Long-Term Monitoring of Landslide Activity in a Debris Flow Gully Using SBAS-InSAR: A Case Study of Shawan Gully, China

Abstract

Shawan Gully historically experienced recurrent debris flow events, resulting in significant losses of life and property. The Nuole and Huajiaoshu landslides are two major high-elevation landslides in Shawan Gully, serving as primary sources of debris flow material. To monitor landslides movements, this study used interferometric synthetic aperture radar (InSAR) and Sentinel-1 SAR imagery acquired between 2014 and 2023 to analyze surface deformation in Shawan Gully. Prior to InSAR processing, we assessed the InSAR measurement suitability of the involved SAR images in detail based on geometric distortion and monitoring sensitivity. Compared to conventional SBAS-InSAR results without preprocessing, the suitability-refined datasets show improvements in interferometric phase quality (1.55 rad to 1.41 rad) and estimation accuracy (1.45 mm to 1.18 mm). By processing ascending, descending, and cross-track Sentinel-1 SAR images, we obtained multi-directional surface displacements in Shawan Gully. The results reveal significant deformation in the NL1 region of Nuole landslide, while the northern scarp and the foot of the slope exhibited different movement characteristics, indicating spatially variable deformation mechanisms. The study also revealed that the Nuole landslide exhibits a high sensitivity to rainfall-induced instability, with rainfall significantly changing its original movement trend.

1. Introduction

Landslides and debris flows are two major natural hazards globally [1,2,3]. In canyon areas with complex topography, landslides are usually the main source of debris flows [4,5]. Significant topographic enhances triggered by landslide events can directly or indirectly amplify subsequent debris flows by changing the availability and distribution of source materials [6,7,8]. In the 2010 debris flow disaster in Zhouqu, large-scale landslides and slope failures transported substantial volumes of loose material into the gullies, significantly enhancing the scale and destructive power of the debris flow [9,10]. Similarly, in August 2014, prolonged heavy rainfall in the Hiroshima region of Japan triggered slopes’ failures, which subsequently led to large-scale debris flows, resulting in massive casualties and property damage [11,12]. Landslide failures are typically preceded by a long-term creeping stage [13,14]. During this time, continuous monitoring of surface deformation on landslides can effectively capture subtle changes, providing early warning and defensive guidance for landslides’ instability and the concomitant debris flows [15,16]. However, as most such landslides usually occur in mountainous regions that are difficult to access, continuous surface deformation monitoring remains a challenge [17,18].

As a well-validated microwave remote sensing methodology, InSAR provides a long-term observation of surface dynamics [19,20]. Compared to other techniques, thanks to its robustness in various weather conditions, continuous temporal coverage, high measurement accuracy, and large-scale observation capacity, modern InSAR methodologies revolutionized our capacity to characterize surface deformation processes [21,22,23,24,25]. In mountain areas where field surveying and traditional remote sensing methods are in limited supply, time series InSAR (TS-InSAR), represented by small baseline subset InSAR (SBAS-InSAR), has become one of the preferred approaches for studying landslides or other geological disasters [26,27,28]. Over the past two decades, continuous improvements in SAR data processing strategies and InSAR analytical methods enabled us to more accurately simulate the landslide displacement trends [29,30,31]. Furthermore, with the growing accessibility of diverse SAR datasets and acquisition modes, the operational efficiency of InSAR significantly improved, allowing for landslide monitoring with higher spatio-temporal continuity [32,33,34].

Receiving echo signals from target areas by SAR sensors is a fundamental prerequisite for successful InSAR applications. In mountainous canyon terrains, the spatial relationship between ground surface and satellite constantly changes, introducing uncertainties in deformation signal transmission and causing geometric distortions in SAR imagery [35]. Shadow and layover phenomena may result in phase loss or overlap, disrupting the continuity of phase unwrapping and reducing monitoring accuracy and reliability [36,37,38]. Changes in slope and aspect may influence the monitoring sensitivity of InSAR, further compromising the reliability of monitoring results [39,40,41]. In addition, InSAR measurements can only reflect displacements in the line-of-sight (LOS) direction [42,43]. In complex terrains, this limitation poses a substantial challenge to accurately analyzing landslide deformation mechanisms and kinematic behavior [44,45]. Therefore, in order to achieve high-reliability monitoring of landslide movement, it is necessary to pre-evaluate the suitability of SAR images for InSAR deformation monitoring in target areas.

Shawan Gully is situated within the Xiaojiang River Basin and is strongly affected by the Xiaojiang fault. In the past few decades, frequent debris flows occurred in this area, severely damaging downstream farmlands and structures. The Nuole and Huajiaoshu landslides, situated in the upper reaches of Shawan Gully, are two large high-elevation landslides that serve as the primary material sources for debris flow events. Under continuous vertical incision and lateral erosion, both landslides experience secondary sliding each year, with volumes ranging from dozens of cubic meters to tens of thousands of cubic meters.

To assess slope activity within the Shawan Gully, this study used SBAS-InSAR and 688 Sentinel-1A acquisitions from three orbits to measure surface displacements in the region. Based on the geometric distortion and monitoring sensitivity, the InSAR monitoring suitability of the involved SAR imagery for landslide analysis in the Shawan Gully watershed was assessed, which identified the applicable and non-applicable datasets for this study. In addition, several permanent scatterer ground control points (PS-GCPs) were extracted by comprehensively considering the phase, scattering, coherence, and deformation trend [46]. During the SBAS-InSAR processing, these PS-GCPs were utilized to adjust baseline errors and eliminate the flattening effect, thereby improving interferogram quality and deformation estimation accuracy. Given the complexity of terrain and the high activity of rainfall-induced landslides, this study provides a representative case for exploring the effectiveness and reliability of InSAR-based monitoring in debris flow-prone mountain gullies.

2. Study Area and Datasets

2.1. Study Area

Shawan Gully is situated within the Xiaojiang River Basin (Figure 1). This major channel spans ~10.32 km in length, with a watershed area of ~17.97 km² [47]. Since 1990, frequent debris flows have been recorded in Shawan Gully due to the continuous material supplying from the Nuole and Huajiaoshu landslides. These disasters resulted in more than 50 casualties and forced the relocation of nearly 400 households [48].

The Nuole and Huajiaoshu landslides serve as principal contributors of material for debris flows within Shawan Gully, accounting for 52.97% of the total volume. These two landslides, with an elevation difference of over 700 m, are typical high-elevation landslides. The Nuole landslide has a maximum thickness of ~130 m, with an estimated volume of solid material at ~1.47 × 10⁸ m³ [49]. A fault of interbedded sandstone and shale intersects the rear edge. Influenced by the cumulative effects of gravitational forces, hydrostatic pressure within the bedrock fractures, and undercutting at the toe, this dip slope landslide is prone to movement along the bedding planes. Continuous headward erosion carved a large, deep channel, effectively dividing the Nuole landslide into two distinct sections, NL1 and NL2. Owing to past geological hazards, a tensile crack (Tc1) developed in the NL1 region of the Nuole landslide, stretching ~1.26 km in length with a maximum width greater than 50 m. Nuole Village is situated at a flat elevation of ~2370 m. Two minor cracks (Mc1 and Mc2) and two landslide steps (Ls1 and Ls2) developed in the surroundings. The length of the landslide scarp (blue line in Figure 1c) in the northern NL1 region was ~1.61 km.

The Huajiaoshu landslide, situated on the northern side of the main gully, has a solid reserve of approximately 3.20 × 10⁷ m³, with its main slip oriented at 230°. The lithology mainly consists of shale and sandstone, featuring rock fragmentation and the development of fissures. A tensile crack (Tc2) approximately 0.66 km long divides the Huajiaoshu landslide into northern and southern sections. Additionally, two minor cracks (Mc3 and Mc4) and two landslide steps (Ls3 and Ls4) developed.

2.2. Datasets

In the study, three groups of Sentinel-1 SAR imagery acquired from 2014 to 2023 were collected. The basic parameters that are listed in

Table 1. Basic parameters of the involved Sentinel-1 SAR datasets.

Path	26	128	62
Flight mode	ascending	ascending	descending
Polarization	VV	VV	VV
Heading angle (°)	347.5028	347.4863	192.5084
Incidence angle (°)	Local incidence angle (LIA)
Acquisition period	19 October 2014 to 29 May 2023	26 November 2015 to 5 June 2023	9 October 2014 to 6 July 2023
Number	223	210	255

: 255 scenes are from the descending orbit path 62 (S1DP62), and the remaining 433 scenes are from two different ascending orbits, path 26 (S1AP26) and path 128 (S1AP128). Observations from three distinct viewpoints facilitated mitigated negative effects by terrains [30]. When importing Sentinel-1 data, orbit errors were corrected with precise orbit data. During InSAR processing, temporal and perpendicular baseline thresholds of 72 d and 300 m, respectively, were applied to attenuate the effects caused by spatial–temporal incoherence. The spatial–temporal baseline networks for the three small baseline datasets are shown in Figure 2.

Moreover, a high-resolution (5 m) DEM was used to precisely assess the InSAR suitability within the Shawan Gully, which was also applied to generate slope and aspect maps. Atmospheric delay corrections were applied using GACOS ZTD data, accessible via http://www.gacos.net/ (accessed on 24 May 2024). Precipitation data from 2014 to 2021 were obtained from daily records at Dongchuan Station, managed by the Chinese Academy of Sciences. For the years 2022 to 2023, we utilized average values from comparable periods in previous years. The optical imagery and base topographic maps were obtained from Google Earth and ArcGIS Pro, respectively, and were used for background mapping and auxiliary interpretation.

3. Materials and Methods

The topography of Shawan Gully is extremely rugged, with large elevation gradients across the region. To ensure the effective acquisition of surface deformation signals from the main targets (Nuole and Huajiaoshu landslides), we assessed the InSAR suitability of Sentinel-1 SAR datasets in the region according to an analysis of geometric distortion and monitoring sensitivity, taking into account the imaging properties of the Sentinel-1 satellite. Then, we selected SAR images from the optimal orbits for SBAS-InSAR time series analysis. Figure 3 provides an overview of the technical scheme.

3.1. Geometric Distortion Identification

The limitations of using InSAR to monitor surface deformation in alpine canyon regions primarily arise from uncertainties related to topographical factors [50,51]. Side-looking radar imaging in highly undulating terrain inevitably leads to geometric distortions (Figure 4) [36,52]. When the slope is oriented towards the sensor, the radar signal compresses in the range direction due to uneven terrain, causing foreshortening or layover (segment b–a in Figure 4). Conversely, if the incidence angle exceeds the complementary slope angle, the satellite cannot capture the surface on the shadowed slope, resulting in an active shadow (segment d–e in Figure 4). Layovers and shadows significantly impact the interpretation and interferometry of SAR images. In the layover regions, the radar signal compression leads to abnormal changes in backscattering intensity and phase values [53]. No radar signal transmission occurs in shadowed areas.

Accurate classification results of geometric distortion are helpful for terrain visibility analysis. Two principal methods for identifying geometric distortions are the Layover and Shadow Map method (LSM) and the R-index model [54,55]. The LSM method is highly accurate in identifying shadows (active or passive) and layovers (active or passive), and the R-index model can effectively separate active layover, foreshortening, and visible areas [39,56]. This study combined the LSM and R-index models to accurately classify geometric distortions by simulating surface visibility. Initially, the LSM method was used to extract shadow and layover (including far- and near-) regions in the SAR imagery. Following the initial classification by the LSM, the R-index model was applied to accurately classify good visibility (GV) areas, foreshortening (FS), active layover (AL), passive layover (PL), active shadow (ASh), and passive shadow (PSh). If the target area predominantly falls within active/passive layover zones or active/passive shadow zones, it indicates that the dataset from this orbit has low applicability in the study area.

3.2. InSAR Monitoring Sensitivity Analysis

InSAR monitoring sensitivity refers to the ability of SAR sensors to detect surface deformations in different terrains [50]. Influenced by topography, vegetation, and atmospheric conditions, the Sentinel-1 satellite exhibits varied deformation detection capabilities across different slope units in both ascending and descending orbits [40]. This capability can be quantified based on sensitivity levels.

Figure 5 shows the spatial relationship between LOS displacement (V_LOS) and actual slope displacement. The down-slope direction along the slope surface is considered the actual direction of slope displacement (V_Slope), and InSAR deformation monitoring sensitivity can be expressed as [57]:

$$C-index=V_{LOS}/V_{Slope}=\overrightarrow{u}\cdot \overrightarrow{\gamma }.$$

(1)

The theoretical values of the C-index range from −1 to 1. A |C-index| value approaching 1 corresponds to an increased monitoring sensitivity and a greater reliability of the InSAR measurement. Conversely, an |C-index| value closer to 0 indicates means that the monitoring sensitivity is lower and the InSAR measurement is less reliable. A negative C-index value indicates that the surface area theoretically should be moving away from the satellite. The $\overrightarrow{u}$ and $\overrightarrow{\gamma }$ are the deformation unit vectors in the LOS and actual directions, while $\overrightarrow{u}=\overrightarrow{\gamma }=1$. Decomposing the vectors $\overrightarrow{u}$ and $\overrightarrow{\gamma }$ into the north-east-up (N-E-U) coordinate system:

$$\overrightarrow{u}=\left[\begin{array}{c}{\overrightarrow{u}}_N\\ {\overrightarrow{u}}_E\\ {\overrightarrow{u}}_U\end{array}\right]=\left[\begin{array}{c}\mathit{sin}\phi \mathit{sin}\theta \\ -\mathit{cos}\phi \mathit{sin}\theta \\ \mathit{cos}\theta \end{array}\right]$$

(2)

$$\overrightarrow{\gamma }=\left[\begin{array}{c}{\overrightarrow{\gamma }}_N\\ {\overrightarrow{\gamma }}_E\\ {\overrightarrow{\gamma }}_U\end{array}\right]=\left[\begin{array}{c}-\mathit{cos}\alpha \mathit{cos}\beta \\ -\mathit{sin}\alpha \mathit{cos}\beta \\ \mathit{sin}\beta \end{array}\right]$$

(3)

where θ represents the radar incidence angle, and φ represents the angle between the satellite flight direction and the north direction. In Equation (3), α is the slope aspect and β is the slope gradient.

3.3. Surface Deformation Measurements

3.3.1. PS-GCPs Selection

We processed all the SAR datasets using the SBAS-InSAR method. The methodological details are not repeated herein, and the specific steps can be found in previous studies [58,59]. The conventional SBAS method may suffer from omissions and misselections due to the sidelobe effect, which can lead to low coherence, phase instability, and potential residual phase jumps during phase unwrapping [60]. To mitigate this issue, we proposed an approach that utilizes stable GCPs to refine the satellite orbit and re-flatten the images [46]. Collecting stable GCPs with high spatial density from the ground using conventional methods is challenging, particularly for study areas that cover broad regions. In this study, we selected pixels with persistent scattering characteristics to serve as PS-GCPs. For Gaussian scatterer pixels, PS-GCPs ensure high coherence, signal-to-noise ratio, and phase stability even in cases of low overall surface coherence.

In this processing, interferograms were first generated using a common master SLC image. Typically, selecting an image acquired near the temporal center of the dataset as the super master image is more advantageous to evaluate the long-term coherence and phase stability of pixels. The super master images used in the study were acquired on 13 July 2019 (ascending orbit 26), 6 September 2019 (ascending orbit 128), and 15 July 2019 (descending orbit 62). Then, temporal coherence (TC) and amplitude dispersion index (ADI) were calculated, and the appropriate thresholds T_TC and T_ADI were set to filter potential PS-GCP candidates. ADI was calculated according to the Equation (5):

$$ADI={\sigma }_A/\overline{\alpha }$$

(4)

where ${\sigma }_A$ represents the standard deviation of the mean amplitude, and $\overline{\alpha }$ represents the mean time series amplitude. A lower ADI value indicates higher phase stability of the pixel [61]. Based on the thresholds T_TC and T_ADI, PS-GCP candidate points were preliminarily detected. Furthermore, the candidate points were refined through the linear deformation residuals to obtain PS-GCPs with high coherence, high signal-to-noise ratio, and stable deformation. PS-GCPs not only enhance the accuracy of data processing as verification and correction tools, but also maintain the consistency and reliability of the analysis in complex environments, such as in urban or densely vegetated areas.

3.3.2. Displacement Components Estimation in Vertical and Horizontal Directions

The surface deformation results calculated by SBAS-InSAR have limitations when analyzing landslide movement patterns. Projecting the LOS deformation into the three-dimensional directions (N-E-U) is a common solution that helps in the refined analysis of landslide activity [62,63]. The projection relationship is shown below [64]:

$${\overrightarrow{V}}_{LOS}=A{\overrightarrow{V}}_{N-E-U}$$

(5)

where ${\overrightarrow{V}}_{LOS}$ represents the LOS deformation vector, and ${\overrightarrow{V}}_{N-E-U}$ represents the three-dimensional deformation vector. A is an m × n coefficient matrix, where m and n, respectively, represent the number of ${\overrightarrow{V}}_{LOS}$ and the three-dimensional deformation components. For any LOS InSAR monitoring result:

$$V_{los}=n_{N-S}{V}_{N-S}+n_{E-W}{V}_{E-W}+n_{U-D}{V}_{U-D}$$

(6)

where V_los, V_N-S, V_E-W, and V_U-D represent the displacements in the LOS, north–south (N-S), east–west (E-W), and up–down (U-D), respectively. The unit vectors for these deformation components are denoted as $n_{N-S}$, $n_{E-W}$, and $n_{U-D}$. When m ≥ n, the optimal or unique solution for the three-dimensional deformation components can be estimated [65]. However, this study only used two datasets from Sentinel-1 ascending and descending orbits. Equation (6) is rewritten as a two-dimensional projection model [66]:

$${\overrightarrow{V}}_{LOS}=A{\overrightarrow{V}}_{E-U}$$

(7)

$$\left[\begin{array}{c}{V}_{los}^{Asc}\\ V_{los}^{Desc}\end{array}\right]=\left[\begin{array}{cc}-\cos {\phi }_{Asc}·\sin {\theta }_{Asc}& \cos {\theta }_{Asc}\\ -\cos {\phi }_{Desc}·\sin {\theta }_{Desc}& \cos {\theta }_{Desc}\end{array}\right]\left[\begin{array}{c}{V}_{E-W}\\ V_{U-D}\end{array}\right]$$

(8)

where θ is the radar incidence angle, and φ is the satellite flight direction.

4. Results

4.1. Suitability Assessment of InSAR Measurements

4.1.1. SAR Imagery Availability Analysis

Figure 6 presents the simulation results of the surface geometric distortions and terrain visibility in the Shawan Gully. A 5 m resolution DEM was used in the simulation step to visualize the terrain details. The local incidence angles of the study area surface relative to the satellite sensor on the three orbits were calculated to simulate a more realistic terrain visibility (Figure 6a–c). In comparison, the surface is less susceptible to geometric distortions on the SAR images from S1DP62, with both the Nuole and Huajiaoshu landslide within the good visibility terrains. Shadow is prone to generate on the collapses BT1 and BT2, while the BT3 region is susceptible to layover. The S1AP26 has good visibility of the Nuole landslide, but is affected by layover at the base of the Huajiaoshu landslide. The Shawan Gully is particularly vulnerable to layover and shadow on the path S1AP128 because of the small incidence angle, making it difficult to effectively measure the whole Nuole and Huajiaoshu landslide.

In the SAR imagery from S1DP62, the proportion of the surface with good visibility is 78.23%, significantly higher than other types of geometric distortions. However, it only accounts for 26.61% and 25.73% in the SAR imagery of ascending orbits S1AP26 and S1AP128, respectively (Figure 6g). Additionally, in S1AP128, the area of layover accounts for 26.85%, surpassing the area with good visibility. Thus, the spatial distribution and quantitative statistics of geometric distortions indicate that the SAR imagery from S1AP128 is severely affected by layover and shadow, making it unsuitable for refined monitoring of the Nuole and Huajiaoshu landslides.

4.1.2. Reliability Analysis of InSAR Measurements

Figure 7 delineates the theoretical model of InSAR monitoring sensitivity. Assuming that only the slope and aspect are considered, higher monitoring sensitivities are observed for terrains ranging from 0° to 168° and from 348° to 360° in the Sentinel-1 ascending S1AP26 and S1AP128. On these surfaces, the monitoring sensitivity gradually increases as the slope increases. In the descending orbit, the Sentinel-1 satellite has a higher monitoring sensitivity for slopes with a gradient between 12° and 198°. With an identical slope gradient, monitoring sensitivity increases as the slope aspect aligns more closely with the LOS direction, reaching its minimum in the satellite azimuth direction. If the slope aspect is constant, the monitoring sensitivity increases with the rise of the slope gradient. The impact of geometric distortions on monitoring sensitivity is illustrated in Figure 7d–f.

As shown in

Table 2. InSAR monitoring sensitivity classification.

Sensitivity	[min, −0.5) and (0.5, 1]	[−0.5, −0.3) and (0.3, 0.5]	[−0.3, −0.1) and (0.1, 0.3]	[−0.1, 0) and (0, 0.1]	0
Classification	High	Middle	Low	Extremely low	insensitivity

, the InSAR monitoring sensitivities of the involved Sentinel-1 SAR datasets were classified into different levels, and the results are presented in Figure 8. The S1DP62 orbit demonstrates a higher monitoring sensitivity to the surface within the Shawan Gully. On this orbit, the area of InSAR monitoring with high sensitivity reaches 64.32%, encompassing most of the slope surfaces of the Nuole and Huajiaoshu landslides. In contrast, the ascending orbits S1AP26 and S1AP128 primarily exhibit low sensitivity, with high-sensitivity areas of less than 25%. Moreover, the areas insensitive to InSAR monitoring across the three orbits are minimal, with even the highest proportion on the S1AP128 orbit being only 0.31%. Therefore, it can be considered that the SAR images from all three orbits are credible in monitoring the study area, with S1DP62 providing the highest reliability. In the study, regions affected by layover and shadow were removed prior to interferometric processing to prevent the propagation of unwrapping errors caused by geometric distortions.

4.2. Deformation Monitoring and Kinematic Patterns Assessment

4.2.1. PS-GCP Detection and Identification

To ensure an even distribution of PS-GCPs, the study area was divided into several 5 km × 5 km sub-regions. The T_TC and T_ADI were set to 0.9 and 0.35, respectively. Subsequently, we utilized terrain phase residuals to thin the PS-GCPs, with the results shown in Figure 9. On orbits S1AP26, S1AP128, and S1DP62, 26, 31, and 40 PS-GCPs were identified, respectively (Figure 9a–c). Only two PS-GCPs had identical spatial locations across the three orbits, whereas ten PS-GCPs were derived from the same locations from the two ascending orbits. These PS-GCPs were primarily selected from the Jinyuan Town and mountainous areas distant from the study area. After corrections with high-precision DEM, the topographic phase residuals of these PS-GCPs were essentially eliminated (Figure 9d–f).

The coherence coefficient was determined by the phase consistency of pixels within a time series. Higher coherence indicates greater consistency of radar observations at different times, with less phase variation. The amplitude dispersion index measures the variability of pixel amplitude within a time series. A lower amplitude dispersion index indicates more stable radar reflectance amplitudes over time. An analysis of the relationship between coherence coefficients and amplitude dispersion indices of the PS-GCPs revealed a negative correlation (Figure 9g–i). This implies that pixels with higher coherence exhibit lower variability in radar amplitude, potentially due to more stable surface characteristics or location in areas with less environmental change. In SBAS-InSAR processing, these PS-GCPs with known positions can be used as reference points to calibrate and adjust the geometric accuracy of interferograms, thereby enhancing the accuracy and reliability of InSAR measurements. Additionally, since PS-GCPs provide stable radar reflection signals, they can overcome the effects of complex environmental changes, such as vegetation cover.

4.2.2. Multi-Orbit SBAS-InSAR Surface Deformation Measurements

Before to the SBAS-InSAR processing, it was necessary to preprocess the SAR images based on terrain visibility and monitoring sensitivity. We masked the layover, shadow, and insensitive areas in the SAR images prior to the interfering step to mitigate the negative impacts of those phenomena on the InSAR measurement. Using the SARscape software, we thinned the SAR images from three orbits and utilized SBAS-InSAR to estimate the surface deformation. Figure 10 presents the LOS deformation velocities for the three orbits. A positive deformation value indicates displacement towards the satellite, while a negative value indicates displacement in the opposite direction. The magnitude of these values is associated with slope activity, where larger values suggest poorer slope stability and more intense activity.

Figure 10 shows that the mid and lower reaches of the Shawan Gully are stable, while significant deformations are primarily detected in the NL1 region of the Nuole landslide. The surface deformation velocity in Jinyuan Town, located downstream, ranges from −20 to 0 mm/year. Due to the impact of layover, effective monitoring of the full surface deformation in Shawan Gully is limited when using Sentinel-1 imagery acquired from S1AP128 (Figure 10a,b). Nevertheless, this orbit can still capture significant deformation at the rear, bottom slopes, and northern scarp of the NL1 region. The fastest deformation velocity at the bottom slope is 94.64 mm/year, and −89.14 mm/year at the northern scarp. The SAR images from the orbits S1AP26 and S1DP62 detected more comprehensive surface deformations of Shawan Gully (Figure 10c,d). For the unstable regions of the Nuole landslide, the LOS deformations calculated from S1AP26 align well with those from the S1AP128, while the S1DP62 reveals a broader extent of significant displacements. Around Nuole Village, the surface deformation velocity in the both S1AP26 and S1AP128 ranges from −20 to 20 mm/year. In the S1DP62, the Nuole landslide exhibits displacement away from the sensor, with the fastest slope deformation velocity of −139.82 mm/year occurring below Nuole Village. The Huajiaoshu landslide appears generally stable in the monitoring results from both the orbits S1AP26 and S1DP62, with only minor significant deformations detected at the bottom, the maximum deformation velocities being 39.42 mm/year and −37.95 mm/year, respectively. Additionally, the S1DP62 dataset also reveals significant displacements of area BT2, with a maximum velocity of −43.32 mm/year.

4.3. Regional Analysis of Landslide Dynamics

In the Shawan Gully, severe geometric distortions and lower monitoring sensitivity result in poor measurement quality for S1AP128, leading to a reduced number of effective deformation pixels. This suggests that this orbit is unsuitable for analyzing the displacement of the Nuole and Huajiaoshu landslides. Therefore, we combined the monitoring results from S1AP26 and S1DP62 to estimate the vertical and horizontal displacement components with Equation (8) (Figure 11).

The majority of the ground surface exhibits deformation velocities in the E-W and U-D directions that do not exceed ±10 mm/yr. However, the Nuole landslide displays significant activity signals. In the E-W direction, the NL2 region remains steady, whereas the lower and middle parts of the NL1 region exhibit strong activity (Figure 11a). The NL1 region is dominated by westward displacement. As shown in Figure 11c, the Ls2 region on the Nuole landslide exhibits the fastest westward deformation rate, reaching −170.43 mm/yr.

Furthermore, in the U-D direction, the Nuole landslide is primarily characterized by subsidence (Figure 11b). At the top of the NL2 region, the maximum subsidence velocity is −33.91 mm/yr, which gradually decreases along the downslope direction. The deformation pattern in the NL1 region demonstrates a steep gradient from the rear edge to the bottom of the landslide, characterized by a sequence of subsidence, stability, subsidence, uplift, and subsidence. This deformation characteristic is more evident in the analysis results of profile AA′ (Figure 11c). The fastest subsidence velocity of the rear edge is −96.75 mm/yr, decreasing to −4.31 mm/yr as it reaches Nuole Village. Moreover, along the downslope direction, after passing through the minor cracks Mc1 and Mc2, the subsidence velocity increases again to −69.72 mm/yr in the Ls2 region. The flat terrain below Ls2 allows the accumulation of loose materials, leading to uplift with a maximum velocity of 27.84 mm/yr. The fastest subsidence velocity is −104.42 mm/yr, as observed on the landslide scarp. In contrast, the Huajiaoshu landslide is relatively stable, with only minor displacements identified at the foot (Figure 11b,d).

Figure 11 illustrates that the local movement intensity within the NL1 area of the Nuole landslide varies, likely due to differences in terrain, landform, geological structure, and vegetation environment across different areas of the landslide body. Additionally, variations in slope gradient and aspect in local areas result in differing primary displacement directions (Figure 12). Specifically, the landslide body (area W-1) predominantly exhibits westward deformation in the E-W direction (Figure 12a,b). The northern slope of the landslide body, oriented approximately 40°, shows eastward movement (area E-2), which is opposite to the main body of the NL1 area. The landslide surface exhibits more complex local displacement characteristics in the vertical direction (Figure 12c,d). Nuole Village shows significant westward movement horizontally, but only slight subsidence vertically. The D-1 area, with slight westward movement horizontally, exhibits pronounced subsidence vertically, indicating that subsidence is the primary movement trend in this area. The D-2, D-3, and D-4 areas show significant displacement in both horizontal and vertical directions. The U-1 area at the toe is primarily characterized by uplift deformation, likely due to the accumulation of unstable materials from other regions, such as areas D-1 and D-2. The U-2, located below the Ls2 slope with relatively flat terrain, shows uplift as the main vertical movement trend.

4.4. Rainfall-Induced Changes in Landslide Activity

As shown in Figure 13, landslide failures are typically preceded by a long-term creeping process that involves three stages: deceleration (primary creep), steady state (secondary creep), and acceleration (tertiary creep) [67,68]. Landslide displacement reflects the dynamic balance between destabilizing and stabilizing forces [69]. Linear and sublinear creep are two common manifestations of a landslide during the stable creep stage. Sublinear creep may indicate progressive damage accumulation before catastrophic failure. Landslides in the accelerated creep stage are subjected to increasing shear forces, manifesting as persistently growing displacements. Under external factors such as rainfall, seismic activity, or anthropogenic disturbances, the slope may become unstable and develop into a landslide or even more severe hazards [70,71].

Multi-orbital InSAR measurements identified the Nuole landslide as the most active slope within the Shawan Gully, particularly below the Ls2 region. We selected deformation pixels P1 and P2 on the Nuole landslide from S1AP26 and S1DP62 measurements (Figure 14a,c). The maximum LOS deformation velocities of P1 and P2 reached 116.54 mm/yr and −139.82 mm/yr, respectively, with cumulative displacements of 957.21 mm and −1103.33 mm. To analyze the impact of rainfall on landslide activity and assess the movement trend of the Nuole landslide, we calculated the relative displacement variation (RDV) between adjacent acquisition times for pixels P1 and P2 as follows:

$$RDV={Disp}_{Total}^{N+1}-{Disp}_{Total}^N$$

(9)

where ${Disp}_{Total}^N$ represents the cumulative surface displacement at time N relative to the initial acquisition. A larger RDV value indicates a higher acceleration and stronger slope activity, and the occurrence of multiple successive high RDV values may signal that the landslide progressed to an accelerated creeping stage.

As shown in Figure 14b,d, the Nuole landslide is generally in a quasi-linear and steady creeping stage, with no significant signs of acceleration or deceleration. During the non-rainy season (November to March), RDV values for P1 and P2 remained mostly between −5 and 5 mm, indicating weak slope activity. However, during the rainfall-intensive period from April to October, RDV values showed noticeable fluctuations. Intense or prolonged rainfall disrupted the previously stable state, with RDV values generally exceeding ±10 mm and reaching a maximum of −23.8 mm. After the rainy season, RDV fluctuations diminished, and the landslide returned to a steady creeping state with quasi-linear or sublinear displacements.

The results indicate a high sensitivity of the Nuole landslide to rainfall-induced instability. Rainfall changes the original movement trend of the Nuole landslide and intensifies its activity behavior in the short term. The RDV metric proves to be a useful indicator for capturing short-term deformation accelerations, offering practical value for early warning and risk mitigation in rainfall-sensitive mountainous regions.

5. Discussion

Prior to the SBAS-InSAR processing, we completed a suitability assessment of the InSAR monitoring scheme for involved datasets. According to the results, we masked areas prone to layover and shadow as well as regions with zero monitoring sensitivity from the SLC images. Meanwhile, we reprocessed the SAR images of S1AP26 using traditional SBAS-InSAR without masking. Figure 15 evaluates the processing results of the two SBAS-InSAR through RMSE, V_precision, 1/ADI, and Velocity, analyzing the impact of the suitability assessment on InSAR measurement accuracy. The results indicate that the proposed suitability assessment method significantly improves data quality and measurement accuracy, specifically demonstrated as follows:

(1)

RMSE: This metric evaluates the phase quality of the interferograms, where a lower RMSE value indicates smaller phase errors and higher quality. Post-masking, the average interferometric phase error decreased from 1.55 rad to 1.41 rad, with the standard deviation reducing from 0.33 to 0.25, signifying a notable reduction in phase error.

(2)

1/ADI: Higher 1/ADI values indicate better phase stability of the pixels and more reliable results. The masking process had minimal impact on the phase mean and standard deviation (1/ADI), with no significant difference observed before and after processing. This suggests that geometric distortions affect the magnitude of interferometric phase errors rather than the phase distribution itself. The result also indicates that reducing sources of phase error is crucial for improving InSAR monitoring quality.

(3)

V_precision: This parameter assesses the precision of deformation measurements, with lower V_precision values indicating higher measurement accuracy. After masking, the mean V_precision value decreased from 1.45 mm/year to 1.18 mm/year, and the standard deviation dropped from 0.33 mm to 0.30 mm, demonstrating that masking reduces error and uncertainty, thereby enhancing the precision of deformation rate measurements.

(4)

Velocity: Masking reduced the variability in average rate measurements, with the standard deviation of mean velocity dropping from 13.38 mm/yr to 12.85 mm/yr, indicating more concentrated and stable measurement results.

In summary, the masking process based on applicability assessment results not only reduces phase error and enhances velocity measurement accuracy, but also stabilizes and makes the data more consistent. Therefore, it is recommended to conduct necessary suitability assessments before further InSAR processing to improve the precision and reliability.

6. Conclusions

Shawan Gully is a large landslide-induced debris flow valley. The Nuole and Huajiaoshu landslides are two high-elevation landslides within this valley, serving as primary sources of debris flow. We used SBAS-InSAR and multi-orbit Sentinel-1 SAR imagery to monitor surface deformation in the Shawan Gully, with a focus on analyzing the displacement and activity of the Nuole landslide. The main conclusions are the following:

• This study used 688 Sentinel-1 SAR scenes collected from three orbits: S1AP26, S1AP128, and S1DP62. Before SBAS-InSAR processing, we proposed a method to evaluate the suitability of the InSAR monitoring scheme based on geometric distortions and monitoring sensitivity. The results show that the SAR imagery from Sentinel-1 ascending path 128 is severely affected by layover and shadow, making it unsuitable for refined monitoring of the Nuole and Huajiaoshu landslides. Among the remaining two orbits, the descending path 62 provides higher reliability for surface deformation monitoring.
• After thinning the ground surface extent, we used SBAS-InSAR to process the refined SAR datasets. To eliminate residual fringes caused by topographical effects, we utilized PS-GCPs with stable phase and high coherence to refine the satellite orbit and re-flatten the interferograms. The refined SBAS-InSAR results of multi-orbits indicate significant surface deformation at the rear scarp and toe of the NL1 section on the Nuole landslide. Additionally, deformation features were observed in the BT2 collapse area, the NL2 section of the Nuole landslide, and the toe of the Huajiaoshu landslide using descending orbit 62 data. For detailed deformation monitoring and analysis of movement characteristics in different regions of the landslide, we estimated the surface deformation rates in both vertical and horizontal directions. The results show significant displacement in the NL1 area in both directions, with consistent movement in the main body of the landslide, while the northern scarp and the foot of the slope exhibited different movement characteristics. Time series displacement demonstrated that rainfall effectively triggers landslide instability and changes the slopes’ movement states, making it a significant factor influencing landslide activity.
• We discussed the role of suitability assessment in enhancing InSAR monitoring accuracy through RMSE, V_precision, 1/ADI, and Velocity. The mean and standard deviation of RMSE and V_precision decreased, indicating improved deformation measurement accuracy. The standard deviation of mean deformation velocity decreased from 13.3788 mm/yr to 12.8514 mm/yr, indicating more stable and consistent measurement results. This also confirms that reducing error sources is an effective way to improve InSAR monitoring accuracy. Future research could explore the applicability of these methods in larger and more diverse study areas and investigate other factors that may influence InSAR monitoring accuracy to further enhance the quality and reliability of surface deformation monitoring.