Post-Disaster High-Frequency Ground-Based InSAR Monitoring and 3D Deformation Reconstruction of Large Landslides Using MIMO Radar

Highlights

What are the main findings?

• Dual-view MIMO GB-InSAR data with a high-resolution DEM reconstructed the 3D deformation field of a large-scale landslide.
• The landslide evolved from gradual accumulation to rapid acceleration, with a maximum vertical deformation of −100 mm.

What is the implication of the main finding?

• The method provides a rapid and reliable tool for post-disaster analysis, early warnings, and risk assessment.
• The method offers a practical framework for integrating radar monitoring with field validation in complex terrains.

Abstract

Landslide InSAR monitoring is crucial for understanding the evolutionary mechanisms of geological disasters and enhancing risk prevention and control capabilities. However, for complex terrains and large-scale landslides, satellite-based SAR monitoring faces challenges such as a low observation frequency and limited spatial deformation interpretation capabilities. Additionally, two-dimensional monitoring struggles to comprehensively capture multi-directional movements. Taking the post-disaster monitoring of the landslide in Yunchuan, Sichuan Province, as an example, this study proposes a method for three-dimensional deformation dynamic monitoring by integrating dual-view MIMO ground-based synthetic aperture radar (GB-InSAR) data with high-resolution digital elevation model (DEM) data, successfully reconstructing the three-dimensional displacement fields in the east–west, north–south, and vertical directions. The results show that deformation in the landslide area evolved from slow accumulation to rapid failure, particularly concentrated in the middle and lower regions of the landslide. The average three-dimensional deformation of the main slip zone was approximately 60% greater than that of the original slope, with a maximum deformation of −100 mm. These deformation characteristics are highly consistent with the topographic structure and sliding direction. Field investigations further validated the radar data, with observed surface cracks and accumulation zones consistent with the high-deformation regions identified by the monitoring system. This system provides a solid foundation for geological disaster early warning systems, mechanism research, and risk prevention and control.

1. Introduction

In recent years, ground-based synthetic aperture radar (GB-InSAR) has become an essential tool for high-precision, all-weather, and continuous deformation monitoring in landslide-prone areas, owing to its high spatial resolution, short revisit intervals, and flexible deployment capabilities [1]. Since its initial application in dam and landslide monitoring in the early 21st century, GB-InSAR has been widely adopted and practically implemented in countries such as Italy [2], Switzerland, Spain [3], and China [4]. In emergency landslide monitoring and early warning systems, GB-InSAR enables minute-scale continuous observation, significantly improving the timeliness and spatial accuracy of deformation detection and precursor identification [5,6,7,8]. In many cases, GB-InSAR can detect precursory deformation several hours before the onset of a landslide, thus providing important support for emergency early warning systems [9]. Moreover, in complex and remote environments such as sinkholes or steep alpine slopes, GB-InSAR can provide high-resolution and high-accuracy surface displacement fields, ensuring reliable continuous monitoring under challenging geological conditions [10,11,12,13]. High-frequency radar systems and MIMO-SAR technology have attracted increasing attention in geohazard monitoring. For example, MIMO-SAR systems such as the MELISSA prototype have been successfully applied to monitor landslides and volcanic slope instabilities, achieving sub-millimeter precision and refresh rates as high as 0.01 s, which greatly surpass the temporal resolution of conventional GB-InSAR systems [14]. Furthermore, field experiments conducted at the Vetto landslide, Stromboli volcano, and Costa Concordia wreck demonstrated the feasibility of near real-time monitoring with acquisition rates up to 125 images per second [10]. These advances highlight the significant potential of high-frequency and MIMO radar technology for the real-time monitoring of rapidly evolving geohazards.

Furthermore, ground-based synthetic aperture radar systems based on multiple-input multiple-output (MIMO) arrays, known as MIMO-SAR, have emerged as an important innovation in the field of GB-InSAR. By utilizing electronic antenna arrays and high-speed switching technology, MIMO-SAR enables SAR imaging without the need for mechanical scanning, substantially increasing data acquisition speed and spatial sampling density [15,16,17]. This allows for high-frequency, high-resolution dynamic surface monitoring [18]. Relevant studies have demonstrated that MIMO-SAR systems exhibit outstanding performance in specific applications, with acquisition times as short as 0.1 s, significantly enhancing the rapid response capability for sudden geohazards [14,19,20,21].

Traditional landslide monitoring methods present certain limitations: contact-based techniques often require equipment deployment on the landslide body, posing significant safety risks, and it is difficult to maintain continuous operation in complex post-disaster environments. Single-station GB-InSAR, constrained by line-of-sight requirements, typically yields only single-component deformation data, failing to comprehensively reflect the three-dimensional movement of the slope. Furthermore, these methods struggle to meet the precision and timeliness requirements for emergency monitoring of sudden, large-scale landslides [22,23,24,25]. At the same time, both conventional GB-InSAR and MIMO-SAR systems continue to face two major scientific challenges in post-disaster emergency scenarios: first, how to optimize radar deployment strategies in rugged terrain to achieve a balance between wide area coverage and the focused monitoring of key zones; and second, how to overcome the limitations of single line-of-sight measurements to achieve the real-time reconstruction of three-dimensional full-field displacements [26,27,28], thereby revealing the true kinematic characteristics and evolutionary mechanisms of the hazard [21].

To address these challenges, this study proposes a high-frequency three-dimensional deformation reconstruction method based on multi-angle GB-InSAR, which is specifically designed for post-disaster emergency monitoring of large-scale landslides. By leveraging the high spatiotemporal resolution of MIMO-SAR, a spatially optimized multi-angle radar deployment strategy and an efficient three-dimensional inversion workflow are developed to achieve high-precision dynamic monitoring during critical post-failure periods. Using a typical landslide in Junlian County, Sichuan Province, as a case study, the system demonstrates the potential of the proposed method for capturing the spatiotemporal evolution of landslide deformation and supporting post-disaster emergency management, thereby providing both theoretical and practical foundations for advancing geohazard emergency monitoring technologies.

2. Study Area and Dataset

2.1. Landslide Characteristics

This study focuses on a catastrophic landslide event that occurred at 11:50 a.m. on 8 February 2025 in Jinping Village, Junlian County, Yibin City, Sichuan Province (Figure 1). Under the influence of complex geological and geomorphological conditions, the landslide evolved into a debris flow, resulting in a deposit extending 1.2 km. The disaster caused the burial of ten residential buildings, with ten fatalities, two injuries, and nineteen missing persons, drawing widespread public attention.

Junlian County lies in the transitional zone between the Sichuan Basin and the Yunnan–Guizhou Plateau, characterized by high mountains, deep valleys, rugged terrain, and pronounced topographic relief. The region is dominated by a subtropical humid monsoon climate, and its landforms are primarily shaped by tectonic and erosional processes. The investigated landslide occurred on the western slope of a distinct “V”-shaped ridge system. The local stratigraphy is mainly composed of interbedded Triassic Feixianguan and Xuanwei Formations. The landslide initiated from a steep slope with an elevation difference of about 229 m and slope angles ranging from 50° to 70°. The slope failure rapidly evolved into a high-speed debris flow, which destroyed multiple houses and farmlands along its path, and finally accumulated to form an “L”-shaped deposit on the lower terrace (Figure 1a).

The affected area of the landslide disaster covers approximately 1.8 × 10⁵ m², with an average width of about 300 m, a deposit thickness of 10–20 m, and a total slide volume exceeding 1 × 10⁵ m³, mainly consisting of angular rock blocks. As the mass moved downslope, the entrainment of loose colluvium and soil facilitated its rapid transformation into a debris flow. The combination of complex topographic conditions and small-scale surface cracks and minor deformations prior to the landslide highlights the limitations of traditional geological hazard investigation methods [29]. To mitigate the risk of secondary landslides in the aftermath, monitoring equipment was deployed at the site to provide real-time detection and early warning, as illustrated in Figure 1a, with Figure 1b showing the overall morphology of the landslide and Figure 1c presenting the ground-based radar deployment perspective.

2.2. UAV Oblique Photogrammetry Model

In this study, data acquisition was conducted on 19 February 2025 using a DJI Matrice 300 RTK (M300 RTK) (manufactured by DJI (SZ DJI Technology Co., Ltd.), Shenzhen, China) unmanned aerial vehicle equipped with a five-lens oblique photogrammetry system. The flight altitude was set at 150 m, with a forward overlap of 80% and a side overlap of 70%. The digital surface model (DSM) of the study area, with a spatial resolution of 3 cm, and a three-dimensional reality model were generated using Pix4Dmapper software, version 4.7.5. During data processing [30], images acquired at different time points were registered using the Scale-Invariant Feature Transform (SIFT) algorithm, achieving a registration accuracy better than 0.5 pixels. CloudCompare software was employed to perform point cloud differencing analysis to identify potential boundaries of the landslide mass.

2.3. GB-SAR Data

Ground-based SAR monitoring in this study was conducted using two R/HYB2000 MIMO slope radar systems manufactured by Suzhou Ligong Leike Sensing Technology Co., Ltd., Suzhou, China, and developed by Beijing Institute of Technology. Operating in the Ku-band with a central frequency of 17.2 GHz, these radars provide a high spatial resolution and strong anti-interference capabilities, making them well suited for landslide monitoring in complex terrains [1,31]. The radar systems offer a spatial resolution of 0.5 m × 0.5 m, enabling the precise detection of subtle deformations within the landslide body [32]. Additionally, they deliver rapid temporal resolution, acquiring four full-field interferograms per hour over a swath width of 1.5 km × 1.0 km (

Table 1. Parameters of the employed radar system.

Parameters	Value
Rail length	1.8 m
Band	Ku
Central frequency	16.2 GHZ
Bandwidth	600 MHZ
Image acquisition time	5 min
Resolution	0.25 m × 5 mrad @ 1 km
Power	200 W

).

The monitoring period was set from 18:00 on 15 February 2025 to midnight on 19 February 2025. The radar systems were installed on stable bedrock located approximately 1.2 km opposite the landslide mass, thereby minimizing the influence of slope instability on the instrumentation itself. These systems utilize Multiple-Input Multiple-Output (MIMO) technology, which employs multiple antenna arrays for the simultaneous transmission and reception of radar signals. This enhances data acquisition efficiency and improves the processing of complex signals. After each scan, atmospheric phase screen (APS) compensation and phase unwrapping techniques are applied to further increase deformation monitoring accuracy. Moreover, a dynamic deformation extraction algorithm allows for the real-time detection of changes in the landslide mass, providing high temporal resolution that is particularly effective for identifying acceleration features during sudden slope failures (

Table 2. Monitoring periods selected for analysis.

Equipment ID	Monitoring Period	Total Time	Image Number
1	15 February, 18:00 to 19 February, 24:00	102 h	965
2	15 February, 18:00 to 19 February, 24:00	102 h	1023

).

3. Methodology

Compared with spaceborne SAR systems, ground-based SAR interferometry benefits from precise rail control, effectively achieving a zero baseline and eliminating the need for external DEM data to compensate for flat Earth and topographic phases. In addition, the GBSAR system allows flexible observation geometry, enabling the radar to be positioned according to the scene and readily detect north–south deformation components [4,13,33].

The primary algorithm for deformation measurement in GB-SAR systems is the PS-InSAR processing algorithm [2,34,35]. Due to extended monitoring periods and potentially significant scene changes, the set of permanent scatterers may vary over time. As a result, the GB-SAR system requires the dynamic selection of PS points. To maintain real-time capability, PS selection is performed using the first n images before conventional interferometric deformation grouping with the designated master image. This real-time selection ensures an adequate number of PS points within the scene at all times [36,37,38,39].

To accurately derive the three-dimensional (3D) surface deformation field of a landslide, this study proposes a joint inversion approach integrating high-precision digital elevation model (DEM) data and dual-view ground-based synthetic aperture radar (GB-InSAR) observations. Employing high-resolution DEM data to constrain topographic characteristics, this method combines line-of-sight (LOS) displacement observations synchronously obtained by two GB-InSAR systems positioned at different locations with complementary geometries, enabling the quantitative retrieval of the 3D landslide displacement (Figure 2 and Figure 3).

In practical implementation, the high-resolution DEM is first resampled and strictly co-registered to ensure consistency with the GB-InSAR data in terms of spatial resolution and the coordinate system. The DEM is then used to calculate the slope and aspect of each pixel, from which the local surface normal is derived. This information is further employed to correct the radar imaging geometry parameters, including incidence and azimuth angles, thereby ensuring that the LOS vectors are consistent with the actual terrain. Based on multi-view LOS observations, a least squares inversion is applied to derive the three-dimensional deformation components. When the observation geometry is insufficient, the surface normal derived from the DEM, together with the assumption that landslide motion typically follows slope-parallel or normal components, is introduced as a constraint to enhance the stability and physical reliability of the inversion results.

Initially, two GB-InSAR systems are strategically deployed opposite or adjacent to the landslide, providing displacement measurements along different LOS directions. Each GB-InSAR collects N1 and N2 period images over the monitoring duration. Through phase unwrapping and differential analysis, pixel-level cumulative LOS deformation time series $d_1(P,t)$ and $d_2(P,t)$ and average deformation rates $v_1\left(P\right)$ and $v_2\left(P\right)$ are derived. Simultaneously, high-precision DEM data covering the study area are acquired to calculate slope angles, slope aspects, and corresponding surface-normal vectors precisely at each pixel $n\left(P\right)=(n_x,n_y,n_z)$. All observational datasets are resampled to a consistent spatial resolution and coordinate reference, followed by rigorous spatial co-registration using ground control points.

In Earth observation, the three-dimensional displacement vector $v(P)={(v_x,v_y,v_z)}^T$ of pixel P at any given time projected onto the line-of-sight (LOS) direction $l_i$=${(l_{ix},l_{iy},l_{iz})}^T$ of the $i$-th GB-InSAR system can be expressed as follows:

$$d_i(P)=l_i^T\cdot v(P)+{\epsilon }_i$$

(1)

In Equation (1), i = 1, 2, and ${\epsilon }_i$ represents observation noise. Since the two LOS observations alone cannot uniquely resolve the three displacement components, additional physical constraints must be introduced [30]. A high-precision DEM provides the surface-normal vector n for each pixel. Assuming that the principal deformation direction aligns with the slope normal or that motion is predominantly concentrated along slope-normal or slope-tangential components, the following constraints can be applied:

$$n^T\cdot v\left(P\right)=v_n\left(P\right)$$

(2)

In Equation (2), the component $v_n\left(P\right)$ represents the principal slip direction, whose physical interpretation can be defined through geological surveys or empirical models. Consequently, the combined observation equations can be established as follows:

$$\left[\begin{array}{lll}{l}_{1x}& l_{1y}& l_{1z}\\ l_{2x}& l_{2y}& l_{2z}\\ n_x& n_y& n_z\end{array}\right]\left[\begin{array}{l}{v}_x\\ v_y\\ v_z\end{array}\right]=\left[\begin{array}{l}{d}_1\\ d_2\\ v_n\end{array}\right]$$

(3)

In Equation (3), the LOS direction vector $l_1$ for each GB-InSAR system can be accurately derived from the system’s deployment parameters and local topographic parameters extracted from the DEM. For pixel P, the LOS vector is defined as follows:

$$l_i=\left(cos{\theta }_{i}sin{\phi }_i,sin{\theta }_{i}sin{\phi }_i,cos{\phi }_i\right)$$

(4)

In Equation (4), ${\theta }_i$ denotes the radar’s azimuth angle relative to true north, and ${\phi }_i$ represents the incidence angle. A high-precision DEM is utilized to correct for the topographic slope’s effect on the actual LOS vector, ensuring the physical accuracy of the multi-view geometric constraints.

The incidence angle, defined as the angle between the radar line-of-sight (LOS) and the ground normal, influences deformation sensitivity. Larger incidence angles increase the sensitivity to horizontal components along the LOS plane, while smaller incidence angles enhance sensitivity to the vertical component. Thus, the measured LOS displacement corresponds to a projection of the true 3D displacement vector $u=(u_E,u_N,u_U)$, where the projection coefficients vary with incidence angle.

$$d\_LOS=u·\hat{\mathrm{s}}=-u\_Esin\theta cos\alpha -u\_Nsin\theta sin\alpha +u\_Ucos\theta$$

(5)

In Equation (5), θ is the incidence angle. The sensitivity coefficient to the vertical component $u\_U$ is cosθ, while the sensitivity coefficient to the “radar-facing horizontal component” is sinθ. Therefore, when the incidence angle θ changes, the projection coefficients also change, leading to the amplification or attenuation of different displacement components in the LOS measurement.

For each pixel, the linear system described above is solved using least squares or regularization methods to retrieve the three-dimensional displacement components $(v_x,v_y,v_z)$. Provided that the covariance of observational noise is known, error propagation is applied to quantify the uncertainty in the three-dimensional deformation results as follows:

$$C_v={\left(A^T{C}_{obs}^{-1}A\right)}^{-1}$$

(6)

In Equation (6), $A$ represents the coefficient matrix described above and $C_{obs}^{-1}$ denotes the observational error covariance matrix. The error associated with each displacement component can thus be extracted, providing a reliable basis for subsequent quantitative analyses of landslide risk.

4. Results

4.1. Site Deployment and Spatial Overview of Post-Disaster Monitoring

In this study, two ground-based radar systems were deployed to monitor the post-disaster evolution of a large-scale landslide. The deployment locations of the ground-based radar were highly constrained by the intense destructive force of the landslide, presenting considerable challenges for site selection. According to the theoretical criteria for optimal radar placement, the monitoring requirements are met when the deformation sensitivity of the slope radar exceeds 0.5. The deformation sensitivity of a slope radar refers to the degree to which the radar’s line-of-sight (LOS) is responsive to the displacement direction within the monitored area, specifically defined as the projection coefficient of the local sliding direction onto the LOS direction. The geometric relationship between the radar LOS and the plane of the monitored area is illustrated (Figure 4a). Since the value of the deformation sensitivity varies with position across the monitored area, its spatial distribution changes accordingly. The spatial distribution of deformation sensitivity is shown in Figure 4b.

Based on the geometric relationship illustrated above, the sensitivity is given by the value of cos ω, which can be calculated as follows:

$$cos\omega =sin\theta sin\phi cos\alpha sin\sigma +sin\theta cos\phi cos\sigma cos\alpha +cos\theta sin\alpha$$

(7)

NOEU refers to the north–east–up (NEU) coordinate system, where the positive directions are due east, due north, and vertically upward. TDKI denotes the local sliding surface coordinate system for the monitored area, with axes oriented along the sliding direction, the direction perpendicular to sliding, and the normal to the sliding surface, respectively [22]. $\omega$ represents the angle between the sliding direction and the radar line-of-sight (LOS); $\alpha$ is the slope angle of the monitored area; $\sigma$ is the slope aspect; $\theta$ is the radar incidence angle; and $\phi$ is the angle between the projection of the radar LOS in the NOE plane and the north direction.

Accordingly, the satisfaction rate of the slope radar deformation sensitivity index is defined as the ratio of the area within the monitored region that meets the sensitivity criterion to the total monitored area. In this study, the two ground-based radar systems achieved deformation sensitivities of 0.75 and 0.77, thus satisfying the monitoring requirements. The following sections will provide a detailed analysis of the spatiotemporal evolution characteristics of secondary disasters triggered by the landslide event.

Figure 5 presents the displacement monitoring results for the landslide area obtained using two ground-based Interferometric Synthetic Aperture Radar (GB-InSAR) systems between 15 February 2025 and 19 February 2025. Figure 6a and Figure 6b correspond to the monitoring perspectives and area coverage of the two devices, respectively. Both primary monitoring areas encompass the main landslide body and its surrounding regions. However, due to differences in device placement and imaging angles, slight variations are observed in the point cloud distribution. The monitoring results are expressed as line-of-sight (LOS) displacements, with color indicating the magnitude and direction of deformation: red to green denotes negative displacement (away from the radar) to positive displacement (toward the radar). The LOS displacement range for Figure 5a is from −54.22 mm to +45.71 mm, while that for Figure 5b is from −58.62 mm to +50.36 mm. The comparable ranges indicate a high degree of consistency and comparability between the monitoring results.

Both images reveal significant deformation responses, particularly in the central and lower parts of the landslide body, where persistent movement toward the radar is detected. This pattern is closely related to the topographic structure and potential sliding direction of the landslide, as the central and lower accumulation zones typically serve as focal areas for stress concentration and material transport. In addition, the widespread distribution of green areas indicates regions of relatively minor or stable deformation. The complementary perspectives provided by the two monitoring systems capture deformation characteristics from different directions, enhancing the comprehensiveness and accuracy of the monitoring. Therefore, integrating the results from both devices improves the spatial resolution and sensitivity of the landslide early warning system, supporting the dynamic identification and risk assessment of landslide evolution.

4.2. Time Series Deformation Characteristics and Stability Evolution of the Landslide

To further elucidate the evolutionary characteristics of secondary failures following a large-scale landslide, this study analyzes time series deformation data acquired from the landslide area between 16 February 2025 and 19 February 2025. These data include the spatiotemporal evolution of three physical parameters: acceleration (a), velocity (b), and displacement (c). The detailed analysis of these datasets enables a comprehensive understanding of the dynamic response characteristics of the landslide area during the monitoring period. As shown in Figure 6a, the time series of acceleration reveals significant fluctuations during specific intervals, particularly between 20:59 on 16 February 2025 and 10:33 on 17 February 2025. These fluctuations reflect abrupt changes in the internal stress state of the landslide mass. Figure 6b indicates that the velocity within the landslide area experienced a sudden increase between 16 February 2025 and 17 February 2025, which is consistent with the acceleration data. The displacement time series shown in Figure 6c provides additional information on the temporal evolution of deformation in the landslide area.

The analysis of the spatiotemporal distribution of deformation indicates that the most pronounced increases occurred between 16 February 2025 and 17 February 2025, with the maximum LOS displacement reaching approximately 15 mm, the peak LOS velocity up to 260 mm/h, and the peak LOS acceleration as high as 1036 mm/h².

Overall, the time series deformation process in the landslide area can be divided into two distinct stages: stage I (18:00 on 15 February 2025 to 17:00 on 16 February 2025) is characterized by slow deformation, which may represent the natural deformation of the landslide body under static or low-stress conditions; stage II (17:00 on 16 February 2025 to 10:00 on 17 February 2025) is marked by a clear acceleration in deformation rate and displacement magnitude, reflecting a deterioration in the dynamic stability of the landslide under external disturbances.

The comprehensive analysis of these time series data reveals the deformation characteristics and stability evolution of the landslide area at different stages. Moreover, thanks to the minute-level monitoring capability of ground-based radar, the evolution of the landslide can be continuously assessed, providing a reliable basis for timely landslide early warning.

4.3. Three-Dimensional Deformation Decomposition

After processing the two sets of ground-based SAR observation data, pixels with high data quality were geocoded, converting the radar image coordinates into a unified geographic coordinate system. This step enabled the consistent integration of the observational data with topographic, geomorphological, and remote sensing information. Subsequently, spatial registration and image resampling techniques were applied to identify and extract the set of common coherent pixels—pixels that exhibit high coherence and data integrity from both viewing geometries. This set of common pixels serves as the core data foundation for subsequent three-dimensional deformation inversion, and its geometric configuration is critical for the resolvability and stability of the derived deformation components. To ensure data accuracy, only pixels with strong complementary viewing angles and high coherence were retained for joint inversion.

Figure 7 illustrates the radar imaging geometry and parameter distributions for the two ground-based SAR observation sites within the landslide area. Figure 7a and b show the radar line-of-sight coverage for observation site 1 and site 2, respectively, with blue and red regions indicating the primary fields of view illuminated by the radar beams. The figure clearly demonstrates that the two sites provide highly complementary perspectives, effectively covering most key deformation zones of the landslide. Figure 7c,d present the radar incidence angle distributions for each observation point, reflecting the angle between the incident electromagnetic wave and the local vertical at the ground surface. The incidence angles range from 1.72° to 37.75° for site 1 and from 0.05° to 30.14° for site 2. Larger incidence angles are observed in the central and rear parts of the landslide body, which are beneficial for capturing vertical deformation signals. Figure 7e,f display the radar azimuth angles (the angle between the radar wave propagation direction and geographic north), which for both sites fall within the range of 53.03° to 89.99°, providing a solid geometric basis for three-dimensional deformation inversion and the identification of the principal sliding direction. The complementary viewing configurations of the two observation sites offer multi-directional and multi-angular radar geometric sensitivity within the landslide area, which is characterized by complex slope orientations and pronounced topographic relief, thereby significantly enhancing the reliability and accuracy of deformation field retrieval.

Based on the methodology described in Section 3, the three-dimensional surface displacement field of the landslide area was reconstructed, including the east–west, north–south, and vertical motion components. Overall, the main landslide zone exhibited significant deformation responses across all three directions. Figure 8 presents the spatial distribution of the 3D surface displacement components in the landslide area, as derived from dual-perspective GB-InSAR and high-resolution DEM inversion. Figure 8a, b, and c correspond to the east–west, north–south, and vertical displacement components, respectively, while Figure 8d provides an integrated 3D model of the main sliding zone superimposed on the landslide area. The black arrows indicate the direction of movement within the landslide region. The main sliding belt (outlined in red) demonstrates pronounced spatial heterogeneity in all three displacement components.

A detailed analysis is as follows: In the east–west component (Figure 8a), the maximum displacement is approximately +85 mm, primarily concentrated in the upper and central portions of the landslide body, indicating significant eastward sliding. Influenced by local slope and the primary sliding direction, negative east–west displacements (−80 mm and below) are mainly observed at the upper margin of the landslide. For the north–south component (Figure 8b), the maximum positive displacement, approximately +72 mm, occurs along the southern boundary and principal accumulation zone, while negative values (−60 mm) are concentrated near the northern toe, reflecting non-uniform movement along the north–south direction. In the vertical component (Figure 8c), the maximum subsidence within the main sliding belt reaches −100 mm, which is substantially greater than that of the surrounding original slope (generally not exceeding −30 mm). Areas of large vertical displacement are concentrated at the lower edge of the landslide deposit, reflecting the combined effects of material accumulation and surface uplift (frontal accumulation uplift/rear edge subsidence) during the landslide event.

The spatial distribution patterns reveal clear synergetic motion and differential deformation among the three components within the main sliding zone. Compared with the original slope, the mean three-dimensional displacement within the main sliding belt is approximately 60% higher, with contiguous high-deformation zones observed at the toe and accumulation areas. This indicates a large-scale shear sliding movement of the landslide mass along the principal sliding direction. Combined with DEM-based terrain analysis, the three-dimensional inversion results are highly consistent with the distribution of the main sliding belt, the geological structure of the landslide body, and surface morphology, thus confirming the validity and engineering applicability of the dual-perspective InSAR–DEM joint inversion for reconstructing 3D displacement fields.

5. Discussions

5.1. Influence of DEM on Three-Dimensional Deformation Retrieval

For small-scale landslides, conventional Shuttle Radar Topography Mission (SRTM) digital elevation model (DEM) data, typically at a 30 m resolution, exhibit limited effectiveness in capturing detailed deformation characteristics at such scales. Moreover, while SRTM DEM is generally used only when fine DEM data are unavailable and is unsuitable for dynamic simulation due to its coarse resolution, this study employed a high-resolution DEM to ensure accurate terrain constraints for the inversion process. To comprehensively reveal the true motion characteristics of landslides, this study implements a joint inversion of the three-dimensional (3D) surface deformation field $v={[v_x,v_y,v_z]}^T$ using dual-view ground-based Interferometric Synthetic Aperture Radar (GB-InSAR) observations and high-resolution DEM data. Specifically, the 3D displacement vector of any pixel within the landslide area is projected onto the respective line-of-sight (LOS) directions of two GB-InSAR systems, establishing the following observation equations:

$$d_i=n_i\cdot v+{ϵ}_i(i=\mathrm{1,2})$$

(8)

In Equation (8), $d_i$ denotes the LOS deformation acquired by the i-th GB-InSAR device, $n_i$ represents the LOS directional cosine vector, and ${ϵ}_i$ is the observational error term. Since the two LOS observations alone are insufficient to uniquely resolve the three displacement components, additional physical constraints are introduced. This study supplements the observation equations with a third independent constraint derived either from the slope-normal vector obtained from the DEM or a known principal sliding direction, thereby forming a solvable linear system. All observational data and DEM parameters are resampled to a consistent spatial resolution and coordinate system and rigorously co-registered using ground control points. The resulting combined equation system is solved pixel-wise via the least squares method, yielding the three-dimensional displacement components $[v_x,v_y,v_z]$.

As shown in Figure 9, SRTM DEM data provide only a limited number of pixels, restricting their capability to accurately represent the detailed topography and geomorphology of landslides. Particularly for high-resolution GB-SAR observations, the use of such DEM data could introduce terrain-induced errors into the three-dimensional deformation retrieval models. Moreover, following a landslide event, substantial elevation changes occur within the collapsed area, which older SRTM DEM data fail to capture due to their earlier acquisition date. When the SRTM DEM data were resampled to match the resolution of high-precision UAV-derived DEM data, significant inconsistencies emerged due to differences in temporal and spatial resolution. Elevation differences calculated on an identical grid reached up to −61.14 m (Figure 9c). This discrepancy further highlights the limitations of moderate-resolution data for monitoring small-scale landslides.

As illustrated in Figure 9d, terrain information extracted from high-resolution DEM data provides a more accurate depiction of landslide topography compared with SRTM DEM data. Such high-precision terrain data significantly reduce topographic errors associated with prior constraints in the three-dimensional deformation reconstruction process. Detailed analysis using high-resolution DEM data thus facilitates a more accurate understanding and prediction of landslide deformation processes.

5.2. Field Investigation and Validation

Due to the significant fracturing of the slope surface following a large-scale landslide, as well as the presence of a steep, rocky slope, a field investigation was conducted on 19 February 2025, to verify the reliability of the deformation monitoring results. Field photographs demonstrate that the rock layers in this region are highly jointed and the slope surface is extensively fractured, indicating that the landslide initiation zone is characterized by high-level tensile cracking and loosened rock masses. These features are consistent with the extensional deformation zone identified at the upper edge in the three-dimensional inversion results. Areas exhibiting clear evidence of secondary collapse at the disaster site were selected for illustration, as shown in Figure 10a, Figure 10b and Figure 10c, which display the displacement distributions of the landslide body in the east–west, north–south, and vertical directions, respectively, while Figure 10d shows the results of field verification. In the white area, newly collapsed zones of the landslide body are clearly visible and are distinct from adjacent undisturbed rock and soil, indicating that secondary collapses occurred following the initial landslide event. These features correspond well with the deformation zones interpreted from the three-dimensional decomposition results and exhibit a high degree of consistency. Additionally, the field photographs reveal that the leading edge and periphery of the landslide, as well as the surrounding forest, were affected, with the ground surface covered by substantial debris and severe damage to crops.

Concurrently, the field verification of the entire landslide area is documented: Figure 11a shows the upper landslide segment, revealing the initial detachment zone and steep slope area; Figure 11b depicts the middle segment, presenting the main sliding body with accumulated debris; Figure 11c illustrates the lower landslide mass, displaying the distal accumulation zone; and Figure 11d shows the primary building damage zone, indicating the most severely impacted location at the alluvial terminus. These images provide comprehensive qualitative validation of the radar monitoring results, covering the entire landslide extent from the upper steep slope to the lower deposition zone, including critical collapse areas. Detailed field observations confirm the effectiveness of the high-frequency GB-InSAR method in capturing the spatial distribution and evolutionary process of large-scale landslide deformation.

6. Conclusions

This study focuses on a large-scale landslide event that occurred in Jinping Village, which is representative of rainfall-induced rapid landslides in the mountainous areas of southwest China. To address the need for high-precision, dynamic post-disaster monitoring and deformation mechanism identification in large landslides, two ground-based synthetic aperture radar (GB-InSAR) systems were strategically deployed based on theoretical site selection criteria and sensitivity analysis. These systems achieved deformation sensitivities of 0.75 and 0.77, effectively overcoming the challenges posed by complex topography and extensive deformation on radar deployment and monitoring capability. Continuous observation data demonstrate that the two radar systems provide highly complementary coverage of the main landslide body and surrounding critical deformation zones, with a high degree of consistency in the observed line-of-sight (LOS) displacement fields.

High-frequency, high spatiotemporal resolution time series data revealed that the landslide underwent a dynamic evolution from gradual accumulation to rapid acceleration between 15 and 19 February 2025. Notably, persistent movement toward the radar was observed in the central and lower sections of the sliding mass, indicating a close relationship between secondary instability and stress concentration. Furthermore, three-dimensional deformation inversion was performed using high-resolution DEMs as terrain constraints, with high-coherence pixels selected from both radar perspectives and precise co-registration employed. This enabled the systematic reconstruction of spatial displacement distributions in the east–west, north–south, and vertical directions. The results show significant displacements along all directions within the main sliding zone, with the mean 3D deformation exceeding that of the original slope surface by approximately 60% and the maximum vertical subsidence reaching −100 mm. The deformation exhibits pronounced spatial heterogeneity and closely corresponds to geomorphic structure and sliding direction.

A comparative analysis further demonstrates that the conventional SRTM DEM (30 m) is insufficient for capturing the complex, small-scale, and highly variable landslide morphology, while the high-resolution DEM (2 m) substantially improves the accuracy of 3D inversion and reduces topographic constraint errors, thereby providing a solid foundation for disaster evolution analysis and landslide mechanism interpretation. Post-disaster field investigations further validated the reliability of GB-InSAR monitoring results, with observed cracks and tensile features closely matching areas of high deformation response and the spatial distribution of surface debris, thereby confirming the accuracy of the monitoring approach.

This study constructed a “dual-view GB-InSAR + high-resolution DEM” three-dimensional dynamic deformation monitoring method for landslides. The monitoring system enables the precise characterization of the spatiotemporal evolution of large-scale landslides after major disasters. While the proposed method provides valuable information on deformation patterns, a complete specification of the landslide mechanism requires additional geological surveys, geotechnical testing, and numerical modeling, which will be considered in future work. This approach not only provides a theoretical and technical basis for the early detection and risk warning of landslide hazards but also serves as a valuable model for multi-scale monitoring and mechanism studies of geological hazards in complex mountainous regions.