GF-2/sentinel-2 fusion for active fault detection: a case study of the yilan-yitong fault (liaoning segment)

1 Introduction

Active tectonics refer to geological structures that have remained active since the Late Pleistocene and may continue to undergo tectonic deformation, including active faults and folds. The activity of these structures directly controls the occurrence and distribution of regional seismic hazards (Deng, 2002; Xu et al., 2023; Wu and Hu, 2024). About 90% of strong earthquakes worldwide are closely related to sudden slippage of active faults. Accurate identification of fault surface distribution and clarification of their activity patterns are essential prerequisites for seismic hazard assessment and regional geological safety (Zhao et al., 2025; Valoroso et al., 2014). Remote sensing technology, with its macro-scale and efficient observation capabilities, has become an indispensable tool for active fault mapping (Ren et al., 2023; Cirillo et al., 2025), however its potential is often limited by the trade-off between spatial resolution and spectral richness.

The domestically produced Gaofen-2 (GF-2) satellite provides panchromatic imagery with a spatial resolution of 0.8 m, enabling the capture of fine spatial details and clear identification of micro-topographic features such as fault scarps and drainage offsets. (Zhang et al., 2016). The multispectral images of the Sentinel-2 satellite operated by the European Space Agency (ESA) consist of 13 spectral bands, among which three distinct red edge bands are included (Phiri et al., 2020; Wang et al., 2018; Choi et al., 2010). These red edge bands can finely characterize the spectral features of lithology, soil and vegetation, thereby providing critical information for geological stratigraphy. However, current GF-2 data are primarily limited to qualitative interpretation, and research on synergistic integration with Sentinel-2 data remains limited. In particular, there is a lack of optimized solutions for interpreting complex surface faults in areas with high vegetation cover and strong human activity interference, which limits the quantitative application potential of domestic high-resolution satellites. Image fusion technology is the core means of connecting data with different resolutions and spectral characteristics. Among them, the pixel-level component substitution (CS) method has become the mainstream path for panchromatic and multispectral image fusion due to their simplicity, high efficiency and significant spatial enhancement effect. In recent years, the academic community has conducted numerous explorations on the optimization of the CS method: Yilmaz et al. (2019) improve the Gram-Schmidt (GS) algorithm using a genetic algorithm to enhance spectral fidelity. Zhan et al. (2021) compare the fusion performance of four methods, namely, Principal Component Analysis (PCA), Nearest Neighbor Diffusion (NNDiffuse), GS, and Brovey transform, in QuickBird and GF-2 satellite data, and confirmed that all these methods achieve a remarkable improvement in spatial resolution yet exhibit distinct discrepancies in their spectral preservation capability. Wang et al. (2018) employ PCA and GS methods to fuse Landsat-8 and Sentinel-1 data, effectively reducing cloud and vegetation interference. In recent decades, remote sensing image fusion has evolved into three major technical branches beyond traditional component substitution-based pixel-level fusion: multi-resolution analysis (MRA) fusion, variational optimization (VO) fusion, and deep learning (DL)-based fusion (Zhang K. et al., 2023). MRA decomposition algorithms can sharpen subtle terrain edges, but often introduce noticeable spectral distortion and incur heavy computational costs for large-area regional mapping (Mangalraj P. et al., 2020). VO constructs mathematical energy functions to balance spatial and spectral information, yet it relies on complex parameter tuning and iterative computation, which limiting its operability for routine geological field surveys (Wang P. et al., 2024). DL-based methods have demonstrated superior performance in various image fusion tasks with sufficient training data (Vivone et al., 2024) and represent the current state-of-the-art. Given the practical requirements of large-area geological mapping, the four classic pixel-level fusion algorithms, including PCA, GS, HSV, and Brovey, were adopted as primary baselines due to their high efficiency, zero training data demand, and favorable interpretability. Furthermore, to ensure a comprehensive evaluation, we additionally compared our PCA-based method with UCGAN (Unsupervised Cycle-consistent Generative Adversarial Networks for Pan-sharpening), a representative unsupervised deep learning approach that requires no labeled training datasets (Zhang et al., 2022). However, there are still significant gaps in existing research: First, there are very few comparative studies on fusion methods for the high-potential combination of GF-2 panchromatic and Sentinel-2 multispectral data, which is difficult to meet the needs of fault interpretation in complex geological backgrounds; Second, there is insufficient practical application verification of fusion results in active fault distribution analysis and activity assessment, and there is a lack of a complete research loop from technical methods to interpretation effectiveness to geological verification.

Research on the Yilan-Yitong fault zone dates back to the 1990s, and previous studies have clarified the tectonic properties and activity of some segments. Wei et al. (2025) determined that the Fangzheng segment is a Holocene right-lateral strike-slip fault. Yu et al. (2024) revealed the tectonic deformation and segmentation characteristics of the late Quaternary period; Chang et al. (2021) inverted the horizontal slip rate of the Heilongjiang section to be about 1.57 ± 0.06 mm/a; Shu et al. (2020) revealed multiple paleoseismic events in the northern section through trenching. However, the Liaoning section of the fault traverses the cities of Shenyang and Tieling, with a total length of about 190 km. Relevant research on this Liaoning segment remains insufficient. Thick vegetation coverage, extensive agricultural cultivation and fluvial erosion have obscured surface tectonic markers, leaving key scientific questions unresolved, including the fault spatial assemblage, Late Quaternary kinematic characteristics and reliable evidence of Holocene fault activity. Existing studies mostly rely on low-resolution remote sensing data or local geological surveys, lacking detailed interpretation and trench verification based on multi-source high-resolution remote sensing fusion. These deficiencies not only hinder comprehensive understanding of fault zone activity but also compromise seismic hazard assessment and major infrastructure planning in Liaoning Province.

Therefore, this study employs fused GF-2 panchromatic and Sentinel-2 multispectral imagery as the core data source. By systematically comparing the fusion performance of four typical algorithms (PCA, GS, HSV, and Brovey) through qualitative observation and quantitative indicator evaluation, we determine the optimal fusion scheme for active fault interpretation in the study area. Relying on the optimized fused data, we carry out full-coverage and refined remote sensing interpretation of fault distribution, geometric characteristics and segmentation attributes, and then conduct field investigations on 28 key geological observation points and high-precision UAV aerial surveys on 7 typical sections. Furthermore, we excavated a 25 m × 8 m large-scale trench in the key section of Kaiyuan where tectonic activity is most significant, carried out detailed dissection of stratigraphic sequence, fault contact relationships, and structural deformation characteristics, and determined the nature of fault activity, movement mode, and Late Quaternary activity age. This study demonstrates the application potential of combined GF-2 and Sentinel-2 data for active fault detection. Additionally, it establishes a replicable workflow for precise fault identification in densely vegetated areas subject to extensive anthropogenic disturbance. In addition, it fills the research gap on the activity of the Liaoning segment of the Yilan-Yitong Fault, clarifies its Late Quaternary activity patterns and paleoseismic event sequences, and provides a high-precision scientific basis for seismic hazard assessment in the eastern segment of the Northeast Circum-Pacific Seismic Belt and for the safety layout of major projects in Liaoning Province.

2 Study area and data

2.1 Geological setting of the Yilan-Yitong Fault Zone

The Liaoning section of the Yilan-Yitong fault zone traverses the cities of Shenyang and Tieling. It belongs to the northern extension of the Tanlu fault system and extends in a NE-NNE direction. It is about 1,200 km long and is an important intracontinental strike-slip fault zone in Northeast Asia, as shown in Figure 1. This segment was initiated as a regional shear zone during the Late Paleozoic, and its peak activity occurred from the latest Cretaceous to the Paleogene, accompanied by the development of splay fault systems and graben structures (Zhang et al., 2019). It was offset by NW-trending transverse faults during the Cenozoic, forming a complex rift architecture. Rifting activity progressively waned subsequent to the Neogene. The regional geomorphology is dominated by mountainous terrain, and can be classified into three categories: depositional landforms, erosional-depositional landforms, and erosional-denudational landforms. The area has been subjected to long-term uplift and denudation since the Mesozoic (Zhou et al., 2023). Geologically, the study area is part of the Liaodong Sub-region of the North China Stratigraphic Region, with strata exposed from the Archean to the Cenozoic. Quaternary deposits are dominated by Holocene and Upper Pleistocene sequences, with minor Middle Pleistocene strata locally developed. The caprock was intruded by Indosinian alkaline complexes and emplaced by Yanshanian granitic plutons, forming an E-W trending magmatic belt. It was further overprinted by multiple phases of tectonism, leading to the development of intricate folds and fault systems. As the tectonic boundary between the Liaodong Uplift Belt and the Songliao Basin, the fault zone has been subjected to compressional forcing from the Pacific Plate since the Mesozoic. The fault zone experienced sinistral strike-slip motion during the Yanshanian orogeny, followed by a shift to dextral strike-slip kinematics in the Himalayan orogeny. At present, dextral strike-slip predominates, with significant along-strike variations in slip rate and kinematic mode across different fault segments. During neotectonic movements, fault zones controlled the differential uplift between the Liaodong uplift zone and the Lower Liaohe-Songliao subsidence zone, resulting in spatially differentiated activity: the graben margins were more active than the central depression area (Yu et al., 2016). During the Quaternary, this fault inherited its early kinematic characteristics of normal dip-slip and dextral strike-slip motion (Wan et al., 2017). The present-day coupling between the regional stress state and fault deformation behavior forms the primary tectonic basis for the occurrence of moderate-to-strong earthquakes (Yu et al., 2016).

2.2 Regional tectonic activity

The regional tectonic activity of Liaoning Segment of the Yilan-Yitong Fault exhibited multi-phase and multi-property complex characteristics. Historical earthquake records indicate that although no strong earthquakes of M ≥ 6 have occurred along the Liaoning Segment of the Yilan-Yitong Fault, paleoseismic investigations revealed its substantial deep-seated seismic potential. A surface rupture zone approximately 70 km long was identified in the Fangzheng–Tonghe area. Radiocarbon (¹⁴C) dating yielded an age of ∼1750 years BP, corresponding to the Western Jin Dynasty, suggesting a major paleoseismic event with a magnitude of ∼M7.5. This event triggered local channel offset of the Songhua River and the formation of a chain of lakes (Li, 1997), with a surface vertical displacement of 1–2 m. The present-day tectonic activity is dominated by dextral strike-slip, with a horizontal slip rate of 1.57 ± 0.06 mm/a, and also has a normal dip-slip component with a vertical slip rate of 0.57 ± 0.11 mm/a, reflecting a dynamic setting dominated by the NW–SE-oriented extensional stress field. GPS geodetic observations reveal that the horizontal deformation rate across the fault ranges from 1 to 2 mm/a, which is consistent with the present-day tectonic stress field in the eastern margin of the North China Craton. Focal mechanism solutions indicate that the dominant orientation of the regional principal compressive stress axis is NE 50°–70° with a plunge of 40°–60°, and the shallow crust is characterized by dominant strike-slip and thrust motion. The junction between this fault and the deep Chifeng-Kaiyuan fault system near the Tieling Basin exhibits significant stress concentration, rendering it a likely locus for future strong earthquakes. The Kaiyuan M5.5 earthquake occurred in 1594. Hidden segments (such as the Shenyang-Kaiyuan segment) may accumulate higher strain energy due to the locking effect. Furthermore, groundwater chemical characteristics indicate active deep fluid activity within the fault zone, suggesting that differences in fault permeability may affect local stress release patterns. In summary, the Liaoning Segment preserves clear evidence of Late Quaternary activity and has relatively high potential for strong earthquakes.

2.3 Data source and preprocessing

The GF-2 panchromatic image (0.8 m resolution) and Sentinel-2 multispectral image (10 m resolution, resampled to 0.8 m) were used as the primary data sources. The GF-2 data were obtained from the China Centre for Resources Satellite Data and Application, and the Sentinel-2 data were downloaded from the ESA Copernicus Open Access Hub. High-precision UAV imagery was employed as an auxiliary data source to extract subtle structural features that are barely discernible in satellite imagery, thus supporting the verification and revision of remote sensing interpretation results.

This study integrates two types of data, namely, satellite remote sensing and UAV mapping, to establish a multi-scale data source system. Two core satellite datasets were selected: Gaofen-2 (GF-2) and Sentinel-2. The panchromatic image of GF-2 has a spatial resolution of 0.8 m, allowing clear identification of micro-geomorphic indicators such as fault scarps and drainage offsets (Zheng et al., 2021; Wang D. et al., 2025). The Sentinel-2 multispectral image consists of 13 bands, including three red-edge bands, which can finely characterize the spectral features of lithology, soils and vegetation, providing a critical basis for geological stratification. High-precision UAV imagery was employed as an auxiliary data source to extract subtle structural features that are barely discernible in satellite imagery, thus supporting the verification and revision of remote sensing interpretation results.

A standardized preprocessing workflow is developed as follows: first, radiometric correction and precision geometric correction are performed separately on GF-2 panchromatic images and Sentinel-2 multispectral images. Ground control point matching is utilized to constrain the spatial registration error to ≤1 pixel. Subsequently, multiple image fusion tests are carried out. The optimal fusion strategy is screened by integrating qualitative visual evaluation of spatial structural clarity and spectral fidelity, alongside quantitative metrics including standard deviation, entropy and the Spectral Angle Mapper (SAM). Finally, overlay analysis between the optimal fused image and UAV imagery is implemented to refine active fault interpretation markers and deliver high-quality data support for follow-up studies.

3 Methodology

3.1 GF-2 and Sentinel-2 data fusion methods

3.1.1 Comparison and selection of baseline fusion methods

To cope with the remarkable spatial resolution discrepancy between GF-2 panchromatic imagery and Sentinel-2 multispectral imagery, and to mitigate the spectral–spatial information degradation induced by distinct sensors, this study adopts an optimized image fusion approach. Fusion yields a fused image with richer ground scene information (Ghassemian, 2016; Simone et al., 2002). We compared classic pansharpening fusion algorithms and integrated the spectral-spatial signatures of GF-2 and Sentinel-2 imagery with the geological conditions of the Liaoning segment, thereby selecting four algorithms with excellent spatial and spectral retention performance for cross-comparison: Principal Component Analysis (PCA), Gram-Schmidt (GS), Brovey Transform (Brovey), and Hue-Saturation-Value (HSV). The method with the best comprehensive performance across multiple quantitative and qualitative evaluation metrics is ultimately determined. GF-2 and Sentinel-2 data are fused to form remote sensing data with both high resolution and multispectral characteristics, hereinafter referred to as “fused data”.

The HSV color system uses Hue, Saturation, and Value as a user-oriented representation (Jiao and Xu, 2010; Cantrell et al., 2010). In HSV-based fusion, the multispectral image is transformed into H, S, and V components; the V component is replaced by the panchromatic image; and after resampling H and S to the panchromatic pixel size, an inverse HSV transform yields the fused image. The Brovey method (or ratio transform) first divides the multispectral image into R, G, and B bands via a color standard transform, then fuses these bands with the panchromatic band to generate new R, G, B bands (Zhang and Wu, 2008; Yuan and Zhu, 2018). It is simple and fast but tends to produce low-brightness fused images with significant spectral loss, and is mostly applied to three-band multispectral images. PCA is a common technique for dimensionality reduction and feature extraction (Li et al., 2021; Zubko et al., 2007). Its core principle involves calculating eigenvalues and eigenvectors of the correlation matrix among multispectral bands; the first principal component (which captures most variance) is then replaced by the panchromatic band, followed by an inverse transform to obtain the fused image (Wang et al., 2015; Uddin et al., 2021). Specific steps include spatial registration, PCA on the multispectral image, histogram matching with the panchromatic band, replacement of the first principal component, and inverse transformation (Luo et al., 2021; Wu et al., 2018). The Gram–Schmidt (GS) algorithm represents a high-fidelity pansharpening technique. It maintains spectral fidelity via orthogonal transformation to remove band redundancy and retains the original spatial, spectral and textural signatures of multispectral data (Tong et al., 2022; Zhang et al., 2022). Its steps involve: spectrally resampling the low-resolution multispectral image to simulate a panchromatic first component; applying GS transform to orthogonalize the multispectral image; replacing the first component with the actual panchromatic band; and finally inverting the transform to produce the fused image with minimal spectral loss.

3.1.2 Experimental parameter settings

Specifically, all image preprocessing and fusion procedures were implemented using ENVI 5.3. Image co-registration between GF-2 panchromatic and Sentinel-2 multispectral images was finished with a root-mean-square error (RMSE) of less than 0.5 pixels. For PCA fusion, the first three principal components (P C 1–P C 3) accounting for over 96% of total variance were retained, and the panchromatic band substituted P C 1. For GS fusion, Gram-Schmidt orthogonalization was applied to multispectral bands, taking the panchromatic band as the high-resolution reference. For HSV fusion, the intensity (V) component was replaced by panchromatic data, followed by inverse HSV transformation. For Brovey fusion, the standard Brovey transform was performed on visible R, G and B bands.

3.1.3 Comparison with UCGAN

To achieve a more comprehensive assessment of the PCA-based fusion method, we further benchmarked it against UCGAN, a state-of-the-art unsupervised deep learning algorithm. This network adopts a two-stream generator to extract modality-specific features from PAN and MS images individually, implement feature-domain fusion, and reconstruct pan-sharpened outputs.

First, to resolve the spatial resolution discrepancy between the two sensors, we extracted four 10 m spectral bands (B2, B3, B4, B8) from Sentinel-2 multispectral imagery and upsampled these bands to match the spatial resolution of GF-2 panchromatic data via bicubic interpolation. This preprocessing step guarantees identical spatial dimensions for both input modalities prior to network feeding, a mandatory condition enabling the two-stream generator to execute feature extraction and fusion. Second, to mitigate the channel count mismatch between single-channel GF-2 panchromatic images and multi-channel Sentinel-2 multispectral data, we duplicated the GF-2 panchromatic image K times along the channel axis, where K represents the number of Sentinel-2 bands adopted for fusion (K = 4 in this work). This duplication scheme enables the network to process panchromatic data as multi-channel input without distorting its spatial information, thereby matching the channel dimension of the upsampled multispectral imagery. Upon finishing the above preprocessing operations, the GF-2 panchromatic input and Sentinel-2 multispectral input exhibit consistent spatial dimensions and channel counts, meeting the input consistency constraints of the UCGAN generator (Zhang et al., 2022). All remaining experimental configurations were fully preserved in accordance with the original publication, including the RCA blocks within the generator, discriminator structural design, and the four-component hybrid loss function (cycle loss, adversarial loss, reconstruction loss, and non-reference loss).

3.2 Remote sensing geological interpretation

Prior to geological remote sensing interpretation, we systematically collate geological and tectonic datasets covering the study area. Taking fused satellite images as the primary core data, we integrate high-precision UAV imagery and DEM products to establish a complete database for subsequent tectonic interpretation (Uysal et al., 2015). Before remote sensing geological interpretation, it is necessary to systematically organize the geological and tectonic data of the study area, using fused data as the core data source, and combining high-precision UAV imagery and DEM data to build the interpretation foundation (Uysal et al., 2015). The fused images achieve substantially enhanced spatial resolution, enabling clear delineation of fine-scale geomorphic features including size, geometry, and surface texture. This capability facilitates detailed analysis of stratigraphic architecture, rock mass geometry, and tectonic spatial relationships, providing robust support for geological interpretation (Zhang et al., 2025; Li, 2023). Interpretation markers can be categorized into direct and indirect ones. Direct interpretation markers refer to the external characteristics that reflect the inherent attribute differences of geomorphic units. Linear features associated with active faults, including linear faulting, linear scarps, geomorphic units with tonal differences, and drainage deflections (Jahantigh and Ramazi, 2025), all fall into this category, enabling direct extraction of fault locations and determination of their tectonic properties. The differences in hue and texture among different strata are also a direct basis for stratigraphic identification (Figure 2). Indirect interpretation markers require inferring the nature of land features through correlation analysis, which is greatly influenced by experience and should be used with caution. For example, color boundaries can be used to distinguish landform boundaries, and texture features can be used to help determine the relationship between the ages of strata (Wang S. et al., 2024).

The interpretation techniques comprehensively integrate texture analysis, geomorphological analysis, and field UAV validation. Texture analysis facilitates the identification of lithological strata based on image textural features, whereas geomorphological analysis employs high-precision DEM data for 3D visualization (Figure 3.), which intuitively presents terrain morphology and helps determine the direction of fault displacement. These techniques combined with other methods to overcome limitations caused by excessive empirical dependence. Field UAV verification focuses on collecting images of fault scarps and micro-topographic structures, and correcting laboratory interpretation results. The specific operation employed the ArcGIS interactive interpretation method, which interpreted faults and different lithological strata through linear features and area features, respectively. A 1:50,000-scale remote sensing interpretation map was compiled, and the interpretation criteria of active faults were summarized, providing accurate basic data for subsequent fault activity analysis and regional geological research.

3.3 Geological and geomorphological survey

Based on fused data interpretation, we identified and investigated 28 key geological observation points through human-computer interactive interpretation and preliminary field reconnaissance. These observation points cover the main tectonic units, lithological distribution zones, and geomorphological evolution characteristic areas of the study area, providing a basic data framework for the analysis of regional geological patterns. Based on this, and considering the research objectives and the complexity of geological phenomena, seven key observation points (P1-P7) were selected. High-precision UAV surveys were conducted on typical geological points such as fault contact zones and lithological phase transition interfaces (as shown in Figure 3). This aerial survey employed the DJI PHANTOM 4 RTK, which is equipped with a GPS measurement system, an inertial navigation system, and an 8.8 mm fixed-focus high-definition camera. It can provide meter-level positioning accuracy and capture 20-megapixel images. The aerial survey was conducted under cloudless and well-lit conditions at an altitude of 100 m. This study employed SfM technology to extract digital 3D models of target objects based on the principle of stereophotogrammetry. The aerial survey data were processed using Agisoft Photoscan software to ensure reliability (Guillaume et al., 2021). The processing results show that the flight paths of different aerial survey flights overlap by ≥ 50 m, with no missing aerial photography areas, and the image overlap density is ≥6 images; the average density of topographic point clouds is 88.9 points/m², and the pixel resolution of orthophotos is 2.5 cm. Subsequently, DEM data with a grid size of 0.1 m were generated via the natural neighbor interpolation method using Surfer software. Its accuracy and resolution are sufficient to fully reflect the actual characteristics of the landform surface and satisfy the requirements for extracting structural geomorphological information. High-resolution UAV imagery provides precise spatial constraints for structural attitude measurements, stress field analysis, and 3D geological modeling by clearly depicting lithological outcrops, tectonic deformation, and micro-scale geomorphology. This approach effectively overcomes the limitations of traditional manual surveys in resolving fine-scale features and ensuring measurement accuracy.

Typical linear scarps were identified at key observation points P1, P2, P4, and P6 (Figure 4). These scarps generally strike NE–NNE, with lengths ranging from several hundred meters to several kilometers and heights mostly ranging between 1 and 2 m. They are stable, straight, and undisturbed by anthropogenic activities, and possess significant structural indicative value. Among them, the steep slope at P2 is accompanied by longitudinal tensile ground fissures, which exhibit right-lateral torsional characteristics and directly reveal the right-lateral strike-slip nature of the fault. Significant differences exist in slope gradient, basement lithology, and Quaternary strata distribution on both sides of the P3 steep slope. UAV imagery clearly reveals the fault characteristics of this steep slope, showing a linear connection with the micro-topographical steep slope developed by the P1 fault. These features collectively confirm the neotectonic activity of the Yilan-Yitong fault zone during the Upper Pleistocene of the Quaternary period. The fault scarps developed at point P6 are arranged in a left-handed, oblique pattern, and the gullies show right-lateral displacement characteristics, further confirming that the fault is dominated by right-lateral strike-slip motion, while also having a vertical slip component. The aforementioned steep slopes are mostly developed within geomorphic units such as piedmont slope aprons and slope-alluvial fans, and their orientation is highly consistent with the strike of the main and secondary structures of the Yilan-Yitong Fault Zone, thus forming an important geomorphic response to fault activity. Small fault scarps are visible along linear geomorphic traces near the surface at Point P5. These scarps have a limited extent and are slightly curved in local segments, reflecting spatial heterogeneity in fault activity. Although the main fracture line at P7 is not fully exposed due to engineering modifications, it still retains faint traces of topography. Integrating fault-associated micro-geomorphic evidence from multiple field observation points, we verify that the Yilan-Yitong fault zone sustained obvious tectonic activity throughout the Late Quaternary. This not only enriches the geomorphic evidence chain of tectonic activity in this fault zone, but also provides important field support for analyzing its kinematic characteristics and potential seismic hazards.

4 Result

4.1 Evaluation and analysis of fusion results

4.1.1 Qualitative evaluation

Image fusion quality directly affects the accuracy of image interpretation, and different fusion algorithms produce distinct fusion results. This study compares the fusion performance of four common methods: HSV, Brovey, PCA, and GS.

Figure 5 Shows the fusion results of GF-2 and Sentinel-2 images, including a comparison of local details of buildings, landforms and other features after fusion. The results show that the spatial resolution of the fused original multispectral images is improved, while the complete spatial information of the panchromatic image is preserved, making the clarity of the fused image close to that of the panchromatic image. This is the advantage of component substitution. Visual assessment indicated that PCA and GS outperformed the other two methods in preserving spatial detail and spectral fidelity, thereby maintaining more authentic surface characteristics. From the perspective of spectral characteristics, the images fused by the Brovey and HSV methods show significant changes in the color of landforms and water bodies.

4.1.2 Quantitative evaluation

Evaluation of fusion performance is an important component of image fusion processing, which involves analyzing and testing fused images, evaluating the efficiency and results of fusion algorithms, and adjusting relevant parameters (Ren J. et al., 2019). This study evaluates the fused images from two aspects: spatial information integration and spectral information fidelity. For spatial information integration, four evaluation indicators were selected: information entropy, average gradient, spatial frequency, and structural similarity (Table 1). For spectral information fidelity, five evaluation indicators were selected: relative deviation, gray mean, correlation coefficient, spectral distortion, and SAM (Table 2).

TABLE 1. Spatial resolution evaluation indice.

TABLE 2. Spectral fidelity evaluation indices.

Visual interpretation struggles to distinguish differences in features such as riverbanks and stratigraphic boundaries within the fused image (Huang et al., 2021). Therefore, quantitative evaluation is needed to further assess the fusion effect. We assessed the four fusion methods (PCA, GS, HSV, and Brovey) using standard quantitative indicators. IDL programming was used to first calculate the single-factor indicators of each band of the original multispectral image and the fused image, and then the average value of each single-factor indicator was calculated. The results are shown in Table 3. As shown in the table, PCA has the highest information entropy, GS has the best average gradient, followed by PCA; the HSV fusion method has the best spatial frequency performance. The larger the entropy and average gradient, the more spatial information the image contains and the better the expression of spatial details. In terms of image sharpness and sensitivity to spatial information, PCA and HSV methods are the best. In terms of spectral characteristics, PCA performs best in terms of relative deviation, gray mean, correlation coefficient, and spectral distortion. The smaller the deviation, the smaller the mean change, the closer the correlation coefficients, and the higher the spectral fidelity. SAM results further confirm PCA fusion’s spectral superiority: its lowest SAM value ensures the closest spectral consistency with original Sentinel-2 data, enabling accurate identification of lithological boundaries and vegetation along fault zones. Conversely, HSV fusion has the highest SAM value, with severe spectral distortion misleading the recognition of fault-related alluvial deposits and microtopography. To quantitatively evaluate the practical utility of each fusion method for active fault identification, we calculated the fault identification accuracy based on the consistency between remote sensing interpretation results and 28 field validation points. Specifically, a site was counted as correctly identified when the fused image allowed clear visual recognition of fault-related linear features at the corresponding field-confirmed location. As shown in Figure 6, PCA fusion achieved the highest accuracy, significantly outperforming the other three methods. This ranking is fully consistent with the visual assessment (Figure 5) and the quantitative image quality metrics (Table 3), confirming that PCA fusion not only produces superior image quality but also translates directly into better geological interpretation performance. Although GS achieves the highest average gradient among the four methods (Table 3), its lower spectral fidelity leads to lower identification accuracy than Brovey, underscoring the importance of spectral preservation in fault mapping. Therefore, the PCA is the most effective at preserving spectral information, and the smaller the spectral distortion, the better the image fusion result. Compared with the other three fusion methods, the PCA retains more spectral information of the original multispectral image, and is more complete in expressing spatial details, resulting in better performance.

TABLE 3. Fusion performance evaluation indices.

4.1.3 Comparison with UCGAN

To comprehensively validate the performance of the PCA-based pansharpening fusion framework against state-of-the-art unsupervised deep learning fusion models, we select UCGAN as the comparative baseline and carry out qualitative visual assessment and quantitative index evaluation simultaneously. Figure 7 presents a qualitative visual comparison between PCA and UCGAN over a representative fault-geomorphic area. As indicated in the figure, PCA fusion preserves slightly clearer structural details, particularly along fault lineaments and micro-topographic boundaries, whereas UCGAN produces relatively smoother edges with marginally reduced sharpness. This visual assessment is quantitatively supported by the evaluation metrics reported in Table 4. Three widely used metrics are adopted: SAM (Spectral Angle Mapper); ERGAS (Erreur Relative Globale Adimensionnelle de Synthèse); and SSIM (Structural Similarity Index). In terms of spectral fidelity, UCGAN achieves a slightly lower SAM value, indicating marginally better color preservation. However, PCA demonstrates a superior SSIM value, reflecting better structural information retention—a critical factor for identifying fault traces and subtle geomorphic features. The ERGAS values are comparable, suggesting that both methods achieve similar overall quality. Considering that active fault identification relies more heavily on structural and textural information than on absolute spectral accuracy, the SSIM advantage of PCA, combined with its computational efficiency and physical interpretability, makes it more suitable for large-scale geological mapping applications in this study area.

TABLE 4. Quantitative comparison between PCA-based fusion and UCGAN.

4.2 Detailed structural interpretation results

Based on the evaluation of the fusion effect, this study used the PCA method to fuse GF-2 and Sentinel-2 images (referred to as PGS data). Since most of the PCA information is concentrated in the first three principal components, the PGS321 false-color composite image was selected as an auxiliary image for remote sensing geological interpretation. The main fault in the study area is the Shenyang-Siping segment of the Yilan-Yitong fault zone, which exhibits a clear linear characteristic in the images. The bedrock and Quaternary strata are linearly distributed, and the tectonic landforms such as fault scarps and gullies are clearly identifiable. The fault strikes NE 30°–40° and extends approximately 190 km from a branch of the Tanlu Fault Zone southwest of Shenyang to the Shiling Basin east of Siping. This study divides the Liaoning segment into three tectonic segments: Shenyang-Tieling. Tieling-North Kaiyuan, and Changtu-Siping. Based on the interpretation results of fault geometry, activity characteristics, tectonic geomorphology, and distribution of different lithologies in the study area, three remote sensing interpretation maps at a scale of 1:50,000 were compiled.

The Shenyang-Tieling section is 70 km long and consists of two parallel branch faults, spaced 2–8 km apart (Figure 8). The faults strike southeast or northwest, and the most recent movement is a normal dip-slip combined with right-lateral strike-slip. The faults are influenced by the Liuhe Fault and the Chifeng-Kaiyuan Fault at their two ends, respectively, and are located at the lower margin of the Liaohe Basin. Macroscale landforms on both sides of the fault show significant differences, whereas its microscale landform expression is weak. The fault zone hosts numerous ponds and depressions, reflecting the structural control on Mesozoic-Cenozoic basin development. Paleogene strata are dissected by the fault, and the upper part of the fault cuts into the Q₁ overburden. Quaternary strata are continuous and intact, and the fault controls their distribution, forming a topographic gradient zone. The crystalline basement is well developed, forming elongated grabens. Gravity and magnetic fields form gradient zones along the fault zone, with significant variations in crustal thickness along the zone. The two sides exhibit distinct neotectonic structural differences. The Lower Liaohe Basin has undergone intense Quaternary subsidence, with the principal compressive stress direction at approximately 70°. The fault zone is large in scale, with well-developed compressional foliation, lenticular bodies, fault breccia, and fault gouge. A Ms 5 1 2 earthquake occurred along the fault zone in historical times, and minor earthquakes are frequent currently, distributed in strips and clusters along the fault zone.

The segment of the Tieling- North Kaiyuan fault zone extends for a length of 50 km (Figure 9). It is predominantly composed of 2–3 parallel or oblique branch faults, which are spaced 3–8 km apart, dip northwestward, and exhibit both normal and strike-slip faulting characteristics. This fault line borders the Liaodong Mountains to the east and low hills and gentle slopes to the west. It controlled the development of the small Kaiyuan Basin and formed Quaternary sediments, but its extent was much smaller than that of the Lower Liaohe Basin. The eastern margin of the Kaiyuan Basin comprises a series of NNE-trending fault-controlled landforms. These include steep slopes, fault scarps, and alternating colluvial and alluvial deposits that occur discontinuously along the fault trace. This fault controlled the evolution of small Quaternary rift basins and drainage systems. Significant differences in macro-geomorphology exist on both sides, with well-developed steep slopes and ridges. The Q3-Q4 strata are displaced, and while the Quaternary strata are continuous, their integrity is poor, forming distinct gradient zones along the fault. The region is distributed across the basement, Mesozoic cover, and Yanshanian intrusive rocks. The magnetic field forms the boundaries of different anomaly zones. The geophysical field is disturbed by the Chifeng-Kaiyuan fault, and the crustal thickness gradient zone diverges. The conjugate activity of faults is complex. The Quaternary subsidence in the Kaiyuan Basin is significant, the fault zones are large in scale, and tectonic rocks such as breccia and siltstone are developed. Fault gouge is common. Historically, there have been 4 earthquakes of M 4 3 4-5 1 2. Nowadays, there are fewer small earthquakes, and they are distributed in strips.

The Changtu–Siping Segment extends northward from the northern side of the Chifeng–Kaiyuan Fault to the vicinity of Siping (Figure 10). It strikes NE–NNE, dips either SE or NW with dip angles ranging from 65° to 80°, and has a total length of approximately 70 km. This segment comprises 3–4 sub-parallel or en echelon major branch faults. Secondary faults with a consistent orientation are scarce, and faults of other strikes are poorly developed. It exhibits excellent along-strike continuity and a simpler structural geometry relative to the other two segments. The total width of the fault zone ranges from 15 to 25 km, displaying a north-widening and south-narrowing pattern. The southern termination of the fault intersects the NW-trending fault. A structural high in the basement is developed within the northern segment of the Tieling-Kaiyuan Fault. This barrier extends ∼20 km along the NE direction and is 15–20 km wide in the NW direction, with a scale comparable to that of the fault zone. The northern termination of the fault is offset by a NW-trending fault. The Shiling Basin acts as a structural barrier, impeding its connection with the Yitong segment to the north, thereby giving rise to pronounced differences in structural distribution and morphology. This fault zone is relatively small in scale, with cataclastic rocks and mylonites developed, and thin layers of fault gouge locally present. The cataclastic rocks are weakly cemented. This section is located in the Northeast seismic zone, where seismic activity is rare and there are no historical records of destructive earthquakes.

4.3 Trench verification and analysis

Remote sensing geological interpretation and field geological-geomorphological surveys provide key spatial data and surface evidence for identifying the distribution and activity indicators of the Liaoning segment of the Yilan-Yitong Fault. However, the accurate determination of the latest activity period of the fault and the assessment of seismic hazard ultimately rely on the comparison of geological profiles of typical sections and the constraints of absolute chronology. Guided by the remote sensing interpretation results (Section 3.3; Figure 4, sites P1–P2), we excavated a 25 m trench at Beiguyu, Kaiyuan City, to expose the fault architecture. This area exhibits distinct linear features in remote sensing images and continuous development of micro-geomorphic scarps. Field investigations have confirmed the occurrence of 1–2-m-high fault scarps striking NE–NNE, together with dextral deflections of gullies. This area lies in the transition zone between the piedmont colluvial plain and the first river terrace, at an altitude of ∼160 m. It is bordered by erosion-denudation hills to the east and a river valley plain to the west. With minimal human disturbance, tectonic features are well-preserved here. The trench is oriented NWW (nearly perpendicular to the fault strike), with a length of 25 m, a surface width of 8 m, a base width of 2 m, and a depth of 4.5–6.5 m. Its northern wall serves as the main observation profile, and the southern wall as the auxiliary profile, both with a dip angle of 75°. Two 1-m-wide T-shaped platforms were constructed in the middle to avoid collapse and ensure the integrity and safety of profile observation.

The trench excavation yielded ideal results, clearly revealing the shallow structural morphology of the fault (Figure 11). The faults at this site are dominated by two NW-dipping normal faults (F1, F2) and two NE-dipping faults (F3, F4), forming a typical flower structure. The flower structure identified in the trench corresponds closely to the linear fault zone interpreted from PGS data. This consistency, validated by UAV-derived microtopography, supports the use of multi-source remote sensing for 3D fault characterization. Meanwhile, the trench clearly revealed distinct Quaternary stratigraphic sequences on the eastern and western sides of the fault: the eastern (footwall) side contains 10 complete strata (① silt layer to ⑩ mud-silty soil layer), whereas the western (hanging wall) side shows obvious stratigraphic missing and a sharp reduction in thickness, with its lithological assemblage highly consistent with the sedimentary characteristics of the piedmont transition zone. The fault’s effect on the strata is very clear. F1 offsets multiple sets of Middle–Late Pleistocene strata (⑤, ⑥, ⑦, ⑧) with a vertical offset of >2 m. F2 induces obvious dislocation of relatively young strata (②, ③, ④), resulting in a horizontal wedge structure on the hanging wall, thereby evidencing multiple episodes of tectonic activity.

Comparison with well-dated trench sections nearby enabled preliminary quantitative constraints on fault activity ages: The latest faulting event of the main fault F1 occurred at approximately 7.21–13.77 ka (Terminal late Pleistocene to Early Holocene). F2 was active significantly later; its youngest faulted strata formed later than 2.02–5.02 ka (Middle to Late Holocene). Structural analysis revealed four main fault segments (F1–F4) on the northern wall, which constitute a typical flower structure:F1 and F2 dip toward the NW, with attitudes of 315°∠35° and 310°∠45°–320°∠50°, respectively; F3 and F4 dip toward the NE, with attitudes of 140°∠40° and 125°∠35°, respectively. F1 and F2 converge in a wedge geometry at the base of the profile to form the main fault segment (310°∠35°). Secondary fault segments either exhibit conjugate intersection with F2 and F3 in the upper part, or intersect with F1 at a small angle in the lower part. Fault offset characteristics are clearly distinct: F2 displaces the ③ medium-coarse sand layer and ④ silty clay layer with medium-coarse sand, with a near-surface offset of ∼0.8 m, showing west-down and east-up differential vertical movement. F4 exhibits an offset of 0.3–0.5 m, and F1 has an offset exceeding 2 m.F1 offsets strata ⑤–⑩ (formed at 7.21–13.77 ka); F2 and F3 displace strata ①–⑤ and ⑧–⑨ (younger than 2.02–5.02 ka); F4 offsets strata formed between 2.02 and 18.91 ka. These ages, constrained by stratigraphic correlation, indicate at least one faulting event during the terminal Pleistocene to Early Holocene. Confirming the full span of fault activity requires additional direct dating of older stratigraphic units. In addition, the profile further exhibits distinct structural characteristics: oblique sand wedges and silty soil wedges are developed within the “bowl-shaped structure”, with bedding significantly altered by tectonic activity. Layers ③ and ④ near F3 exhibit folding and rotation, and triangular wedge deposits (analogous to thrust-collapse clastic deposits) are identified near the surface steep slope. Based on the cross-sectional morphology and stratigraphic undulation characteristics, it is preliminarily inferred that at least one paleoearthquake event occurred, which is consistent with the tectonic setting associated with the M7.5 earthquake of the Western Jin Dynasty and the M5.5 Kaiyuan earthquake in 1594. The “bowl-shaped” tension crack filling and other features observed in the profile are preliminarily interpreted as possible paleoearthquake remains, suggesting that at least one seismic event may have occurred during the Holocene.

5 Discussion

This study employs a comprehensive methodology that integrates multi-source remote sensing data fusion, geomorphological interpretation, and trench verification to investigate the activity of the Liaoning section of the Yilan-Yitong Fault from the Late Pleistocene to possibly the Holocene, particularly within its Tieling-Kaiyuan segment. The following discussion will focus on the effectiveness, geological significance, and limitations of this methodology.

5.1 Effectiveness of the multi-source remote sensing data fusion method

The PCA fusion algorithm demonstrated the best overall performance in this study. It effectively integrated the spectral information of Sentinel-2 with the high-spatial-frequency details of GF-2 by extracting principal components through orthogonal transformation, a process that minimized spectral distortion, a finding consistent with Uddin et al. (2021). The fused data thereby clearly revealed critical micro-geomorphic indicators—such as linear fault features and fault scarps—significantly improved fault interpretation accuracy, particularly in vegetated areas. Our results showed that combining GF-2 and Sentinel-2 data via PCA fusion offered a practical, low-cost approach for mapping active faults in vegetated terrain.

Rather than merely listing isolated image quality metrics, this study jointly interpreted Figure 5 and Table 3 to analyze how each fusion algorithm affects active fault interpretation. The results demonstrated that Brovey and HSV transforms introduce severe spectral distortion, which interfered with lithological discrimination along fault zones. Though the GS method preserved abundant spatial textures, prominent color deviation blurred subtle fault offset micro-landforms. Among all tested schemes, PCA fusion achieved the optimal trade-off between fine micro-topography retention and spectral fidelity, making it the most suitable fusion strategy for large-scale active fault mapping along the Liaoning segment of the Yilan-Yitong Fault Zone.

5.2 Characteristics of fault activity and dynamic implications

The interpretation of active fault activity in this study followed a stepwise, multi-evidence chain that progressively integrated remote sensing lineaments, geomorphic indicators, UAV surveys, field observations, and trench verification (Figure 12). First, PCA-fused GF-2/Sentinel-2 imagery (Figure 5) provided the primary basis for identifying linear fault traces, drainage offsets, and lithological variations. These remotely sensed lineaments and lithological contrasts served as the initial guide for targeting areas with potential tectonic expression. Second, high-precision UAV-derived DEMs and orthophotos (Figure 4) were used to validate and refine the interpreted lineaments at the meso-scale, revealing micro-topographic features such as fault scarps and deflected gullies. Third, field investigations at 28 key observation points (Figure 4) ground-truth the remote sensing and UAV interpretations by directly documenting fault attitudes, lithological contacts, and stratigraphic relationships. Finally, the large-scale trench at Kaiyuan (Figure 11) cross-cut the interpreted fault trace, revealing the subsurface fault geometry and providing chronological constraints that confirmed the Late Pleistocene to Holocene activity inferred from the surface evidence. This integrated workflow ensured that each stage of interpretation was anchored by independent evidence from different spatial scales, thereby minimizing reliance on experience-based criteria and establishing a robust foundation for active fault identification.

The “flower structure” and evidence of multiple faulting episodes revealed in the trench indicate that fault activity here is a continuous, multi-phase process. The identified F1 (Terminal Late Pleistocene to Early Holocene) and F2 (Middle to Late Holocene) activity phases correspond to the multiple paleoseismic events documented by Shu et al. (2020) in the northern segment of the fault zone, which collectively highlighted the temporal heterogeneity of activity along the Yilan-Yitong Fault Zone. The trench evidence from this study—particularly the two activity phases at ∼7.21–13.77 ka and ∼2.02–5.02 ka—provided new constraints on the Late Quaternary behavior of the Liaoning segment of the Yilan-Yitong Fault Zone (YYFZ). These findings are broadly consistent with recent paleoseismic studies along other segments of the fault zone. For instance, a ∼150 km-long pristine surface rupture trace has been identified along the Fangzheng–Tangyuan segment, with the most recent event estimated at Mw ≈ 7.6 (Wei et al., 2025). The late Quaternary right-lateral slip rate of the Tangyuan segment has been constrained to 1–2 mm/a, with the most recent paleoearthquake (∼M 7.0) occurring between 2800 ± 600 a BP and 1700 ± 200 a BP (Yu et al., 2024). Multiple paleoseismic events have also been documented in the northern segment of the fault zone (Shu et al., 2020). Our trench results added to this growing body of evidence by providing the first direct constraints on Holocene activity along the Liaoning segment, which had previously remained poorly documented. The flower structure and multi-phase faulting events observed in our trench were consistent with the dextral-reverse strike-slip kinematics documented in other segments of the YYFZ during the Late Quaternary, forming consistent regional tectonic characteristics across the whole fault belt. Furthermore, the estimated magnitudes and recurrence patterns inferred from our study aligned with the regional seismic hazard framework suggesting that the YYFZ is the primary seismotectonic structure (M ≥ 7) posing significant earthquake threats to Northeast China (Yu et al., 2018). Combined with the continuous fault scarps and offset gullies extracted from our GF-2/Sentinel-2 fused remote sensing data, the chronological results from trenching solidly verify the Late Pleistocene to Holocene active property of the Liaoning segment from both macroscopic geomorphic and microscopic stratigraphic perspectives. These comparisons highlight that the Liaoning segment, although previously considered weakly active, exhibits clear evidence of Late Pleistocene to Holocene activity that warrants inclusion in regional seismic hazard assessments. The segmented nature of fault activity along the YYFZ—with different segments showing distinct slip rates, recurrence intervals, and timing of the most recent events—underscores that the combination of full-coverage high-resolution remote sensing interpretation and targeted paleoseismic trenching is an effective way to quantify the seismic potential of each subsection of the fault zone.

The characteristic right-lateral strike-slip with a normal faulting component aligned well with the NW-SE extensional stress field inferred from GPS data and focal mechanism solutions, reflecting the contemporary tectonic dynamic background influenced by the far-field effects of the Pacific Plate subduction. Notably, the possible Holocene activity evidence suggested that the activity level of this fault segment was higher than that implied by historical earthquake records, indicating that relying solely on historical seismicity catalogs may underestimate its long-term seismic hazard. The intersection zone between the Tieling-Kaiyuan segment and the Chifeng-Kaiyuan Fault constituted a potential strain accumulation area, warranting a higher weighting in seismic hazard models.

5.3 Study limitations and future perspectives

This study has certain limitations. Remote sensing interpretation poorly resolves fault segments buried beneath thick sediments, and trenching verification was only implemented at one key site. These limitations make it difficult to comprehensively characterize the spatial variation in activity behavior and slip rates along the entire Liaoning section of the Yilan-Yitong Fault. Future research should focus on the following areas to address these gaps: conducting more geophysical exploration and trenching across buried segments to construct a complete three-dimensional structural model and an activity segmentation map of the fault; integrating the geological evidence from this study with long-term InSAR time-series data and dense GPS measurements. This multidisciplinary approach will bridge temporal scales, improve our understanding of fault locking and strain accumulation, and ultimately provide stronger constraints for seismic forecasting research. In addition, although UAV data were primarily used for validation in this study, their high spatial resolution offers great potential for direct integration with satellite imagery. In future work, we plan to develop a multi-scale fusion framework that incorporates UAV-derived orthophotos and DEMs with GF-2 and Sentinel-2 data at the feature or decision level. Such an approach could further enhance the identification of subtle fault-related micro-topography, particularly in areas with complex surface cover or intensive anthropogenic modification.

In addition to demonstrating the effectiveness of PCA-based GF-2/Sentinel-2 fusion for active fault mapping, this study provided new geological constraints on the activity of the Liaoning segment of the Yilan-Yitong Fault. The trench evidence, combined with remote sensing interpretation and field investigation, suggests possible Holocene activity along this segment, which has remained poorly constrained in previous studies. Together, these multi-source lines of evidence offer a more complete framework for understanding the Late Quaternary behavior of this fault segment and contribute to improved seismic hazard assessment in the region.

6 Conclusion

This study targets the Liaoning section of the Yilan-Yitong Fault, a key segment of the northern Tanlu Fault Zone. By taking full advantage of the high spatial resolution of GF-2 panchromatic imagery and the rich spectral information of Sentinel-2 multispectral data, this study compares four classic pansharpening algorithms (PCA, GS, HSV, and Brovey). The PCA method is finally determined as the optimal scheme due to its superior comprehensive performance in spatial detail retention and spectral fidelity, which provides high-quality data support for refined active fault identification.

Integrating interpretations of PCA-fused imagery, high-precision UAV topographic surveys, and field geological investigations, this study clarifies the two-dimensional distribution and three-dimensional structural features of the Yilan-Yitong Fault in the Liaoning segment. The fault was divided into three structural segments: the Shenyang-Tieling, Tieling-Kaiyuan, and Changtu-Siping segments. These segments exhibit significant differences in fault assemblage patterns, geomorphic expression, and Quaternary control. Among them, the Tieling-Kaiyuan segment displays the clearest linear landforms and the strongest Late Quaternary activity.

The key advancement of this research is the trenching verification at Beiguyu Village, Lianhua Town, Kaiyuan. This work provided, for the first time, direct geological evidence for possible Holocene activity on this fault segment. The trench clearly revealed a “flower structure” composed of the F1–F4 faults and identified multiple episodes of fault displacement. Dating results indicate that the most recent activity along the main fault extended into the Middle to Late Holocene (later than 2.02–5.02 ka), with paleo-earthquake traces preliminarily identified. This study precisely identified micro-geomorphic features generated by tectonic activity since the Late Pleistocene and clarified their temporal attributes. It significantly enhanced the understanding of the seismic hazard on the Tieling-Kaiyuan segment and effectively addressed the challenges of identifying and evaluating active faults in complex surface terrains. Furthermore, it confirmed the effectiveness of domestic high-resolution satellite data, provided crucial evidence for regional seismic hazard assessment, and offered a replicable technical pathway for research in similar tectonic settings.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Tongwen Li, Sun Yat-sen University, China

Reviewed by: Chong Xu, Ministry of Emergency Management, China

Liupeng Lin, Wuhan University, China