Introduction

The Xiaojiang River Basin, situated in the mountainous region of the upper Yangtze River in southwestern China, has gained renown for its recurrent geological hazards, particularly debris flows and landslides1. Recognized as the ‘Natural Museum of Debris Flows’, this area has attracted extensive attention from the research community2,3,4. In recent years, the implementation of effective preventive measures such as sediment trapping dam construction, river channel reformation, and slope reinforcement has led to a gradual reduction in the frequency of debris flow disasters in the Xiaojiang River Basin. However, landslides remain a persistent and covert threat. Field surveys combined with remote sensing interpretation revealed that 176 landslides exceeding 50,000 m3 in magnitude have occurred in the basin5. These landslide events have caused significant degradation of local ecosystem integrity, with pronounced geomorphological alterations exacerbating hydrological vulnerability and sediment transport potentials. Understanding the evolutionary processes of landslides, particularly their deformation characteristics, is crucial for effective hazard monitoring and early warning systems.

It is well-established that slopes experience primary, constant, and accelerated creep before rupture6. During the creep stage, the slope manifests gradual, uniform, and minor subsidence or deformation, accompanied by the appearance of cracks on the slope surface and the gradual tensile rupture of the rear edge. The duration of this stage can vary from several days to decades, contingent upon geological conditions and other factors. Thecontinuous deformation or subsidence over these periods changes the original surface structure of such slopes. When subjected to external factors such as heavy rain, prolonged precipitation, or earthquakes, slopes may experience accelerated deformation, ultimately leading to slope instability and triggering landslides. Previous endeavors primarily concentrated on qualitatively analyzing disasters in specific areas or statically describing individual landslides, often overlooking unlisted potential hazards. The extensive area, severe soil erosion, and difficulties in data acquisition pose significant challenges to the application of traditional geodetic techniques for landslide research in this region. These limitations are primarily reflected in the insufficient spatiotemporal resolution of data, the reliance on a single observation method, and challenges in covering large areas. Therefore, there is an urgent need to adopt advanced remote sensing technologies and multi-source data fusion approaches to enhance the ability to identify and monitor landslides in this region.

Surface deformation is widely recognized as the most intuitive macroscopic expression indicating landslide development, thereby serving as the primary indicator for hazard assessment and stability analysis7,8. An in-depth analysis of the structural changes on the slope surface not only helps to accurately assess the boundary extension and current stability of the slope but also serves as an efficient and reliable method to identify potential landslides. Demonstrating proficiency in monitoring subtle surface movements, Interferometric Synthetic Aperture Radar (InSAR) is a critical tool for monitoring surface movements, essential for assessing geological hazards9,10,11,12,13. In regions with severe weather and complex terrain, SAR imagery often represents the sole viable technique for observing active landslides and identifying previously unrecognized ones14,15. However, InSAR measurements only reflect the displacement in a Line-of-Sight (LOS) direction16,17. This limitation poses challenges in accurately analyzing the mechanisms and dynamics of landslide displacement, particularly in complex terrains18,19. To overcome the directional limitations, InSAR measurements can be projected in a specific direction to enhance landslide analysis3,20,21.

Over the last two decades, landslides resulted in over 300 fatalities and substantial economic losses in the Xiaojiang River Basin. In response to these safety concerns, the ‘Poverty Alleviation Resettlement’ project was initiated, which relocated residents to safer areas, away from the threat of future landslides. Consequently, the early identification and consistent monitoring of potential landslides in this region are essential not only for ensuring the safety of residents and the protection of property but also for disaster prevention and environmental conservation22. This paper presents a Sentinel-1 SAR dataset-based case study that identified 59 potential landslides in the Xiaojiang River Basin through advanced InSAR analyses. Our findings not only pinpoint potential landslide sites but also offer insights into slope stability by integrating InSAR time-series displacements with precipitation data, crucial for informed disaster management and environmental conservation.

Study area and datasets

Geological setting

The Xiaojiang River Basin in Yunnan Province is located in southwestern China, at the southeastern margin of the Tibetan Plateau (102.941° E to 103.349° E, 25.763° N to 26.569° N; red rectangle in Fig. 1a,b). Its unique geological background makes it a high-risk area for geological hazards. The basin is characterized by well-developed fault structures, particularly the Xiaojiang Fault Zone, which is an active fault system with a pronounced dextral strike-slip feature (Fig. 1d). The interaction between the Xiaojiang Fault Zone and adjacent secondary faults results in complex geological structures and high-stress accumulation in the region. Geomorphologically, the Xiaojiang River Basin is dominated by steep mountainous terrain and deep gorges, with a fragmented landscape and significant elevation differences ranging from approximately 648 m at the confluence of the Xiaojiang and Jinsha Rivers to nearly 4,000 m at the highest peaks (Gongwang Mountain). the Dongchuan District has an average elevation of about 1,200 m, situated within the transitional zone of the basin23. These dramatic topographic variations provide the fundamental conditions for landslides and debris flows.

The region experiences a subtropical monsoon climate, with annual rainfall concentrated during the summer months. Due to the steep topography, intense rainfall easily leads to rapid surface runoff, causing strong erosion on slopes and triggering landslides or debris flows. The geological formations in the Xiaojiang River Basin are primarily composed of Mesozoic to Cenozoic sedimentary, volcanic, and metamorphic rocks, which exhibit complex lithological characteristics and unstable rock masses. The long-term tectonic evolution has resulted in fractured fault rocks and breccia zones, which are highly susceptible to collapse and sliding under external forces. Moreover, the area has experienced frequent mining activities, particularly since the late 20th century, with large-scale anthropogenic engineering disturbances further exacerbating the instability of geological bodies, thereby significantly increasing the risk of landslides and debris flows.

The combined effects of well-developed faults, complex lithology, steep topography, and significant human activities make the Xiaojiang River Basin highly prone to frequent landslides24,25. A 2018 field survey confirmed 248 landslides, posing threats to at least 127 villages and communities along the Xiaojiang River and its tributaries (Fig. 1c).

Fig. 1
figure 1

(a) Location of the Xiaojiang River Basin (red rectangle); (b) Coverage of Sentinel-1 SAR images (black rectangles); (c) Distribution of debris flows and landslides (Field investigations and documentation); (d) Xiaojiang fault and typical debris flow gully. ①: Taipingcun Gully, ②Jiangjia Gully, ③Daqiaohe Gully, ④Tianbagan Gully, ⑤Shen Gully, ⑥Shiyang Gully, ⑦Lali Gully, ⑧Dabaini Gully, ⑨Xiaobaini Gully, ⑩Xiaohe Gully, ⑪Pingzidi Gully, ⑫Shawan Gully (Nuole landslide), ⑬Xiaojiang Fault Zone, ⑭Jinsha River, ⑮Yili River, ⑯Kuai River, ⑰Xiaojiang River. (e–g) Slopes within the Xiaobaini Gully.

Full size image

Sentinel-1 A SAR imagery

Typically, advanced InSAR analysis requires a minimum of 15–20 SAR images, with several tens of images used in most InSAR applications26. Over the past few years, Sentinel-1 C-band radar imagery has been used extensively in various research fields associated with surface deformation27,28,29,30. In this study, two Sentinel-1 SAR datasets were used, comprising 100 from an ascending orbit and 92 from a descending orbit. The coverage of these images is depicted in Fig. 1b. Two contiguous SLC images from the same orbit should be initially mosaicing to adequately span the study area. Subsequently, all images underwent resampling and coregistration against a designated super-master SLC image, setting the stage for subsequent time-series InSAR processing. Table 1 includes specific details such as the heading direction and incidence angle, derived from the central image of each respective orbit.

Table 1 Basic parameters of Sentinel-1 imagery data usedin this paper.
Full size table

Figure 2 shows the temporal-spatial baseline distributions of the interferograms generated by SBAS-InSAR. The thresholds for the temporal and spatial baselines were set at 90 d and 200 m, respectively. From ascending orbit 26, 195 interferograms were generated, with a maximum temporal baseline of 84 d and a spatial baseline reaching up to 158 m. From descending track 62, 178 interferograms were generated, with maximum temporal and spatial baselines of 84 d and 192 m, respectively.

Fig. 2
figure 2

Spatio-temporal baseline distribution of interferograms obtained using SBAS-InSAR. (a) Ascending orbit 26. (b) Descending orbit 62. Green solid circles represent data, yellow solid circles represent super-master images, and black straight lines represent interference pairs.

Full size image

Other data

A reliable landslide inventory map is an essential initial step for conducting comprehensive hazard and risk analyses. However, such maps are available only for some countries, primarily in specific regions31. Unfortunately, no landslide inventory is available for the Xiaojiang River Basin. To compile a landslide inventory map for this area, we reviewed previous investigations. The mapping process involved deduplication and spatial correlation analysis to identify landslide occurrences. Data were primarily sourced from two origins: (1) documented results of surveys or studies (indicated by origin triangles in Fig. 1c)32,33, and (2) field investigations conducted in 2018 (represented by green triangles in Fig. 1c). Utilizing these data, duplicates were removed in the spatial dimension, resulting in the creation of a landslide inventory that includes 339 landslides (Fig. 1c).

In this study, Precise Orbit Ephemerides were downloaded to correct satellite orbit errors and improve the registration accuracy. A digital elevation model (DEM) with a spatial resolution of 12.5 m was obtained from data collected by the ALOS-1 satellite, accessible at https://vertex.daac.asf.alaska.edu/. Rainfall data provided by the Dongchuan Debris Flow Observation and Research Station, Chinese Academy of Sciences, were used to analyze the response mechanism of the slope stability to rainfall. All optical images used in this study were sourced from Google Earth to assist in the analysis of InSAR measurement results.

Methods

We utilized SBAS-InSAR to process Sentinel-1 SAR images, and detected potential landslides in the Xiaojiang River Basin. The processing workflow is illustrated in Fig. 3.

Fig. 3
figure 3

The technical scheme of SBAS-InSAR with PS-GCPs for measuring surface deformation and detecting potential landslides in the Xiaojiang River Basin.

Full size image

SBAS-InSAR processing

In this study, we used SBAS-InSAR to process the Sentinel-1 A ascending and descending acquisitions. The standard SBAS method establishes a model for deformation-phase estimation based on high-coherence pixels, a robust algorithm for time-series analysis34,35. However, 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 unwrapping36. To mitigate this issue, we proposed an approach that utilizes stable ground control points (GCPs) to refine the satellite orbit and re-flatten the images37. In our SBAS-InSAR processing, these GCPs are used to fit the orbital refinement polynomials. In our SBAS-InSAR processing, these GCPs were integral in fitting the orbital refinement polynomials, aiming to remove residual fringes arising from topographical effects. However, 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 stable ground control points (GCPs), referred to as PS-GCPs. Initially, it was necessary to generate interferograms using the same master SLC image. We then calculated the average amplitude index (ADI) of the SAR images and employed Eqs. (1) and (2) to refine the selection of the GCP:

$$\bar{\alpha }\ge \bar{A}+\sigma$$
(1)
$${{\Lambda }}_{j}=\frac{{\sigma }_{A}}{\bar{\alpha }}$$
(2)

where \(\bar{A}\) and \({\sigma }_{A}\) represent the mean amplitude and the standard deviation of the mean amplitude image, respectively, and \(\bar{\alpha }\) refers to the mean time-series amplitude. Equation (1) indicates that pixels with higher amplitudes exhibit better coherence. The mean-amplitude image was the average of the amplitudes obtained from multiple SAR images. Equation (2) represents the ADI for pixel j. Decreasing ADI values correspond to increased phase stability of the target pixels. In InSAR processing, the ADI can be used to identify pixels that are likely affected by noise or other sources of error, enabling their exclusion from further analysis38. For Gaussian scatterer pixels, PS-GCPs ensure high coherence, signal-to-noise ratio, and phase stability even in cases of low overall surface coherence.

Multi-directional surface deformation estimation

InSAR directly provides surface displacement in the LOS direction. However, in most cases, this singular measurement often failsto capture the entirety of the actual deformation. To address this issue, it is necessary to estimate the deformation components in a specific direction. In the study, we estimated the deformation components of the surface in the direction of east-west (E-W), vertical (U-D), and downward sliding (D-S) along the slope.

Vertical and horizontal deformation

The deformation measured by the time-series InSAR represents the projection of actual 3D displacements into a LOS direction39,40. This projection relationship is mathematically defined as:

$${V}_{los}={n}_{N-S}{V}_{N-S}+{n}_{E-W}{V}_{E-W}+{n}_{U-D}{V}_{U-D}$$
(3)

where, Vlos VN−S, VE−W, and VU−D represent the deformations of the ground surface in the LOS, N-S (north-south), E-W, and U-D, respectively. The unit vectors for these deformation components are denoted as nN−S, nE−W, and nU−D. At least three distinct InSAR measurements are required to establish a three-dimensional deformation field41,42. However, this study utilized only two SAR datasets. Consequently, Eq. (3) was reformulated into a two-dimensional deformation model because of the insensitivity of satellite observations to the settlement in the N-S direction43,44:

$$\left[\begin{array}{c}{V}_{los}^{A}\\ {V}_{los}^{D}\end{array}\right]=\left[\begin{array}{cc}-\text{cos}{\phi }_{A}\cdot \text{sin}{\theta }_{A}& \text{cos}{\theta }_{A}\\ -\text{cos}{\phi }_{D}\cdot \text{sin}{\theta }_{D}& \text{cos}{\theta }_{D}\end{array}\right]\left[\begin{array}{c}{V}_{E-W}\\ {V}_{U-D}\end{array}\right]$$
(4)

where, θ is the radar incidence angle, and φ is the satellite flight direction. A and D represent the ascending and descending orbits, respectively.

Downward-sliding deformation

With some assumptions, the LOS deformation can be transferred to measurements from a specific angle and azimuthal direction21. In this study, the translational sliding movement in the D-S direction was estimated using the following equations (VD−S)21,45:

$${V}_{D-S}=\frac{{V}_{los}}{{n}_{los}\cdot {n}_{D-S}}$$
(5)

where vector \({n}_{D-S}\) is the unit vector in the D-S direction and can be rewritten in the north–east–up format:

$${n}_{D-S}=\left[\begin{array}{c}{\gamma }_{N-S}\\ {\gamma }_{E-W}\\ {\gamma }_{U-D}\end{array}\right]=\left[\begin{array}{c}-\text{cos}A\cdot \text{sin}\omega \\ -\text{sin}A\cdot \text{sin}\theta \omega \\ \text{sin}\omega \end{array}\right]$$
(6)

where A represents the slope azimuth angle, and ω represents the local slope gradient.

Results

Ground surface deformation analysis

LOS deformation derived by SBAS-InSAR

Figure 4 provides an overview of the mean velocity field from 2014 to 2018, with Panel (a) representing the ascending orbit and Panel (b) the descending orbit. Positive values indicate movement towards the satellite, while negative values indicate movement away from the satellite.

In Fig. 4a and 81.23% of the surface velocities ranged from − 10 to 10 mm/yr, compared to 83.15% in Fig. 4b. A total of 13 prominent deformation areas were observed in the ascending monitoring results (a1-a13 in Fig. 4a), while 11 were in the descending monitoring results (b1-b11 in Fig. 4b). According to the terrain map, areas a11, a12, and a13 were flat, whereas the other areas were located on slopes. These regions were distributed along the banks of the Xiaojiang River and its tributaries, as well as along the slopes of typical debris flow gullies, such as Jiangjia, Dabainihe, and Xiaobainihe Gully. In the ascending monitoring results, the NuoLe landslide exhibited the highest deformation velocity of 105.39 mm/yr towards the satellite, while the north slope of the Menqian Gully showed the highest deformation velocity of -147.81 mm/yr away from the satellite. In descending monitoring results, The Menqian Gully displayed the fastest deformation velocity towards the satellite at 84.67 mm/yr, whereas the Nuole landslide had the quickest deformation velocity of − 110.54 mm/yr away from the satellite. Significant deformation signals on the middle and lower sections of the Nuole landslide were detected by both Sentinel-1 ascending and descending datasets, with an areas of 0.94 km2 and 1.70 km2, respectively. However, in the Menqian Gully, InSAR measurements through Sentinel-1 ascending datasets revealed extensive deformation characteristics on both the northern and southern slopes, covering an area of 4.18 km2, while the Sentinel descending datasets only detected deformation from northern slope, with an area of 1.27 km2. This difference is likely attributed to shadow distortion in the Sentinel-1 descending imagery on the southern slope, which prevented effective deformation monitoring.

Fig. 4
figure 4

LOS deformation velocities in ascending (a) and descending (b).

Full size image

Deformation components in the E-W and U-D directions

Figure 5a and b respectively show the deformation components in the E-W and U-D directions. The results reveal significant deformations on old landslides or slopes in the debris-flow gullies in U-D and E-W directions. Significant subsidence deformations were identified at the Bahe River, Taipingcun Gully, Jianshan Gully, Lao’ao Gully, Jiangjia Gully, Xinshui Gully, Dabaini Gully, and Shawan Gully. The subsidence velocity detected on the southern slope of Duozhao Gully reached − 108.25 mm/yr. Horizontally, the maximum eastward deformation velocity of 111.32 mm/yr was observed on the northern slope of the Menqian Gully in the E-W direction, and the maximum westward velocity of –144.82 mm/yr occurred at the Nuole landslide in Shawan Gully. Along the Bahe River, we identified eight significant deformation areas in the E-W direction, while only two were detected in the U-D direction. The slopes of Xinshui Gully exhibited pronounced deformation in the E-W direction while remaining stable in the U-D direction. Conversely, the slopes of Maojia and Caohaizi Gully experienced maximum deformation velocities exceeding 100 mm/yr in the U-D direction, while remaining stable in the E-W direction.

Deformation components in the D-S direction

When calculating slipping velocity in the D-S direction, we use ALOS DEM to simulate terrains and exclude areas where landslides are less likely to occur. The D-S deformation of the remaining terrain was preserved, and pixels with a VD−S greater than 0 were discarded. A grid covering the entire study area was created in ArcMap, with each grid cell having the same dimensions as the VD−S pixels. The slope sliding velocities from ascending track 26 and descending track 62 were assigned to the corresponding grid cells, and grids without any pixels were removed. Different assignment methods were employed based on the specific values obtained if more than one value was distributed within a single grid. Pixels with sliding velocities of less than − 16 mm/yrwere assigned the minimum value in the corresponding grid to highlight the potential instability of the landslide, while pixels with sliding velocities greater than − 16 mm/yr were assigned the maximum value to emphasize the slope stability. The integrated slope-sliding velocity estimation results for Xiaojiang Basin obtained after applying the procedure above are shown in Fig. 5c. A total of 11 areas with significant sliding signals were identified along the downslope direction, including The Nuole landslide, and slopes within Jiangjia Gully, Dabaini Gull, Xiaobaini Gully, and Taipingcun Gully.

Fig. 5
figure 5

Projection of LOS surface deformation in the three directions. (a) in the E-W direction, where positive values represent eastward movement, and negative values represent westward movement. (b) in the U-D direction, where positive values indicate eastward displacement and negative values indicate westward displacement. (c) in the D-S direction, where negative values indicate that sliding occurred downwards along the slope.

Full size image

InSAR suitability assessment

Terrain visibility and geometric distortions in assessing InSAR monitoring effectiveness

The complex terrain of alpine canyon regions poses significant challenges for disaster monitoring. A fundamental requirement for employing InSAR technology to detect surface deformations is the capability to accurately monitor displacements in the LOS direction. However, the uncertainty involved in acquiring deformation information due to terrain relief can lead to various geometric distortions during the formation of SAR images. These distortions include foreshortening, layover (both active and passive), and shadow (both active and passive), which can substantially diminish the efficacy of InSAR deformation monitoring efforts46. In this study, we integrated the LSM (Layover and Shadow Map) method and R-index model to assess the terrain visibility of Sentinel-1’s ascending and descending imagery in the Xiapjiang River Basin47,48,49. Slope and aspect were estimated using the ALOS DEM, with satellite flight parameters detailed in Table 1.

Geometric distortions generated in Sentinel-1 SAR images are depicted in Fig. 6. The proportion of areas with good visibility (GV) in ascending imagery was 48.37%, which was higher than the 36.61% observed in descending imagery. Foreshortening (Fs) emerged as the most prevalent type of distortion, affecting more than 40% of the area in both ascending and descending images. In ascending imagery, the proportions of layover and shadow were 3.86% and 3.72%, respectively, primarily occurring along the eastern slopes of the Xiaojiang River, Kuaihe River, and Qinglong River, as well as in some mountainous regions. Conversely, in descending imagery, both layover and shadow percentages were higher, accounting for 8.72% and 10.75%, respectively.

In this study, we used Eq. (7) to calculate the density of deformation pixels in various areas and to discuss the accuracy of InSAR deformation monitoring results:

$${\text{A}\text{v}\text{a}}_{GD}=\frac{{InSAR}_{GD}}{GD}\times 100 {\%}$$
(7)

where AvaGD represents the availability of InSAR monitoring for a specific type of geometric distortion, GD denotes the total pixel number of each geometric distortion, and InSARGD represents the number of InSAR monitoring pixels in the region with that type of geometric distortion.

Fig. 6
figure 6

(a, b) are terrain visibility in the ascending track 26 and the descending track 62. (c, d) are coverage areas for each distortion type in the ascending track 26 and the descending track 62. GV: good visibility, Fs: foreshortening, ASH: active shadow, PSH: passive shadow, AL: active layover, PL: passive layover.

Full size image

Table 2 provides statistics on the distribution density of InSAR deformation monitoring pixels in different distorted regions for Sentinel-1’s ascending and descending images. Pixels located in areas with good visibility were considered valid monitoring values, whereas those in regions affected by layover and shadow were deemed unreliable. Generally, the higher the proportion of valid values and the lower the proportion of unreliable values, the more effective the InSAR monitoring is considered. In this study, the densities of valid monitoring values in ascending and descending orbit images reached 93.06% and 86.20%, respectively, which were higher than those in all geometrically distorted regions.

Table 2 Statistics of InSAR deformation pixel density in Sentinel-1 ascending and descending images.
Full size table

Monitoring sensitivity in assessing InSAR reliability and applicability

The Sentinel-1 satellite exhibits varied deformation-detection capabilities across different slope units, applicable in both ascending and descending orbits. Researchers have suggested the significance of the spatial correlation between LOS deformation and actual slope deformation for assessing InSAR monitoring sensitivity50. This sensitivity defines the ability of SAR sensors to discern surface deformations across varied terrains51. The C-index, calculated as follows, serves as a quantitative measure of InSAR monitoringsensitivity and aids in evaluating the reliability of deformation estimates:

$$C - index = - \sin \phi \sin \theta \cos \alpha \cos \beta + \sin \phi \cos \theta \sin \alpha \cos \beta + \cos \theta \sin \beta$$
(8)

where θ is the local incidence angle of the radar signal, and φ is the angle between the satellite flight direction and the north direction, approximately 12° in ascending orbit and 168° in descending orbit. Furthermore, α and β denote the slope aspect and gradient, respectively. The theoretical value of C-index ranges from − 1 to 1. LOS displacements derived from InSAR are typically defined as positive (negative) values when the landslide moves towards (away from) the SAR sensor52. Consequently, a positive C-index correlates with a positive LOS displacement in InSAR monitoring, while a negative C-index corresponds to a negative LOS displacement, signifying the displacement pattern of reliable pixels.

The C-index map of the study area for the ascending and descending Sentinel-1 images are shown in Fig. 7a and b, respectively. Pixels with a C-index approaching ± 1 are deemed to have more reliable deformation monitoring values. The study area is segmented into four zones based on the C-index values: High Applicability Zone (HAZ, [− 1 to − 0.6] and [0.6 to 1]), Medium Applicability Zone (MAZ, [− 0.6 to −  0.3] and [0.3 to 0.6]), Low Applicability Zone (LAZ, [−  0.3 to −  0.1] and [0.1 to 0.3]), and Non-Applicability Zone (NAZ, [−  0.1 to 0.1]). As shown in Fig. 7c, the applicable areas for InSAR monitoring in ascending and descending images accounted for 86% and 82%, respectively, with the HAZ covering the largest proportion at 36% and 35%. The NAZ accounted for 14% and 18%, substantially lower than the applicable areas.

Figure 7d,g display the surface deformation monitoring results for the Xiaobaini and Dabaini Gully using ascending and descending Sentinel-1 SAR images, respectively, while Fig. 7e,h depict the surface classification based on the C-index. The slopes of both gullies are identified as potential landslide areas. From Fig. 7d,g, ascending images detected significant displacements on the northern slope, while descending images observed pronounced displacements on the southern slope. Most northern slopes in both Dabaini and Xiaobaini Gully fell within the InSAR HAZ areas in the ascending images, but within the LAZ in the descending images (Fig. 7e,h). Conversely, the majority of the south slope belonged to the HAZ areas in descending images but the LAZ areas in ascending images. Based on these results, we assessed the consistency between the LOS deformation and C-index calculated from Sentinel-1 ascending and descending images (Fig. 7f,i). It’s indicated that the deformation-prone areas on the slopes of Dabaini and Xiaobaini Gully exhibit LOS deformation and C-index values with the consistent positive or negative signs, thereby demonstrating the reliability of the InSAR monitoring and potential landslide identification in this study.

Fig. 7
figure 7

(a, b) are the C-index maps of the study area for the ascending and descending Sentinel-1 images, respectively; (c) presents the statistical results of different InSAR applicability zones for the ascending and descending Sentinel-1 images, respectively; (d)–(i) depict the LOS deformation velocity maps, C-index maps, and the consistency between LOS deformation values and C-index values for ascending and descending orbit images, respectively.

Full size image

Potential landslides identification

Spatial distribution and boundary determination

The decompositions of InSAR measurements in different directions allowed us for the extraction of significant surface deformation, effectively filtering out areas with low deformation velocity (-10 mm/yr < Velocity < 10 mm/yr). By utilizing ALOS DEM to generate topographic and slope maps, we excluded flat terrain units, retaining only the slopes. In accordance with the research by Zhang et al. (2018), we defined areas with a slope of less than 8° as flat terrain units. Finally, we used Google Earth optical imagery to precisely delineate slope boundaries within the deformation zones. This process yielded the final results of potential landslide identification in the Xiaojiang River Basin, as depicted in Fig. 8. The study area, covering approximately 1845.33 km2, revealed 59 potential landslides. Among these, 27(45.76%) landslides exhibited strong deformation signals in all three directions (E-W, U-D, D-S), while 12 (20.34%) were detectable in any two directions. Detection efficiency was highest in the E-W direction with 48(81.36%) potential landslides, followed by the U-D direction with 41(69.49%) landslides, and the lowest in the D-S direction with 36(61.02%) landslides.

We calculated the area of the 59 landslides, with the largest covering 5.34 km2. Since InSAR is to measure ground surface deformation, directly determining the depth of landslides remains challenging. However, by integrating Google Earth optical imagery (e.g., cracks, scarps, and vegetation covers), we could preliminarily classify these landslides as either shallow or deep-seated (Table 3). In this study, we defined shallow landslides as those with higher deformation velocities, smaller spatial extents, clear boundaries and scarps, and exposed soil surfaces. In contrast, deep-seated landslides were characterized by slower deformation velocities, larger affected extents, diffuse boundaries, and partial vegetation cover. Following these criteria, we classified 41 landslides as shallow and the remaining 18 as deep-seated.

Slope #43 is a typical shallow landslide and has been documented in the landslide inventory (26° 07′ 00.44″ N, 103° 09′ 07.47″ E). Collapse features are observed obviously on the slope surface, with accumulated rock debris at the toe. InSAR measurements indicate that its maximum deformation velocities in the U-D and E-W directions are − 30.84 mm/yr and 22.64 mm/yr, respectively. The landslide is less than 200 m from the bridge, and poses a significant risk of severe damage to the road and bridge infrastructure. Additionally, six deep-seated landslides were identified within the Jiangjia Gully watershed, with the largest one covering an area of approximately 4 km2. InSAR monitoring indicates that these deep-seated landslides exhibit annual average deformation rates exceeding 20 mm/yr in the E-W direction and 35 mm/yr in the DS direction. Historically, Jiangjia Gully has been frequently affected by debris flow disasters, resulting in prolonged erosion and geomorphic reshaping of the watershed. This long-term geomorphic evolution has led to the continuous removal, transport, and redeposition of loose surface materials on the slopes, creating favorable conditions for the development of deep-seated landslides.

Fig. 8
figure 8

Spatial distribution of identified potential landslides. (a)–(c) display landslides detected in the U-D, E-W, and D-S directions, respectively, and (d) presents the consolidated results following deduplication.

Full size image
Table 3 The identified landslides in this paper.
Full size table

The landslide inventory indicated that 25 of the potential landslides had previously been recorded, while 34 were newly identified. Areas susceptible to landslides are predominantly situated along the banks of typical debris-flow gullies and rivers, characterized by sparse vegetation cover. Historically, debris flows have frequently occurred in the Xiaojiang River Basin, sourcing their materials primarily from slope deposits in the middle and upper reaches of the gullies. In Fig. 8, slopes within typical debris flow gullies, such as those in Jiangjia, Dabaini, Xiaobaini, and Taiping Gully, are identified as potential landslide sites in all three directional analyses. The Nuole landslide, known for recurring secondary slides averaging tens of thousands of cubic meters annually, has resulted in over 50 fatalities in the past 100 years.

The boundaries of potential landslides were confirmed through Google Earth optical imagery. The total area of these landslides is approximately 55.91 km2, with 71.18% covering less than 1 km2. Multi-temporal optical imagery from Google Earth verified that 17 potential landslides exceeded 1 km2. Among these, 14 slopes exhibited significant movement between 2014 and 2018. The spatialdistribution and extent of these historical displacements are depicted in Fig. 9.

Fig. 9
figure 9

The spatial distribution of 14 potential landslides with an area exceeding 1 km2. According to the InSAR results and optical images, the red lines delineate the landslide boundaries, and blue polygons with orange centers are areas where the landslides have experienced significant displacements in the past.

Full size image

Quantitative statistics of potential landslides

Elevation and slope are two key factors that significantly influence landslide occurrence. Our results indicate that landslides primarily occur within specific elevation bands, particularly in the mid-to-high altitude regions ranging from 1,500 to 3,000 m, which account for 66.10% of all landslide events. In this elevation range, steep slopes and complex geological structures increase regional instability, thereby enhancing susceptibility to landslides. In contrast, in lower elevation areas (below approximately 1,000 m), landslide events are relatively rare, with only two potential landslides identified. This distribution trend can be attributed to the relatively flat terrain, less distinct landform features, and lower frequency of landslide activity compared to higher elevations.

In analyzing the impact of slope, we identified 49 potential landslides within slopes ranging from 8° to 40°, representing 83.05% of the total. When the slope is less than 8°, only four potential landslides were identified. This is because the driving force for slope movement decreases with a reduction in slope, making landslides unlikely on such gentle slopes. Similarly, landslides are also rare on slopes exceeding 40° due to the scarcity of unstable surface deposits as the slope steepens. These findings are consistent with previous studies by Zhang et al. (2018), affirming the reliability of our results. These findings underscored the critical role of topographic factors in governing landslide dynamics.

Discussion

Precipitation effect on landslide stability

In this study, we focused on two specific deformation points, labeled P1 and P2 in Fig. 10b, located on the slopes within Jiangjia Gully, to assess their geostability. Point P1 was strategically selected from the central section of the northern slope near Lijiayakou, which was indicative of typical slope deformation characteristics. Point P2 was selected from an active landslide zone within the Menqian Gully. The stability of these points was evaluated across various periods using the relative displacement variation as a primary indicator. Precipitation is recognized as a key factor influencing landslide stability and is commonly employed as a variable for landslide stability assessment or susceptibility evaluation53,54. Our analysis incorporated this variable. By utilizing rainfall observation data provided by the Chinese Academy of Sciences, we conducted a detailed examination of the precipitation’s impact on the stability of the landslide-prone regions delineated in Fig. 10c,d.

Our study detailed the influence of concentrated rainfall occurring from March to October each year on landslide dynamics. During periods of sparse rainfall, the slopes exhibited near-uniform motion, with relative deformation at Points P1 and P2 generally remaining below 10 mm. During this time, P1 showed irregular deformation with a downward trend, while P2 remained relatively stable, with a maximum relative deformation not exceeding 20 mm. However, during episodes of heavy rainfall, the sliding velocity of the slopes increased markedly, resulting in significant changes in relative deformation and activity intensity. The influence of precipitation on slope stability manifests in the following aspects:

(1) Increased sliding intensity: As shown in Fig. 10c, slope activity during non-rainy seasons was characterized by uniform subsidence with low intensity. Heavy rainfall led to soil saturation and increased pore water pressure, which, during periods TA1, TA2, and TA3, caused rapid subsidence of P1, with a corresponding rise in activity intensity. However, rapid changes in landslide subsidence velocity did not necessarily coincide with the periods of peak rainfall intensity (e.g., TA2), but often exhibited lag effects (e.g., TA1 and TA3). After continuous rainfall ceased, activity intensity returned to a low state.

(2) Disruption of inherent slope stability: As shown in Fig. 10d, P2 maintained a steady state over an extended period, with a lower activity intensity. During periods TB1 and TB2, continuous rainfall compromised stability, resulting in a rapid transition to either uplift or subsidence. From April to November 2017, the maximum relative deformation of P2 exceeded ± 50 mm, indicating the highest intensity of slope activity. Following continuous rainfall, the activity intensity returned low.

Fig. 10
figure 10

(a) Location of Jiangjia gully; (b) distribution of precipitation observation stations (YZL03-09, and 11) and deformation points (P1 and P2); (c, d) represent respectively the influences of precipitation on the stability of P1 and P2. TA1, TA2, TA3, TB1, and TB2 represent a period respectively.

Full size image

We conducted a drainage analysis in the Jiangjia Gully watershed to evaluate the impact of rainfall on surface runoff and the stability of the two landslides (Fig. 11). The results reveal distinct hydrogeomorphic effects on landslide behavior: Landslide P1, located in the upper reaches of the debris flow channel, is characterized by steep topography and elevated terrain (P1 in Fig. 11a), which limits the development of extensive catchment areas. Following rainfall events, surface runoff primarily infiltrates or disperses along the slope, resulting in a relatively minor influence on slope stability. In contrast, Landslide P2 is situated in the transport zone of the debris flow channel, proximal to multiple pour points, and receives concentrated runoff from a significantly larger upstream catchment area (P2 in Fig. 11b). During intense rainfall events, rapidly converging surface runoff exacerbates toe erosion and elevates pore water pressure, accelerating slope deformation. The combined effects of rising groundwater levels and increased pore water pressure reduce the shear strength of the landslide toe, ultimately compromising the overall stability of Landslide P2. This hydrogeomorphic process is strongly corroborated by InSAR-derived surface deformation, which exhibit remarkable spatial and temporal consistency with the predicted hydrological dynamics. Specifically, InSAR measurements indicate that during intense rainfall events, the toe region of landslide P2 exhibits significant accelerated deformation, with deformation rates fluctuating within a range of ± 100 mm/yr (Fig. 8d). In contrast, landslide P1 maintains a relatively stable deformation pattern, with the velocity variations consistently remaining below ± 10 mm/yr (Fig. 8c). These findings underscore the critical role of rainfall-induced runoff convergence in triggering slope instability, particularly in areas with concentrated drainage pathways and high catchment efficiency.

Fig. 11
figure 11

(a) Slope and elevation of landslides P1 and P2. H: elevation; S: slope. (b) A drainage analysis of Jiangjia Gully watershed.

Full size image

Ability of InSAR to detect potential landslides in mountainous areas

This study used SBAS-InSAR and Sentinel-1 SARimages from ascending and descending orbits to measure surface ground deformation in the Xiaojiang River Basin. The results indicate that these datasets are highly suitable for the study area, and 59 potential landslides were successfully identified, which demonstrated the significant advantages of SBAS-InSAR in monitoring landslide activity over large and complex terrains. Through the identification and stability analysis of these potential landslides, this study deepens our understanding of the deformation mechanisms during the landslides’ creeping stage, and provides data support for the construction of future landslide early warning systems.

Our findings are consistent with studies in other similar regions around the globe. For example, Liu et al. used a refined InSAR method to identify 317 active landslides in Deqin County on the Tibetan Plateau, and Zhang et al. detected 217 potential landslides with time-series InSAR in the Bailong River Basin52,55. These results further validate the widespread ability of InSAR in monitoring landslides in mountainous areas. However, it is important to note that the results of this study exhibited certain time lag effect when analyzing landslide activity changes under the influence of precipitation.

Although InSAR technology played a significant role in this study, due to relying only on a single remote sensing technology, it may lead to the omission of some landslide deformation details. For instance, surface deformation in shadow or overlap areas of SAR images could not be effectively observed. In future research, we are going to integrate multi-source remote sensing data, such as GNSS, LiDAR, and UAV technologies, to improve the accuracy and timeliness of landslide monitoring. The fusion of multi-source remote sensing data holds the potential to provide a more comprehensive and precise analysis of landslide dynamics, thereby enhancing the performance of landslide disaster early warning systems.

Conclusion

Landslides remain a significant geological hazard in the Xiaojiang River Basin, and accurately understanding their spatial distribution and stability is essential for effective detection, prevention, and management. This study demonstrates the capabilities of C-band Sentinel-1 SAR imagery in monitoring surface deformation patterns and identifying potential landslides across the basin. By applying the SBAS-InSAR to 100 ascending and 92 descending SAR images acquired between 2014 and 2018, we successfully monitored ground deformation across an area of approximately 1,845.33 km2. 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. Furthermore, we validated the applicability of Sentinel-1 SAR datasets in complex mountainous canyon areas and the reliability of InSAR monitoring results through analyses of geometric distortion and monitoring sensitivity.

A total of 59 potential landslides were identified, covering 55.91 km2, among which 25 were previously documented and 34 were newly detected. In the U-D, E-W, and D-S components, 41, 48, and 36 potential landslides were identified, respectively. By integrating InSAR-derived time series displacements with ground-based rainfall data, we found that precipitation plays a critical role in destabilizing slopes—not only disrupting the intrinsic stability of landslides but also accelerating the movement of already active slopes.