-Slip Faults

2.1. Basic Characteristics

The Tarim Basin is a typical superimposed basin that encompasses an area of 40 × 10⁴ km², and it is the largest petroliferous basin in China (Figure 1a) [16]. The central part of the Tarim Basin has a long history of uplift. The research area is in the north-central Tarim Basin, which is controlled by faults (Figure 1b) [18]. Fault systems can be divided into two sets based on their fault scales, time of breakage, and faulted horizons: the Tazhong I tectonic belt, which developed primarily during the Caledonian, trends NW–SE and is associated with secondary thrust faults, and the strike-slip fault systems, which developed primarily during the Hercynian and trend NE–SW [19,20].

The strike-slip faults found in the Tarim Basin are geographically distant from active plate boundaries and result from the reactivity of pre-existing structures under stress concentrations within the craton [21,22]. While the Anderson model provides a reasonable explanation for the mechanics of shallow crustal strike-slip faults, the complex nature of deep cratonic basins introduces several factors that influence fault formation and development [23]. These factors encompass regional and local stress fields, pre-existing structures, pre-existing faults, and rock physics, and they result in non-Anderson mechanisms such as fault end expansion, sectional friction, and dynamic rupture [24,25]. The strike-slip faults within cratons exhibit a high degree of structural complexity and are often associated with folds, secondary faults, and fault blocks. Within fault zones, secondary faults and extensional faults develop heterogeneously, typically initiating in regions experiencing the highest tensile stress [26]. These faults serve as crucial connectors, bridging otherwise discontinuous fault segments and playing a key role in fault development [27,28]. Minor- and meso-scale faults are generally not planar but manifest as trapezoidal fault zones interconnected at various scales [18]. The weaker portions of pre-existing structures facilitate fault nucleation and connection growth, not only leading to constraints on the development and distribution of subsequent faults but also influencing the underlying mechanism of fault occurrence [29]. Additionally, fluid flow within different stress tectonic zones along the fault may display heterogeneity, resulting in the formation of channels within highly faulted rock zones [30].

2.2. Structural Styles

From past outcrop studies and structural physics simulations, we have observed that strike-slip faults display a variety of structural patterns. Their intricate formation processes lead to multiple fault zones and the presence of smaller branch faults. This complexity contributes to the formation of intricate geological structures, petrophysical attributes, seepage characteristics, and pronounced heterogeneity [31]. Notably, the outcrops reveal the presence of fault zones within strike-slip faults, showing distinct variations in their distribution, with these fault zones primarily concentrated at fault intersections, transitions, overlaps, and terminations [32] (Figure 2). Moreover, physical simulation experiments demonstrate that the upward growth of strike-slip faults in deep cratonic carbonate basins initially leads to the development of echelon structures that manifest as a complex “spiral-drag” pattern, which subsequently connect through Y-shear faults [30]. Such sectional connections and interactions represent significant mechanisms that drive fault growth [33]. Additionally, the differences in the rock’s composition, structure, and physical attributes play a crucial role in governing the formation and development of these faults within the geological setting.

2.3. Fault-Controlled Reservoir

The advancement of seismic datasets in the cratonic basin has led to an increasing number of scholars recognizing the significant influence of tectonic evolution, fault stage differences, and strike-slip fault systems on the modification of physical attributes of carbonate reservoirs, oil and gas migration pathways, and the control of oil and gas traps [34]. Under the extrusion effects of tectonic stress, deep strike-slip faults, particularly within brittle carbonate strata, give rise to a complex fault system that evolves into fault zones of considerable scale [35]. Subsequent dissolution processes further reshape these early fault zones, with karst water infiltrating along fault strikes, dissolving carbonate rocks. The presence of large strike-slip faults renders deep and surrounding strata susceptible to localized deep hydrothermal upwelling, resulting in further dissolution and the transformation of fault zones and adjacent strata, consequently forming paleokarst reservoirs exhibiting diverse spatial structures [36]. Notably, reservoirs in the Tarim Basin and Llanos Basin display oil- and gas-bearing characteristics, with localized enrichment occurring along the fault zones. The strike-slip faults serve as principal pathways for longitudinal oil and gas migration and accumulation, and the multistage activities of the faults manifest in the characteristics of multistage accumulation [16].

3. Seismic Resolution Improvement

The seismic data used in this research were derived from carbonate rock formations that possess pronounced wave impedance properties. This leads to a strong damping effect on the seismic reflection energy. Aside from the distinctive reflection features arising from low-velocity fault-karstic bodies, the interior of carbonate rocks primarily displays blank or weak reflection characteristics. Therefore, identifying the reflection features associated with sequence-level strike-slip faults proved challenging from the seismic profiles, exacerbating the difficulty in interpreting and identifying strike-slip faults. The mechanism governing strike-slip faults within cratons is intricate, often involving multi-phase tectonic movements, leading to suboptimal seismic imaging effects and a low signal-to-noise ratio in the seismic data. Thus, seismic profiles with “fault + chaotic reflection” characteristics are frequently being obtained (Figure 3a). This challenge is particularly pronounced in deep strata, in which the seismic reflection energy is attenuated and the frequency is low, hindering the detection of smaller faults, typically in the range of 3~5 m. To address these limitations, this paper employed a combination of principal component filtering, structure-oriented filtering, and fault-focused imaging techniques to enhance the quality of the seismic dataset, facilitating improved accuracy and resolution in the identification of strike-slip faults.

3.1. Principal Component Filtering

Strike-slip faults manifest in seismic profiles as distinct discontinuity points along the same phase axis, thus enhancing the continuity of effective seismic reflections and bolstering the fault identification capability of the seismic data. However, the presence of noise can significantly impact the quality of the seismic data and subsequently hinder accurate interpretation. Consequently, denoising seismic datasets is of critical importance.

One effective approach employed for denoising is the principal component analysis (PCA) algorithm, which leverages eigenvalues as features in orthogonal signal space orthogonal decomposition. This enables the enhancement of coherent energy while suppressing interference [37]. The methodology involves calculating and analyzing the covariance matrix of seismic traces within a defined time window, facilitating a robust denoising process and improving the overall quality of seismic data for further analysis and interpretation.

$$R=\left[\begin{array}{cccc}{r}_{11}& r_{12}& \cdots & r_{1p}\\ r_{21}& r_{22}& \cdots & r_{2p}\\ \cdots & \cdots & \cdots & \cdots \\ r_{p1}& r_{p2}& \cdots & r_{pp}\end{array}\right]$$

(1)

$$r_{ij}={\displaystyle \frac{\sum _{k=1}^n\left(x_{ki}-\overline{x}\right)\left(x_{ki}-\overline{x_j}\right)}{\sqrt{\sum _{k=1}^n{\left(x_{ki}-\overline{x}\right)}^2\sum _{k=1}^n{\left(x_{ki}-\overline{x_j}\right)}^2}}}$$

(2)

In these expressions, the eigenvalues of the covariance matrix are obtained, and $n$ is the number of samples. $k$ is the serial number of the $i$ variable. $p$ eigenvalues are obtained by solving the eigenequation of the correlation coefficient matrix $R$: $\left(R-\lambda I_p\right)=0$. Then, the unit eigenvector $b_j^o$ is obtained by solving the equation $Rb=\lambda j$ [38].

Finally, the features of the weighted reconstruction signal of the eigenvector corresponding to each eigenvalue are analyzed as follows:

$$U_{ij}=z_i^T{b}_j^o$$

(3)

The reconstructed signal representing the interference signal is eliminated to achieve noise filtering.

3.2. Structure-Oriented Filtering

The structure-oriented filtering technique is a significant approach for leveraging the dip and azimuth information of geological targets, thereby enhancing the attribute prediction accuracy and target detection capabilities [39]. By maintaining the essential characteristics of the original seismic signal while improving its signal-to-noise ratio, this method effectively emphasizes the discontinuity along the in-phase axis. Consequently, it yields a more pronounced image effect of faults and enhances fault identification, even for fine faults. This technique proves valuable in its ability to highlight fault-related features and supports the accurate identification of subsurface structures.

The principle of construction-guided filtering is to add a parameter $\epsilon$ that is sensitive to the discontinuous region of the structure based on the tensor diffusion model. The value range is close to 1 in smooth regions and close to 0 in special structure regions, such as faults and pinch-offs [40]. First, the structure tensor is used to extract the local structure information of the image. Then, the diffusion tensor is designed according to the structure tensor. The formation azimuth and dip are obtained by multi-window dip scanning, and directional filtering is performed along the formation by parameter $\epsilon$, which is called the structure-oriented filtering equation [41]:

$$\left\{\begin{array}{c}{\displaystyle \frac{\partial u}{\partial l}}\left(x,y,t\right)=div·\left(\epsilon ·D\left(J_p\right)\nabla u\left(x,y,t\right)\right)\\ u\left(x,y,0\right)=u_0\left(x,y\right)\end{array}\right.$$

(4)

where $\epsilon ={\displaystyle \frac{Tr\left(S_0{S}_{\rho }\right)}{Tr\left(S_0\right)Tr\left(S_{\rho }\right)}}$ represents a diffusion continuity factor, $Tr$ represents the trace of the matrix, and $S_0$ represents the initial structure tensor matrix. $S\rho$ represents the gradient structure tensor for the current number of iterations. The equation was discretized, and the iterative formula of filtering was obtained as follows: $u_{\left(n+1\right)}=u_n+∆t·div({\epsilon }_n{D}_n \nabla u_n)$.

$n$ represents the number of iterations, and $∆t$ represents the time step of the iteration. Generally, noise in seismic data can be effectively removed after 5–10 iterations. Comparing the original seismic profile and seismic profile after PCA+ structure-oriented filtering, the latter shows that the background noise is better suppressed, the breakpoint is clearer, and the fault location is clearer (Figure 3b).

3.3. Fault-Focused Imaging

Following the implementation of structure-oriented filtering, notable improvements in the fault resolution and clarity have been achieved, albeit with certain faults still exhibiting issues such as unclear boundaries and weak continuity. To address these limitations, applying fault-focused imaging technology has become essential for enhancing fault continuities and highlighting fault boundaries [42].

To attain a more refined filtering effect on the features of geological structures and effectively reduce seismic noise, the pivotal step in this approach involves the derivation of the diffusion tensor [43]:

$$\left\{\begin{array}{c}{\displaystyle \frac{\partial U}{\partial t}}=div\left(D·\nabla U\right)\\ {\left.U\right|}_{t=0}=U_0\end{array}\right.$$

(5)

where $t$ is the diffusion time; $div$ is the divergence operator; $D$ is the tensor diffusion coefficient of the diffusion filter; $U$ is the result of tomography focusing imaging at time $t$; and $U_0$ is the original data at time $t=0$.

Moreover, $\left|r\right|$ is the distance of the data point from the data centre.

The gradient structure tensor can be constructed as follows:

$$J_p\left(\nabla U\right)=G_{\rho }\ast J_0=G_{\rho }\ast \left(\nabla U{\left(\nabla U\right)}^T\right)$$

(6)

where $U$ is the original seismic data and $J_0$ is the tensor product of the gradient vector.

The Gaussian function with scale $\rho$ is

$$G_{\rho }\left(r\right)={\left(2\pi {\rho }^2\right)}^{-{\displaystyle \frac{3}{2}}}\exp \left(-{\displaystyle \frac{{\left|r\right|}^2}{2{\rho }^2}}\right)$$

(7)

The eigenvector of the structure tensor can be obtained as follows:

$$J_{\rho }\left(\nabla U\right)=\left(v_1 v_2{ v}_3\right)\left[\begin{array}{ccc}{\lambda }_1& 0& 0\\ 0& {\lambda }_2& 0\\ 0& 0& {\lambda }_3\end{array}\right]{\left(v_1 v_2{ v}_3\right)}^T$$

(8)

where $v_1$, $v_2$, and $v_3$ are the three feature vectors of the gradient structure tensor, and they are regarded as the local orthogonal coordinate system. Then, $v_1$ points to the gradient direction of the signal, and the plane composed of $v_2$ and $v_3$ is parallel to the local structural features of the signal. ${\lambda }_1$, ${\lambda }_2$, and ${\lambda }_3$ represent the corresponding eigenvalues.

Linear structure confidence metric:

$$C_{line}={\displaystyle \frac{{\lambda }_2-{\lambda }_3}{{\lambda }_2+{\lambda }_3}}$$

(9)

Planar structure confidence metric:

$$C_{plane}={\displaystyle \frac{{\lambda }_1-{\lambda }_2}{{\lambda }_1+{\lambda }_2}}$$

(10)

Focusing tensor:

$$D=\left(v_1 v_2{ v}_3\right)\left[\begin{array}{ccc}{u}_1& 0& 0\\ 0& u_2& 0\\ 0& 0& u_3\end{array}\right]{\left(v_1 v_2{ v}_3\right)}^T=\left[\begin{array}{ccc}{D}_{11}& D_{12}& D_{13}\\ D_{21}& D_{22}& D_{23}\\ D_{31}& D_{32}& D_{33}\end{array}\right]$$

(11)

where $u_1$, $u_2$, and $u_3$ are the three non-negative eigenvalues of the focusing tensor. In the interval [0,1], they represent the filtering intensity of the fault-focused imaging along the three characteristic directions $v_1$, $v_2$, and $v_3$.

The fault-focused imaging algorithm effectively retains the distinctive “feathery zone” and “flower-like structure” associated with craton strike-slip faults while significantly enhancing the imaging capabilities of seismic data concerning branch faults (Figure 3c). Concurrently, the algorithm exhibits noise suppression properties, leading to an improvement in the transverse continuity of the in-phase axis and a higher signal-to-noise ratio in the seismic signals.

4. Seismic Attribute Generation

Seismic attributes, which include geometric, kinematic, dynamic, or statistical characteristics derived from pre-stack or post-stack seismic datasets through mathematical transformations, are valuable tools for fault identification. In the context of identifying strike-slip faults within cratonic basins, 12 attribute bodies that are categorized into four groups based on underlying principles are focused on in this paper (

Table 1. Seismic attributes applied to the identification of strike-slip faults in cratonic basins.

Categories	Types of Attributes	Results	Score
Energy	Root mean square amplitude	Not clear fault	3
	Low-frequency energy	Clear fault with the large noise disturbance of the whole area	6.5
	High-frequency attenuation	Not clear fault	4
Curvature	Maximum positive curvature	Clear fault with the large noise disturbance of the whole area	6
	Minimum negative curvature	Clear fault with the large noise disturbance of the whole area	7
	Dip	Clear fault with the large noise disturbance of the partial area	7.5
Correlation	Coherent	Clear fault with the large noise disturbance of the partial area	7
	Likelihood	Not clear fault	3
	Ant tracking	Clear main fault with false branch fault	8.5
	Auto fault extraction (AFE)	Clear fault with the large noise disturbance of the partial area	7
Gradient	Amplitude variance	Clear fault with the large noise disturbance of the whole area	7.5
	Amplitude gradient	Clear main fault and clear branch fault	9

). These attributes yield dimensionless parameters that signify specific physical properties. Through a comparison of the planar effects of strike-slip faults, the most suitable seismic attributes for deep strike-slip faults are then optimized to facilitate accurate fault identification.

4.1. Energy Attributes

The amplitude attribute, indicating reflection intensity, directly captures changes in the reflection coefficient (at the wave impedance interface), amplitude variation with offset (AVO), and fluid presence within reservoir pores. It reflects the variation in adjacent lithology underground or the existence of oil-bearing rock strata [44]. The gradual change in the amplitude in the lateral direction usually reflects the change in lithology or the change in lithofacies, and its sudden change may indicate the existence of faults or the edges of oil and gas rocks. Therefore, the amplitude attribute can be directly used to describe the formation changes and predict oil and gas attributes.

(1)

Root mean square amplitude

The root mean square (RMS) amplitude is the most commonly used attribute in seismic datasets. It has a good correlation with the rock density and is often used in lithologic phase transition analysis [45]. According to the calculation formula, the RMS amplitude is positively correlated with the amplitude. The amplitude information is correlated with the reflection coefficient of the stratum [46]. When the seismic wave is vertically incident, the relation between the amplitude and density is as follows:

Incident wave reflected wave:

$$R={\displaystyle \frac{A_{\mathrm{i}}}{A_{\mathrm{r}}}}={\displaystyle \frac{{\rho }_2{v}_2-{\rho }_1{v}_1}{{\rho }_2{v}_2+{\rho }_1{v}_1}}$$

(12)

where $R$ is the reflection coefficient of the interface; ${\rho }_1$ is the density of the medium above the reflecting surface; ${\rho }_2$ is the medium’s density below the reflecting surface; $v_1$ and $v_2$ are the velocities above and below the reflecting surface, respectively; $A_{\mathrm{r}}$ is the amplitude of the reflected wave; and $A_{\mathrm{I}}$ is the amplitude of the incident wave.

According to Gardne’s law,

$$\rho =a{v}^{0.25}$$

(13)

where $\rho$ is the density, g/cm³; $v$ is the velocity; and $a$ is the constant, 0.31. By substituting Equation (13) into Equation (12), we obtain

$$R={\displaystyle \frac{{\rho }_2^5-{\rho }_1^5}{{\rho }_2^5+{\rho }_1^5}}$$

(14)

According to the above formula, the root mean square amplitude is positively correlated with the rock density. The RMS amplitude extracted from the fault development model is significantly higher than that from the non-fault development model.

According to the identification results of strike-slip faults in the cratonic basin with root mean square amplitude attributes, the main faults of the strike-slip faults are relatively clear, but the branch faults are not clear (Figure 4a). In addition, there is much non-fault interference in the whole area, which is not conducive to the effective identification of strike-slip faults.

(2)

Low-frequency energy and high-frequency attenuation

When strike-slip faults are developed in a cratonic basin, the wave absorption capacity of the strata is enhanced, and it is mainly characterized by high-frequency absorption and low-frequency energy enhancement. Therefore, frequency attenuation is considered to be the combined effect of fault development and hydrocarbon reactions. The amplitude energy decreases with time [47]. Due to the existence of fractures, the elastic wave captures more energy when it passes through the fault-developed zone. Moreover, the attenuation is large, while the high-frequency energy is absorbed. Therefore, the elastic wave shows weak amplitude at a high frequency, and the attenuation characteristics of the non-fault-developed zone are not obvious. The increase in low-frequency energy is a relative increase due to the loss of high-frequency energy [48]. Due to the seismic response characteristic of high-frequency absorption shown by the faults, characteristics similar to “low-frequency enhancement” appear, and the waveform characteristics show little variation with depth.

According to the identification results of the strike-slip faults of the cratonic basin with low-frequency energy attributes and high-frequency attenuation attributes, most areas are significantly disturbed by “non-fault” areas, and a small number of branch faults can be shown, but they are not as obvious (Figure 4b,c).

4.2. Curvature Attributes

The curvature attributes of the seismic dataset reflect the degree of plane bending when the stratum is compressed by tectonic stress. The characteristics of small disturbances, folds, bulges, and differential compaction are identified from the curvature attributes [47]. The large faults show an obvious stagger of the in-phase axis in the conventional seismic profile, while the faults with small displacements show slight stagger, distortion, and sudden weakening of the amplitude in the seismic profile.

(1)

Maximum positive curvature and minimum negative curvature

The maximum positive curvature and minimum negative curvature attributes are 3D plane attributes used to quantify the degree of layer interface deviation from the plane. The largest positive curvature in the normal curvature is called the maximum positive curvature. This curvature can amplify the fault information and some small linear structures in the plane [48,49]. This curvature property is very useful for defining faults and their geometries. Faults represented by this attribute exhibit positive curvature values (high values). The minimum negative curvature in the normal curvature is called the minimum negative curvature, and its function is similar to that of the maximum positive curvature [50]. Combined with the maximum positive curvature, the characteristics of the surface can also be determined.

The rate of change in the reflecting surface along different directions is reflected by the gradient $grad\left(u\right)$:

$$grad\left(u\right)={\displaystyle \frac{\partial u}{\partial x}}\overrightarrow{i}+{\displaystyle \frac{\partial u}{\partial y}}\overrightarrow{j}+{\displaystyle \frac{\partial u}{\partial t}}\overrightarrow{k}=p\overrightarrow{i}+q\overrightarrow{j}+r\overrightarrow{k}$$

(15)

where $p$, $q$, and $r$ are the components of apparent inclination along the $x$, $y,$ and $t$ directions, respectively.

By substituting the apparent inclinations $p$ and $q$ into the equation, the curvature components along the $x$ and $y$ directions can be written as follows:

$$\left\{\begin{array}{c}{\kappa }_x={\displaystyle \frac{\frac{{\partial }^{2}u\left(t,x,y\right)}{\partial x^2}}{{\left[1+{\left[\frac{\partial u\left(t,x,y\right)}{\partial x}\right]}^2\right]}^{\frac{3}{2}}}}={\displaystyle \frac{\frac{\partial p}{\partial x}}{{\left[1+p^2\right]}^{\frac{3}{2}}}}\\ {\kappa }_y={\displaystyle \frac{\frac{{\partial }^{2}u\left(t,x,y\right)}{\partial x^2}}{{\left[1+{\left[\frac{\partial u\left(t,x,y\right)}{\partial x}\right]}^2\right]}^{\frac{3}{2}}}}={\displaystyle \frac{\frac{\partial q}{\partial x}}{{\left[1+q^2\right]}^{\frac{3}{2}}}}\end{array}\right.$$

(16)

According to the identification results of the strike-slip faults of the cratonic basin with the maximum positive curvature and the minimum negative curvature attributes, the general trend of the main fault can be seen, but the internal details of the branch fault and the main fault cannot be identified and are significantly disturbed by the non-fault areas (Figure 5a,b).

(2)

Dip

The dip attribute reflects the change in the dip angle, and it is effective in depicting the dominant section with a large fault distance but is insensitive to small faults, which are affected by the data quality [51].

To calculate the inclination property, the instantaneous wavenumbers of the horizontal direction and the depth direction, $L_x\left(x,y\right)$ and $L_y\left(x,y\right)$, are first calculated:

$$\left\{\begin{array}{c}{L}_x\left(x,y\right)={\displaystyle \frac{u\left(\frac{d^{H}u}{dx}\right)-u^H\left(\frac{du}{dx}\right)}{{\left(u\right)}^2+{\left(u^H\right)}^2}}\\ L_y\left(x,y\right)={\displaystyle \frac{u\left(\frac{d^{H}u}{dy}\right)-u^H\left(\frac{du}{dy}\right)}{{\left(u\right)}^2+{\left(u^H\right)}^2}}\end{array}\right.$$

(17)

where $u$ is the input data, the superscript $H$ represents the Hilbert operator, and the instantaneous inclination $\theta$ can be calculated with

$$\theta ={tan}^{-1}\left({\displaystyle \frac{l_x}{l_y}}\right)$$

(18)

According to the identification results of the strike-slip faults in the cratonic basin based on the dip angle attribute, the strike-slip faults are relatively obvious. However, there is little difference between them and the data points in the non-fault area; that is, the boundary between the faults and non-faults is not clear (Figure 5c). At the same time, the internal details of strike-slip faults are unclear and difficult to describe.

4.3. Correlation Attributes

Correlation attributes reflect the similarity of seismic datasets at different locations. The essence is to reflect the continuity of strata in a certain property by analyzing the proximity and dissimilarity between data. Correlation functions can be used to detect seismic channel discontinuities and predict stratigraphic discontinuities due to different changes in the lithology, overcomplexities, and unconformities.

(1)

Coherent

The coherence attribute is used to calculate the similarity between adjacent seismic tracks and analyze the transverse changes in strata and lithology in the same phase axis to achieve fault identification. Many methods can be used to calculate the coherent volume, including the cross-correlation method, multi-channel similarity method, feature-based structure method, and gradient structure tensor (GST) method [52]. The energy of the mean amplitude of all seismic channels in the time window divided by the sum of the energies of all seismic channels represents the discontinuity of seismic waveforms, as shown in Equation (19):

$$c_s={\displaystyle \frac{\sum _{k=K}^{+K}{\left(\frac{1}{J}\sum _{j=1}^J\left[u\left(k∆t-p{x}_j-q{y}_j\right)\right]\right)}^2}{\sum _{k=K}^{+K}\left({\frac{1}{J}\sum _{j=1}^J\left[u\left(k∆t-p{x}_j-q{y}_j\right)\right]}^2\right)}}$$

(19)

Faults are often mixed with other geological sequence features and structural features that can cause seismic amplitude changes (Figure 6a). Especially when the signal-to-noise ratio of the seismic dataset is low or the geological structure is complex, the fault or other discontinuity features shown by the coherence attribute are generally effective in horizontal sections or stratigraphic sections but difficult to identify and interpret in vertical sections.

(2)

Likelihood

The likelihood attribute enhances the difference between fault and non-fault responses. The likelihood attributes of the inclination and dip of each data sample point are scanned, and the maximum value is obtained when accurate inclination and dip are scanned [52].

The semblance value range is 0~1, and the formula is as follows:

$$\left\{\begin{array}{c}Semblance={\displaystyle \frac{{\left[{\left(g\right)}^2\right]}_f}{{\left[{{(g}^2)}_s\right]}_f}}\\ Likelihood=1-Semblance\end{array}\right.$$

(20)

In the equation, $Semblance$ is the similarity attribute; $g$ is the 3D data volume; the subscript $s$ represents the constructed information, which guides smoothing; and the subscript $f$ is the dip angle enhancement.

According to the identification results of the strike-slip faults in the cratonic basin with the maximum likelihood attribute, the strike-slip faults are not clear, and there is much pseudo-fault information. Thus, it is not suitable for fine identification (Figure 6b).

(3)

Ant tracking

The principle of ant tracking technology is to distribute a certain number of “artificial ants” in the seismic dataset according to the initial boundary setting value and let each “ant” move forward along the possible fault plane under the constraints of the tracking parameters [53]. At the same time, a “pheromone” will be implemented to make obvious marks if a fault or fracture is encountered.

If a certain number of ants are placed into the seismic dataset, the initial migration probability of each path is the same, which is a constant close to zero. The ant selects the location of the next destination point according to the pheromone concentration on the path, the specified step size, and the distance between points, and its transfer probability $P(i,j)$ is

$$P\left(i,j\right)={\displaystyle \frac{{\tau }^{\alpha }\left(i,j\right)·{\eta }^{\beta }\left(i,j\right)}{{\sum }_{\left(u,v\right)\in R}{\tau }^{\alpha }\left(i,j\right)·{\eta }^{\beta }\left(i,j\right)}}$$

(21)

where $P\left(i,j\right)$ is the probability of finding the next node after the selection of the starting point and before the end of the termination condition, $R$ is the set of all the points within the visible range at the current position $(i,j)$, $\tau$ is the pheromone, $\eta$ is the heuristic, and $\alpha$ and $\beta$ are the weight coefficients of the pheromone and heuristic, respectively.

The pheromone concentration of ants along the path increases and evaporates over time. The pheromones along each path are updated as follows:

$$\left\{\begin{array}{c}{\tau }_{ij}\left(t+n\right)=\left(1-\rho \right)·{\tau }_{ij}\left(t\right)+∆{\tau }_{ij}\\ ∆{\tau }_{ij}={\sum }_{k=1}^m∆{\tau }_{ij}^k\end{array}\right.$$

(22)

where $\rho (0<\rho <1)$ is the evaporation coefficient of the pheromone on the path, $1-\rho$ is the persistence coefficient of the pheromone on the path, $∆{\tau }_{ij}^k$ is the amount of pheromone left by ant $k$ from node $i$ to node $j$, and n is a certain period of time. If ant $k$ does not take this route, $∆{\tau }_{ij}^k=0$, and m is the number of ants.

According to the results of the identification of the strike-slip faults in the cratonic basin by ants, the main faults are clear, and the identified faults are more detailed (Figure 6c). Although several major branch faults can be seen, the whole branch is a network that is significantly affected by false faults, and it is difficult to distinguish true and false faults.

(4)

AFE

Auto fault extraction (AFE) is a directional weighted coherence, which is obtained by further directional filtering based on sharpening [54]. The specific principle is as follows: through preliminary processing of the seismic dataset, the fault and the direction of the stress and tectonic fault development zone are determined, a certain algorithm is used to detect and determine the gradient change direction of the fault coherence value, and finally, directional filtering is implemented. The algorithm flow is as follows:

(1)

Linear enhancement processing. Strip noise caused by environmental factors is eliminated by a linear filter. Then, the linear contour (fault) is enhanced on the time slice of the data volume to enhance the fault recognition ability.

(2)

Direction-weighted processing. The coherent linear characteristics of the favourable direction are retained by the directional filter, and the coherent characteristics of other directions are removed. Combined with other information, such as the logging analysis, the direction of the stress, fault, and fracture development zone can be determined, and the fracture along the favourable direction can be enhanced by the directional filtering method.

(3)

Fault enhancement processing. Further denoising and plane enhancement are carried out on the linear enhancement data body generated in the first step. Relevant plane parameters are set through the fault dip angle and azimuth angle, and non-vertical linear strips on the fault time slice are filtered out, thus eliminating the discontinuity caused by non-faults and formation occurrences in the data body. After these two steps of processing, linear reinforced bands reflect the faults and fractures.

(4)

Generation and processing of vertical fault seed vectors. Vertical fault profiles are processed with fault enhancement, linear enhancement, and dip enhancement data generated in the first two steps, setting relevant parameters to search for linear structural features with dips greater than 45 degrees and transforming them into seed vectors. The minimum length of the vector is set, and any seed vector smaller than the set value is removed.

(5)

The horizontal and vertical fault vectors are connected to form the fault plane. In this step, the vertical vector and the intersecting horizontal vector are searched and connected by setting relevant fault parameters as the new initial fault.

According to the results of the AFE identification of strike-slip faults in the cratonic basin, the whole area is significantly disturbed by non-faults, and the branch faults are not clear, which is not suitable for fine fault characterization (Figure 6d).

4.4. Gradient Attributes

The amplitude spectrum gradient of the seismic dataset refers to the change rate of the seismic reflected wave amplitude with frequency in the effective frequency band, which highlights the variation characteristics of the seismic dataset amplitude at different frequencies. The strike-slip fault will lead to the wrong movement of strata on both sides, which makes the original uniform physical properties discontinuous. The gradient-like attribute body can sensitively capture the seismic waveform changes caused by discontinuities.

(1)

Amplitude variance

The seismic variance body attribute is based on error analysis, which describes the geological structure data mainly through the similarity attribute of adjacent seismic signals [55]. Based on the seismic variance body attribute, the discontinuous fault and fold relationship between geological structures can be expressed.

The specific steps for calculating the amplitude variance of this point are as follows [56]: (1) take the upper and lower half of the sampling points in the window, and first calculate the average amplitude value of all sampling points in each line of $n$ channels of the seismic dataset in the window; (2) calculate the sum of the variance in the amplitude value and the mean value of the amplitude of each sampling point and the $n$-channel data at the same time; and (3) multiply the weighting coefficient and normalize to obtain the variance value of this point, move by the window and iterate steps 1, 2, and 3 to obtain the variance value of each sampling point of the whole data body of the work area, and then obtain the variance body.

According to the identification results of strike-slip faults in the craton basin with amplitude variance attributes, the main and branch faults are relatively clear, but the whole area is disturbed by a large number of noise points, which makes the attribute body as a whole appear chaotic and not suitable for the fine characterization of strike-slip faults (Figure 7a).

(2)

Amplitude gradient

The core idea of the fault prediction method based on amplitude gradient vector disorder detection is to assume that a fault plane is a plane in the local area. By searching the disorder of the seismic amplitude gradient vector in each azimuth and dip angle in 3D space, the most disordered surface is found to be the fault location [57].

The messiness of the amplitude vector needs to be searched along a certain direction of a seismic sample point, and the gradient construction tensor field expression is constructed as shown in Equation (23):

$$\left(s,v\right)=\left[\begin{array}{ccc}\int D_x^{2}S{W}_N\left(s,v\right)& \int D_x{D}_{y}S{W}_N\left(s,v\right)& \int D_x{D}_{t}S{W}_N\left(s,v\right)\\ \int D_x{D}_{y}S{W}_N\left(s,v\right)& \int D_y^{2}S{W}_N\left(s,v\right)& \int D_y{D}_{t}S{W}_N\left(s,v\right)\\ \int D_x{D}_{t}S{W}_N\left(s,v\right)& \int D_y{D}_{t}S{W}_N\left(s,v\right)& \int D_t^{2}S{W}_N\left(s,v\right)\end{array}\right]$$

(23)

where $D_x$, $D_y$, and $D_z$ are the rates of change in the seismic amplitude along $x$, $y$, and z with time $t$, respectively; $S{W}_N\left(s,v\right)$ is the smoothing factor along the direction $v$; and the smoothing function is the multi-point Gaussian function. After establishing the matrix in direction $s$, Equation (24) is used to obtain the disorder property of the amplitude vector:

$$F\left(s,v\right)={\displaystyle \frac{3}{2}}{\displaystyle \frac{{\lambda }_2\left(s,v\right)+{\lambda }_3\left(s,v\right)}{{\lambda }_1\left(s,v\right)+{\lambda }_2\left(s,v\right)+{\lambda }_3\left(s,v\right)}}$$

(24)

${\lambda }_1$, ${\lambda }_2$, and ${\lambda }_3$ are the first, second, and third eigenvalues of $T$, respectively.

Based on the seismic amplitude data, this method can directly search the distribution law of faults, and it is simple and efficient. The results show that the fault can be interpreted well both in the horizontal section and vertical section, and it is a good 3D automatic fault tracking scheme.

According to the identification results of strike-slip faults in the craton basin with amplitude gradient attributes, both main and branch faults can be clearly shown (Figure 7b). At the same time, many details in the main fault can be detected with minimal interference from non-breaks. This attribute is a suitable attribute type for strike-slip faults in the cratonic basin.

5. Fault Identification through Attribute Fusion

5.1. Fault Identification Comparison of Preferred Attributes

The ant tracking attribute is the industry’s most commonly used algorithm for fault identification, facilitating automated recognition of faults and fractures in seismic profiles. This algorithm offers clarity in identifying and characterizing faults while also facilitating the analysis of their direction and distribution through profiles and sections, including strike, azimuth, and other features. Nonetheless, when applied to the heterogeneous deep carbonate reservoirs in the Tarim Craton Basin, the ant tracking attribute exhibits limitations, such as a reduced recognition ability for carbonate faults and challenges in identifying fine faults (Figure 8a,b). For instance, in horsetail sections of strike-slip faults, the presence of geological information beyond fractures leads to the identification of numerous pseudo-faults with disordered forms, yielding unreliable results. Furthermore, the ant tracking attribute proves less accurate in identifying small-scale faults, which are often of significant interest in tight carbonate reservoir exploration.

The amplitude gradient attribute proves highly effective in enhancing the identification of branch faults and in capturing secondary faults and most fractures that are often undetectable through coherence and edge detection methods. Compared to conventional ant tracking techniques, this attribute-based detection technology significantly reduces noise interference and false fault information, providing the precise localization of fault development. Notably, the amplitude gradient attribute outperforms the conventional ant tracking attribute in accurately revealing the trend of the main strike-slip fault and further uncovering the structural style of the fault (Figure 8c,d). Though the highly irregular amplitude gradient attribute can act as a reference pinpointing the main fault’s location within deep carbonate rocks of strike-slip faults, ant tracking offers a more intricate depiction of the fault’s internal structure. Regarding branch faults of strike-slip faults, the amplitude gradient attribute successfully reflects the trends and locations of these faults, whereas ant tracking often exhibits excessive disorder and interferes with the accurate determination of branch fault locations, presenting false information. Consequently, a combined utilization of the advantages of both attributes is recommended for the identification of craton strike-slip faults.

5.2. Attribute Fusion Based on Image Processing

The ant tracking and amplitude gradient attributes were combined to obtain high-precision and accurate fault images. First, the grid subdivision method was used to divide the amplitude gradient attribute image into many small cells. To better reflect the law of data change, a relatively dense grid was used. Then, the wavelet transform was applied to fully highlight the characteristics of the fault structure; the spatial frequency could be localized for analysis, and the signal (function) could be gradually refined at multiple scales through the expansion and translation operation to distinguish fine faults [58]. Then, because K-means can cluster similar data based on the distance between data points, the data were classified into two clusters through K-means clustering: fault clusters and non-fault clusters [59]. Moreover, the locations of the attribute points (fault data points) were extracted. The data points of the fault cluster were set to 1, and the data points of the non-fault cluster were set to 0 (Figure 9a). Then, the ellipsoid expansion method was applied to treat the attribute points that were higher than the threshold value. Due to the anisotropy of the ellipsoid, the original normal line may not be perpendicular to the new ellipsoid after expansion (Figure 9b). By setting the structural unit, the range of the extracted attribute points could be expanded according to geological theory to obtain the data body of the fault location.

To integrate the ant tracking attribute with the data body of the fault location, a uniform grid size was applied to process the ant tracking attribute. The “or and no” rule was employed to multiply the ant tracking attribute and the fault location data body, resulting in a data body with limited fault location information. Subsequently, the data volume and amplitude gradient attribute were combined using the wavelet transform method, yielding a new fusion attribute. This fusion attribute precisely identifies the fault’s location and offers a detailed representation of the fault through high resolution.

5.3. Comprehensive Characterization of Strike-Slip Faults

(1)

Interpretation map of the strike-slip faults

Based on fusion attributes, the segmentation of strike-slip faults becomes evident in plane images. Typically, the strike-slip section exhibits a linear structure with a straight plane extension and lacks secondary faults. However, some segments display oblique distributions, forming tensioning structures that converge downward, leading to a unified cross-section (Figure 10). Shallow parts of the fault are primarily formed by oblique structures, while deeper sections are predominantly influenced by compressional and torsional faults. Small grabens may arise due to the formation of oblique fault assemblages with left-lateral or right-lateral steps, resulting from local tension–torsional stress. The presence of fault clavicles with significant displacements can lead to the formation of grabens. In this study area, most derived faults manifest as small faults along the NW R shear plane, forming normal landform fault blocks with the main faults. At the divergent tail end of strike-slip faults, horsetail structures, which are characterized by a series of slanted faults arranged in a divergent pattern, commonly occur. Due to the varied types and styles of different segments, a complex sectional combination emerges, comprising vertical linear segments, oblique segments, overlying segments, graben barrier segments, and horsetail segments.

(2)

Section interpretation of the strike-slip faults

In the analyzed section, the strike-slip fault exhibits a high and steep nature, extending down to the basement. Although some minor faults may display unclear sections, most exhibit flexes of wave groups. Additionally, the larger strike-slip fault zone displays positive and negative flower-like structures. Notably, this area shows the development of relay/cline faults and braided structures, horsetail structures at the southern end of the fault zone, and typical faults and short axis/dome anticlines (Figure 11f). These distinct strike-slip fault markers aid in the identification of strike-slip faults, with minimal vertical displacement and slight convexity, indicating a compression–torsion structure (Figure 11b). The main fault is nearly vertical, reaching depths of up to 6000 m (Figure 11d). Smaller secondary strike-slip faults are usually vertical, high, and steep, with a single section breaking down to the basement and typically ending at 6500 m, extending in a straight line (Figure 11e). Moreover, the main faults induce an evident structural amplitude at depths of 6000–7000 m, with many exhibiting a “normal flower-like structure” dominated by upper convexities. Branch faults are formed in the upper part of the carbonate rocks, spreading upward through back thrust, while fault barriers form on the top of the carbonate rocks. The section features high steepness, downward convergence, and merging, with distinct “extrusion, reverse fault, and back shape” structural characteristics (Figure 11c). In regions influenced by the combination of left-lateral and right-lateral faults, clear grabens and negative floral structure features of “tensile, positive fault, and trending” are evident (Figure 11a). At depths of 7000–8000 m, a “negative flower-like structure” is more prevalent, resulting from inherited fault activities due to tensional torsion.

6. Discussion

6.1. Strike-Slip Fault within the Craton Identification Workflow

The accurate characterization of seismic datasets in deep cratonic basins with strike-slip faults is a challenging task. This difficulty arises from the steep fault sections and closely spaced faults, which make it challenging to identify smaller faults. Furthermore, the misalignment of rock layers on either side of the fault and substantial lateral variations add complexity to the precise imaging of the target layer in seismic datasets. To enhance the signal-to-noise ratio and spatial resolution, principal component filtering, construction-oriented filtering, and fault-focused imaging were employed in this study to protect low-frequency effective signals and extract residual effective signals from noise. Despite achieving a high degree of in-phase axis continuity in the seismic dataset following quality control, the intricate structure and pronounced heterogeneity within the deep cratonic basin make it challenging to pinpoint breakpoints and fault locations. This complicates the direct utilization of seismic datasets for structural interpretation and reservoir prediction. To overcome these challenges, seismic attributes, encompassing parameters like amplitude, phase, frequency, continuity, and morphology, provide a quantitative foundation for discerning geological features and assisting in fault interpretation across various scales. Nonetheless, when we identify and compare multi-attribute strike-slip faults, it becomes clear that traditional attributes have limitations in effectively depicting faults within the craton basin. These limitations are especially pronounced in the deep carbonate strata experiencing intricate tectonic transformations, resulting in less accurate characterizations of the strike-slip faults.

Recognizing the difficulties involved in pinpointing strike-slip faults within a cratonic basin, this study introduces a hierarchical identification approach. It classifies the faults into primary and subsidiary categories. Both outcrop and seismic dataset analyses validate the prevalence of strike-slip faults, which play a pivotal role in shaping fault combinations and influencing reservoir development. The high-angle development of the main fault section, large fault distances, strong transverse and longitudinal connections, and significant deformation differences on both sides of the local section allow most conventional seismic attributes to accurately locate the main fault. Therefore, a synthesis of various attributes can determine the development location of the main fault. Identifying branch faults is a challenge for many traditional attributes due to the presence of brief faults and closely spaced faults, and their dense arrangement, resulting in either detection issues or a high number of false positives. Branch faults play a pivotal role in oil and gas development and accumulation, making the identification of craton strike-slip faults a crucial and complex research area, with the new attribute being the focus of investigations.

6.2. Seismic Attribute Fusion Technology

Seismic attribute fusion can synthesize the advantages of multiple attributes, and it is a favourable tool for identifying strike-slip faults in cratons with complex tectonic evolutions. Based on the analysis and comparison of twelve seismic attributes, two attributes that are most suitable for the identification of deep strike-slip faults in a craton were selected in this paper: the amplitude gradient attribute and the ant tracking attribute. The ant tracking method is a fully developed fault identification attribute in industry. Ant tracking was used to analyze meso-scale fractures as natural fracture corridors; however, for branch faults with complex and small slip spacings, ant tracking predicted a lot of false fault information. The amplitude gradient attribute measures the rate of change in amplitudes along a seismic trace and has also been used for fault identification [10]. The amplitude gradient attribute was used to identify faults to predict the position of reservoirs and obtained results that are clearer and more accurate in the section; however, the detailed characterization is insufficient in the plane section [12]. The main reasons are as follows: (1) in the ant tracking results, faults of different scales are often mixed with sequence and sedimentary characteristics, especially when seismic dataset noise or geological sedimentary background is complex, and it is difficult to distinguish the fake from the real; (2) the interpretation of the ant tracking data is poor in vertical sections of deep strata with complex structures; and (3) the amplitude gradient attribute is based on searching the most messy direction of the amplitude gradient to find the reliable location of the fault, which also leads to a low resolution of the fault inside [60].

Image fusion technology, through image processing and computer techniques, enables the extraction of optimal information from each channel, enhancing the utilization rate of image data and improving interpretation accuracy and reliability [61]. Drawing from the success of image fusion in various fields, mesh subdivision, threshold analysis, ellipsoid expansion, wavelet transform, and other algorithms are implemented in this study to fuse the ant tracking and amplitude gradient attributes, resulting in high-precision and high-resolution fault images. The fused attributes produce a single, detailed image that highlights potential fault locations within the seismic dataset. Utilizing image fusion technology, the subsurface geology can be depicted with greater detail and accuracy compared to individual seismic attributes alone. The reliability of prediction outcomes is confirmed through comparisons with actual drilling and production data in various aspects [62]. But the primary disadvantages of this research include not incorporating deep learning, limitations in feature extraction, dependence on expert domain knowledge, difficulty in handling noisy data, scalability issues with large datasets, limited adaptability to varying geological conditions, potential interference from other geological features, longer processing times, and challenges in capturing non-linear relationships within the data. Furthermore, the fault prediction results obtained from this method offer significant guidance for fault mode studies and oil and gas production endeavours.

The results of this study have significant implications for geological exploration and oil and gas production, particularly in improving the accuracy of fault identification within complex cratonic basins. By enhancing the prediction of reservoirs, especially in deep carbonate strata, our methods provide critical guidance for more efficient resource extraction. However, the approach also has limitations, including challenges in handling noisy data, adapting to diverse geological conditions, and capturing non-linear features. Additionally, the reliance on expert domain knowledge may limit its generalizability. To address these issues, future research could explore integrating deep learning techniques with our multi-attribute fusion method, potentially leading to breakthroughs in fault detection accuracy and scalability. Such advancements would improve the method’s applicability across different geological settings and enhance its robustness against noise, ultimately contributing to more reliable subsurface imaging and resource development.

7. Conclusions

In cratonic basins, distant from active plate boundaries, strike-slip faults emerge from the reactivation of old structures under cratonic stress. Their importance has been accentuated by recent hydrocarbon discoveries in these basins, underscoring their role in reservoir and hydrocarbon dynamics. Leveraging post-stack seismic data and mathematical theories, we have categorized 12 seismic attributes and particularly focused on ant tracking and amplitude gradient attributes for their fault-delineating prowess. Through techniques like principal component analysis and structure-oriented filtering, we have enhanced data resolution, producing a novel seismic attribute for detailed fault identification. A comprehensive model was established, combining profile interpretation and outcrop studies, to deepen our grasp on cratonic basin faults. We suggest prioritizing the improvement of automatic fault identification in subsequent research.