Deformation Mechanisms and Rainfall Lag Effects of Deep-Seated Ancient Landslides in High-Mountain Regions: A Case Study of the Zhongxinrong Landslide, Upper Jinsha River

Abstract

In high-mountain canyon regions, many settlements are located on large, deep-seated ancient landslides. The deformation characteristics, triggering mechanisms, and long-term developmental trends of these landslides significantly impact the safety and stability of these communities. However, the deformation mechanism under the influence of human engineering activities remains unclear. SBAS-InSAR (Small Baseline Subset-Interferometric Synthetic Aperture Radar) technology, UAV LiDAR, and field surveys were utilized in this study to identify a large ancient landslide in the upper Jinsha River Basin: the Zhongxinrong landslide. It extends approximately 1220 m in length, with a vertical displacement of around 552 m. The average thickness of the landslide mass ranges from 15.0 to 35.0 m, and the total volume is estimated to be between 1.48 × 10⁷ m³ and 3.46 × 10⁷ m³. The deformation of the Zhongxinrong landslide is primarily driven by a combination of natural and anthropogenic factors, leading to the formation of two distinct accumulation bodies, each exhibiting unique deformation characteristics. Accumulation Body II-1 is predominantly influenced by rainfall and road operation, resulting in significant deformation in the upper part of the landslide. In contrast, II-2 is mainly affected by rainfall and river erosion at the front edge, causing creeping tensile deformation at the toe. Detailed analysis reveals a marked acceleration in deformation following rainfall events when the cumulative rainfall over a 15-day period exceeds 120 mm. The lag time between peak rainfall and landslide displacement ranges from 2 to 28 days. Furthermore, deformation in the high-elevation accumulation area consistently exhibits a slower lag response compared to the tensile deformation area at lower zones. These findings highlight the importance of both natural and anthropogenic factors in landslide risk assessment and provide valuable insights for landslide prevention strategies, particularly in regions with similar geological and socio-environmental conditions.

1. Introduction

The sliding surface of deep-seated landslides is typically buried deeper than 30 m [1,2]. These landslides are usually slow-moving and often involve substantial volumes of soil and rock, with deformation rates generally measured over extended temporal scales, ranging from years to decades or even longer [3]. The study of large, slow-moving deep-seated landslides remains a significant challenge. Their low deformation velocity and the deep burial of the sliding surface limit the effectiveness of conventional monitoring techniques, which struggle to identify them with the necessary precision and comprehensiveness. Moreover, the complex interplay of multiple factors influencing the deformation process further complicates robust quantitative analyses.

The transition from slow to rapid deformation in large, deep-seated, slow-moving landslides is of critical significance. The Baige landslide, which caused substantial economic losses due to two major sliding events in 2018 and 2020, had undergone more than 50 years of slow deformation before the onset of the rapid sliding phase [4,5]. A novel framework for predicting landslide deformation and providing automatic early warnings, combining machine learning with physical models, was developed by Wang et al. [6]. This framework has been shown to accurately predict the entire sequence of deformation stages, from slow creeping to critical sliding. Guo et al. [7] examined the reactivation mechanisms of the Jiangdingya ancient landslide in Gansu Province, China, using remote sensing interpretation and geological surveys. Their study found that intense rainfall may have accelerated river erosion, thereby intensifying the deformation process. However, research on the transitional phases of landslide deformation, particularly under the combined influences of rainfall and human engineering activities, remains limited.

Large ancient landslides are typically located in deeply incised canyon regions, where traditional field survey methods often fail to detect slow deformation effectively. Recent advancements in landslide monitoring technologies have significantly improved the identification and analysis of these large, ancient landslides. InSAR technology, with its broad spatial coverage, high precision, and efficiency, has become a crucial tool for the early detection and continuous monitoring of landslide displacement. Kang et al. [8] identified 50 deep-seated, slow-moving landslides in the Jinsha River Basin by using displacement and topographic data from InSAR and Google Earth imagery. The slope angles of the landslide slip surfaces were inferred from the displacement fields, thereby estimating two-dimensional depth and volume. These results were validated against field data, confirming their reliability. In a study by Vařilová et al. [9], InSAR was applied to analyze landslides in the Ethiopian Highlands, concluding that both rainfall and human activities were the primary triggers of reactivation for four typical landslides. Gao et al. [10] used InSAR and LiDAR to analyze the long-term surface deformation of the Hongyanzi landslide and concluded that intense rainfall was the main controlling factor, leading to a terraced deformation pattern characterized by typical long-term, gradual, and seasonal movement. Currently, commonly employed InSAR techniques for landslide deformation monitoring include D-InSAR [11], PS-InSAR [12], and SBAS-InSAR [13]. These methods have proven effective not only for the early detection of potential landslides but also for monitoring their spatiotemporal evolution, making them indispensable tools for surface deformation analysis and hazard assessment.

Due to the tectonic uplift in the Tibetan Plateau and intense river erosion, large ancient landslides are extensively developed throughout the Jinsha River Basin. Human engineering activities and extreme rainfall events are often identified as the direct triggers for the reactivation of these landslides. However, the influence patterns of human engineering activities and the mechanisms behind rainfall lag remain poorly understood. In the complex geological context of the Jinsha River Basin, further research is needed to figure out the precise mechanisms by which various factors contribute to landslide reactivation. This study focuses on a large, deep-seated, slow-moving ancient landslide in Zhongxinrong Township, located upstream of the Jinsha River. Through a combination of field surveys, optical remote sensing interpretation, InSAR-based surface deformation monitoring, UAV LiDAR surveys, and other advanced methodologies, the relationship between deep-seated, slow-moving landslides and precipitation in the Jinsha River Basin was analyzed. Moreover, these findings will enhance the understanding of the deformation mechanisms of large, deep-seated landslides in high-mountain canyon regions under coupled rainfall and construction conditions. In addition, they will also advance quantitative analyses of rainfall lag effects, providing essential data and scientific support for large-scale engineering planning in deeply incised canyon areas. Furthermore, this has significant implications for disaster prevention and mitigation in high-mountain canyon regions during continuous rainfall, offering theoretical support for anticipating the risks of landslides, slope collapses, and other hazards under specific conditions and for taking appropriate countermeasures.

2. Study Area

2.1. Geological Background

The Zhongxinrong landslide is located on the left bank of the Renniang River in Batang County, upstream of the Jinsha River, within the southern section of the Jinsha River Fault Zone (see Figure 1b). The Jinsha River Fault Zone extends from Baiyu in the north, through Batang, and intersects with the Honghe Fault to the south. This tectonically complex belt spans approximately 700 km in length and 80 km in width, oriented nearly north-south [14]. The fault zone is primarily characterized by thrust and strike-slip faulting, exhibiting high seismic activity and frequent earthquakes [15]. Faulting and intense river incision have shaped the river valley into a typical “V”-shaped morphology, with elevations ranging from 2290 m to 5170 m, and vertical incision depths exceeding 1000 m from the mountain top to the valley bottom. Seismic activity in the region is relatively infrequent, with the most recent significant earthquake being the 2016 Batang Ms 5.1 earthquake. According to available data, the Zhongxinrong landslide is approximately 115 km from the epicenter of this earthquake and lies outside the six-degree seismic impact zone. Recent seismic events around the landslide are scarce.

The average annual precipitation in the region is 753 mm, with rainfall concentrated primarily between June and September. River runoff is predominantly derived from atmospheric precipitation. The exposed surface strata are primarily composed of Quaternary loose deposits, while the underlying bedrock consists mainly of the Upper Triassic Jiapi La Formation (T₃j) and the upper and lower sections of the Zhongxinrong Group (T_1-2zh¹, T_1-2zh²), which include greenish-gray schist, slate, phyllite, mafic volcanic rocks, and limestone, with extensive development of slate and schist. Due to the active tectonic processes in the Jinsha River Fault Zone and the intense river incision, the riverbanks are steep and narrow, and the rock masses are highly fragmented. The strata on both sides of the fault are disordered, with intense folding, and features such as joints, cleavage, and schistosity are well-developed. Furthermore, magma intrusion has caused alterations in the rock masses, and surface rocks exhibit strong weathering, resulting in poor mechanical properties of the rock masses. This geological setting has fostered the development of numerous landslides within the fault zone, including the Gonghuo landslide (see Figure 1d) and the Zhongxinrong landslide (see Figure 1e).

2.2. Development Characteristics of Zhongxinrong Landslide

Field surveys indicate that the Zhongxinrong landslide is a large-scale ancient landslide, predominantly an accumulation landslide, with an overall triangular shape in plan view. The landslide is confined by bedrock gully features on both sides. It extends 1220 m in length, 945 m in width, with a front edge elevation of 2753 m, a rear edge elevation of 3305 m, and a relative elevation difference of 552 m. The main sliding direction is 330°, with an average slope of approximately 25°, and a planar area of 9.9 × 10⁵ m². The bedrock primarily consists of slate from the Middle and Lower Triassic Zhongxinrong Group, characterized by well-developed joint fractures. It falls under the compound slide classification according to the Varnes classification system [16,17].

Field surveys also revealed significant deformation characteristics on the landslide surface, including steep scarps, cracks, and collapse induced by road cuts. The landslide is classified as a rockslide with a longitudinal slope structure. The landslide mass comprises Quaternary colluvial debris and gravels, with particle sizes ranging from 2 to 250 cm and a stone content of approximately 40–60%. The lithology is predominantly slate and phyllite.

3. Research Methods and Data

In this study, a multi-disciplinary approach integrating remote sensing interpretation, InSAR-based surface deformation monitoring, field surveys, and UAV LiDAR surveying was employed to investigate the spatial structure, surface deformation, and key controlling factors of the Zhongxinrong landslide, located upstream of the Jinsha River (see Figure 2).

3.1. Remote Sensing Interpretation and Field Survey

Building on an analysis of existing data on geological structures, stratigraphy, lithology, and geological hazards, a combination of high-resolution optical remote sensing imagery interpretation, UAV LiDAR surveying, and field surveys was employed to investigate the developmental characteristics of the Zhongxinrong landslide. Landslide identification in this area was accomplished through visual interpretation of 1.5 m-resolution Spot 6 optical remote sensing images, supplemented by field surveys to delineate the landslide boundaries. Satellite remote sensing imagery provides accurate measurements of gully length and overall distribution on a macro scale, effectively outlining the gully’s general outline, while ground-based LiDAR data excels in measuring microform indicators such as gully depth and slope, achieving a measurement accuracy of up to 0.2 m.

3.2. UAV LiDAR Surveying

Unmanned aerial vehicle (UAV) LiDAR (Light Detection and Ranging), which utilizes laser-based distance measurement and imaging, has overcome the limitations of traditional remote sensing methods constrained by light and weather conditions. UAV LiDAR can penetrate clouds, vegetation, and buildings, enabling the acquisition of detailed surface information and the production of high-precision digital elevation models (DEM) with a resolution better than 0.2 m × 0.2 m [18,19]. This technology has made significant advancements in geological hazard surveys and studies in recent years [20,21,22]. However, LiDAR methods require careful consideration of horizontal and vertical accuracy values when acquiring DEM, as these are significantly influenced by terrain and atmospheric conditions: simpler terrains and minimal atmospheric interference lead to higher accuracy, whereas complex terrain or strong atmospheric disturbances can reduce accuracy. Particularly in steep slopes and dense vegetation areas, uncertainty in surface point extraction may lead to DEM errors, thus affecting the results of landslide deformation analysis.

In this study, the D20 photogrammetry system from Feima Robotics Technology Co., Ltd. (Shenzhen, China) was adopted as the aerial platform, with the D-OP3000 oblique photography module and the DV-LiDAR10 laser LiDAR sensor employed for data collection. An oblique flight was conducted over the Zhongxinrong landslide, attaining an average point cloud density of 49 points/m². Seven laser echoes were utilized to generate high-resolution digital orthophoto maps (DOM) with an accuracy of 0.1 m × 0.1 m, and DEM with a resolution of 0.2 m × 0.2 m after filtering out surface vegetation and buildings. The high-resolution DEM terrain data obtained through UAV LiDAR enabled the generation of landslide profile maps and facilitated detailed observations of the landslide’s deformation characteristics.

3.3. Landslide Surface Deformation Monitoring Based on SBAS-InSAR

InSAR is a high-resolution imaging radar system that belongs to microwave remote sensing. It can, to some extent, penetrate clouded or rainy regions and generate SAR data via radar signals. Compared to traditional methods such as ground-based measurements and GNSS surface displacement monitoring, InSAR technology has significant advantages in terms of monitoring distance, data coverage density, and deformation recognition accuracy. The primary InSAR methods employed for monitoring and analyzing landslide deformation include D-InSAR, SBAS-InSAR, and PS-InSAR. D-InSAR, which uses differential interferometry with two sets of data, is prone to errors such as phase unwrapping issues and atmospheric delays between SAR images taken at different times [23]. Conversely, SBAS-InSAR utilizes a small baseline subset of interferometric pairs, enabling the efficient tracking of surface deformation over time [24]. This technique optimizes the use of interferometric images and provides high-precision surface displacement measurements, making it particularly effective for landslide monitoring [25,26,27].

3.3.1. SAR Data and Sources

The Sentinel-1A satellite, launched in 2014 as part of the European Space Agency’s Copernicus program (GMES), is an Earth observation satellite equipped with a C-band Synthetic Aperture Radar (SAR). Operating in TOPS imaging mode, it features a short revisit cycle of 12 days, enabling continuous, all-weather, and all-day surface observation [24,25]. The quality of data from the Sentinel-1 satellite’s ascending and descending orbits is influenced by the landslide’s slope direction (see Figure 3). Descending orbit data is more suitable for analyzing slope deformation in the western and northern directions [24]. The Zhongxinrong landslide predominantly slopes to the north. In this study, Sentinel-1A descending orbit data were selected to monitor and analyze the deformation of the Zhongxinrong landslide. The data range from November 2014 to October 2023, covering a total of 222 scenes (see

Table 1. The basic parameters of SAR data.

Parameters	Sentinel-1A
Direction	Descending
Path	33
Frame	492, 497
Band	C
Radar wavelength(cm)	5.6
Incident angle (°)	38.71
Image interval (days)	12
Time	3 November 2014 to 8 October 2023
Number of images	222

).

3.3.2. SBAS-InSAR Surface Deformation Monitoring Method

SBAS-InSAR technology acquires N + 1 SAR images captured at different times within the landslide area. Appropriate reference and secondary images are selected for spatial co-registration, and M interferometric radar images are generated. The SAR images used for analysis are sorted chronologically. Let t represent the radar satellite imaging time. After eliminating the effects of flat-earth and topographic phases, the interferometric phase is generated at pixel point (x, r) for two images captured at times t_A and t_B [26,28] (see Equation (1)):

$$\begin{array}{l}\delta {\phi }_{\mathrm{j}}(x,r)=\phi (t_B,x,r)-\phi (t_A,x,r)\\ \approx {\displaystyle \frac{4\pi }{\lambda }}\left[\Delta d_{disp}(t_{B-A},x,r)\right]+{\displaystyle \frac{4\pi }{\lambda }}\cdot {\displaystyle \frac{B_{\perp \mathrm{j}}\Delta h}{r\sin \theta }}+{\displaystyle \frac{4\pi }{\lambda }}\left[\Delta d_{atm}(t_{B-A},x,r)\right]+\Delta n_{\mathrm{j}}\\ (j=1,\dots ,M)\end{array}$$

(1)

where r represents the range coordinate, and x represents the azimuth coordinate. δφ_j(x, r) is the interferometric phase at the pixel point (x, r). φ(t_B, x, r) and φ(t_A, x, r) denote the phase values of the SAR satellite at times t_A and t_B for the pixel (x, r), respectively. j is the index of the differential interferogram sorted by time. Δh represents the topographic phase difference, and Δd_disp is the displacement phase difference. λ is the wavelength of the radar image, and Δd_atm accounts for atmospheric phase errors.

After removing the topographic and atmospheric phase errors, the Equation 1 leads to a system of M equations with N unknowns, φ(t_i, x, r) (where i = 1, …, M), which can be expressed in the form of Equation (2).

$$A\phi =\delta \phi$$

(2)

where A is an M × N matrix, and φ is the unknown vector representing the displacement phase. By transforming Equation (2) and replacing the unknown phase parameters with the average velocity v between two different time points, Equation 3 is derived:

$$Bv=\delta \phi$$

(3)

where B is an M × N matrix. It has been observed that partitioning the SAR image data of the target area into several independent small baseline subsets leads to a loss of rank in matrix B, resulting in an infinite number of solutions for Equation (3). However, in the SBAS processing workflow, the Singular Value Decomposition (SVD) method is employed to compute the pseudo-inverse of matrix B, yielding the least squares solution for Equation (3). This procedure introduces a linear assumption for the average velocity to be solved. Furthermore, prior to solving Equation (3), it is necessary to remove orbital errors, DEM errors, and atmospheric errors.

In the data processing for this study, the initial SAR image from 24 November 2014, was selected as the reference image. After data registration and interferometric processing, interferograms with poor coherence were excluded. A total of 351 interferometric pairs were obtained from descending orbit data (see Figure 4), with time baselines of less than 48 days and spatial baselines of less than 300 m. The SVD method was applied to solve for the small baselines, integrating the phase rate obtained for each interferogram into the time domain. Finally, the surface deformation velocity in the radar line-of-sight direction for the Zhongxinrong landslide was derived.

4. Results Analysis

4.1. Development Characteristics of the Zhongxinrong Landslide

4.1.1. Planar Morphological Characteristics

The landslide can be divided into two main areas in planform (see Figure 5a): the initiation zone (I) and the accumulation zone (II):

(1)

Initiation zone (I)

The initiation zone of the Zhongxinrong landslide, which has experienced large-scale sliding events in geological history, is located in the middle to upper part of the landslide body. The elevation ranges from 3314 to 3024 m, with a length of 690 m along the north-south direction and a slope range from 27 to 33°. Remote sensing interpretation and field geological surveys indicate that Cracks are observed in the road at the middle part of the landslide, with a maximum width of 0.08 m and a length of approximately 12 m (see Figure 5c). Despite these features, this area is stable.

(2)

Accumulation zone (II)

Based on field surveys [29,30], the accumulation zone can be further subdivided into two sub-regions: Accumulation Body II-1 and Accumulation Body II-2:

Accumulation Body II-1: Located on the eastern side of the Zhongxinrong landslide, Accumulation Body II-1 has a primary sliding direction of 207°, with an elevation range of 2803–3176 m and a front edge facing the Renniang River. The average slope in this region is approximately 21–27°, with multiple roads crossing through the middle to upper parts of the landslide body. In the middle section of Accumulation Body II-1, there is a slope cut engineering project resulting from road construction. Four secondary landslides, including L06 and L09, have been identified in this area, with a large portion of farmland visible on L06. The aspect of L06 is consistent with the main sliding direction of Accumulation Body II-1.

Accumulation Body II-2: Located on the lower-western side of the Zhongxinrong landslide, extending from the lower boundary of the initiation zone to the Renniang River. It has an elevation range of 2679–3024 m, with an average slope of 22–25°. Five secondary landslides have been identified in this zone (see

Table 2. Basic characteristics of secondary landslides in the Zhongxinrong landslide.

Landslide	Planar Morphology	Rear Edge Elevation/m	Elevation Difference/m	Area/m2	Thickness/m	Volume/m3
L01	Irregular	2766	87	$2.3\times$104	10	$2.3\times$105
L02	Tongue	2824	132	$9.6\times$104	12	$1.2\times$106
L03	Irregular	2760	47	$1.6\times$104	10	$1.6\times$105
L04	Irregular	2904	142	$1.3\times$105	24	$3.2\times$106
L05	Irregular	2781	40	$1.1\times$104	8	$9.1\times$104
L06	Tongue	2987	206	$1.2\times$105	21	$2.6\times$106
L07	Tongue	3024	197	$3.3\times$104	14	$4.6\times$105
L08	Tongue	2871	39	$3.8\times$103	5	$1.9\times$104
L09	Tongue	2949	95	$8.8\times$103	7	$6.1\times$104

). Field surveys reveal that the rock structure in this area is heavily fractured due to faulting, with significant surface weathering, which has facilitated the development of secondary landslides. The front edge of the landslide is also severely eroded by the river, contributing to notable sliding and collapse phenomena (see Figure 5b).

4.1.2. Spatial Development Characteristics

Based on the high-resolution UAV LiDAR-derived DEM, with an accuracy of 0.2 × 0.2 m, engineering geological plan and profile maps of the Zhongxinrong landslide were drawn (see Figure 6). As shown in Figure 1b, the Zhongxinrong landslide is located within the Jinshajiang Fault Zone, a region characterized by significant tectonic activity that has caused extensive rock fracturing, complex jointing, and fissuring. The resulting weathering processes have facilitated the accumulation of residual slope deposits on the surface. The bedrock is primarily composed of Triassic shale. Landslide movement has induced extensional fractures in the bedrock, particularly at the interface of the sliding surface, some of which have been weathered.

Through ground topographic analysis and engineering geological characteristic analysis, in conjunction with field surveys, the engineering geological profile of the landslide was plotted. The average thickness of the landslide is estimated to be approximately 15.0 to 35.0 m, with a total volume ranging from 1.48 × 10⁷ to 3.46 × 10⁷ m³, classifying it as a large-scale landslide. Specifically, the area of II-1 is approximately 4.5 × 10⁵ m², with an average thickness ranging from 20.0 to 35.0 m, and a volume ranging from 0.9 × 10⁷ to 1.6 × 10⁷ m³. The area of II-2 is approximately 4.0 × 10⁵ m², with an average thickness of 15.0 to 25.0 m, and a volume ranging from 0.6 × 10⁷ to 1.0 × 10⁷ m³.

4.2. Surface Deformation Characteristics of Zhongxinrong Landslide

4.2.1. Overall Deformation Characteristics

Based on the radar line-of-sight (LOS) deformation results for the Zhongxinrong landslide from November 2014 to August 2023 (see Figure 7), combined with remote sensing image analysis, it was found that the surface vegetation coverage of the landslide is relatively sparse, resulting in a high density of coherent points and consequently good interferometric results. Notable accumulation of deformation is observed at the front edge of the landslide, where the maximum deformation velocity reached −65.65 mm/a, which indicate that the Zhongxinrong landslide belongs to the very slow landslide [16,17]. Negative values of V_LOS indicate motion away from the satellite, while positive values indicate motion towards it. The deformation is more pronounced in the upper and middle-lower sections of the landslide, particularly in secondary landslides L02 and L04, as well as in human activity zones near Zhongxinrong Township and the middle-back sections of the landslide.

The high deformation zones are primarily distributed in the upper section of Accumulation Body II-1 and throughout Accumulation Body II-2, with the deformation velocity in II-2 being significantly higher than that in II-1 (see Figure 7b). Remote sensing analysis indicates that the deformation is concentrated along areas with dense road networks, with eight roads intersecting along profile A-A’ and five roads along profile B-B’ (see Figure 8). Analysis of these profiles suggests that areas with more roads also exhibit greater deformation velocity. It can be inferred that the deformation factors differ between the two accumulation bodies:

(1)

Deformation Characteristics of Accumulation Body II-1:

The strong deformation zone in Accumulation Body II-1 is primarily concentrated in the upper section. Remote sensing images show that the corresponding area is highly weathered, with fractures developed in the bedrock due to intense weathering. These fractures gradually propagate under the influence of rainfall infiltration, reducing the shear strength of the rock. Multiple roads pass through the area, and the deformation velocity near monitoring point Z4 is higher than that near Z5 and Z6. This is due to the presence of two wide roads near Z4, where vehicle loading and vibrations significantly impact the slope. A collapse (see Figure 9a) and multiple steep steps (see Figure 9b) are also present in the high deformation region of Accumulation Body II-1, indicating that the body predominantly exhibits high-elevation deformation characteristics. The landslide mass is mainly composed of loose Quaternary sediments, predominantly consisting of unstructured coarse rock fragments and soil. In the central portion of the landslide, near a road, the sliding surface is exposed, revealing clear scratches (see Figure 9f).

(2)

Deformation Characteristics of Accumulation body II-2:

The overall deformation in Accumulation Body II-2 is more pronounced. Remote sensing analysis reveals that this region contains a significant amount of farmland, and extensive groundwater irrigation has led to an elevated groundwater table, softening the soil and reducing its strength. Field surveys found that the area features steep steps, and the rock mass beside the roads is severely cracked, with clear signs of deformation (see Figure 9c,d). Additionally, the erosion of the slope’s toe by the Renniang River has induced strong traction, causing further deformation. The accumulation body features multiple terraces, which facilitate the collection of rainwater. Prolonged rainfall infiltration has kept the slope in a continuous state of deformation, with the deformation velocity in this area being significantly higher than that in Accumulation Body II-1.

Therefore, the Zhongxinrong landslide is experiencing slow deformation, with an overall maximum deformation rate of 65.65 mm/yr. High deformation rates are primarily concentrated in areas with human activity and dense road networks.

4.2.2. Time-Series Deformation Characteristics

Time-series analysis of the Zhongxinrong landslide was conducted using SBAS-InSAR technology with descending orbit data (see Figure 10). The surface deformation of the landslide is primarily concentrated in the two accumulation zones. In Accumulation Body II-1, deformation is mainly located in the upper section, with the maximum cumulative deformation reaching 464.21 mm. In Accumulation Body II-2, the maximum cumulative deformation in the central section is 600.18 mm.

From October 2015, when monitoring began, to October 2018, the cumulative deformation over the first three years was 193.53 mm, with an average annual growth rate of 64.51 mm/a. From October 2018 to October 2021, the cumulative deformation over three years was 315.27 mm, with an average annual growth rate of 105.09 mm/a, which is 1.63 times that of the previous three years. From October 2021 to October 2023, the cumulative deformation over two years was 52.86 mm, with an average annual growth rate of 26.43 mm/a. The annual growth rate gradually decreased, indicating that the landslide had entered a creeping deformation phase.

Six typical monitoring points were selected on the Zhongxinrong landslide to analyze the cumulative deformation trends at these points (see Figure 11). The deformation at Z5 and Z6, located in Accumulation Body II-1, was relatively stable, with the maximum deformation values being −32.2 cm. The deformation at Z4 was moderate, with a value of −38.6 cm. At the foot of the slope, in the area of strong deformation, the deformation at Z1, Z2, and Z3 was relatively intense. Starting from July 2016, the deformation at these three points began to accelerate. By November 2023, the cumulative deformation at these points had reached −53.5 cm, −54.4 cm, and −48.8 cm, respectively.

The deformation at the Zhongxinrong landslide was found to be significantly correlated with rainfall, showing a stepwise growth pattern. Further analysis of rainfall and cumulative deformation revealed that from July 2016, the landslide experienced five phases of accelerated deformation, with the maximum rate reaching 250.03 mm/a. The deformation velocity in Phase ② was significantly higher than that in Phase ④ compared to the previous accelerated deformation phase. Although the rainfall peak in Phase ② was not particularly high, the number of days with daily rainfall exceeding 15 mm was greater and more concentrated than in the previous phase. In Phase ④, the daily rainfall peak exceeded 30 mm for the first time, leading to a significant increase in the average deformation rate during this phase.

5. Discussion

5.1. Discussion on the Mechanism of Rainfall Lag in Large Deep-Seated Slow-Moving Landslides

The influence of rainfall on slow-moving landslides is governed by several factors, including the increase in pore water pressure, the reduction in shear strength along the slip surface, the depth of the sliding surface, and the permeability and diffusion coefficients of the materials [31]. As a result, landslides of varying volumes and depths exhibit different responses to rainfall [32,33]. Short, intense rainfall typically triggers shallow landslides, whereas deep-seated landslides are more commonly triggered by prolonged, low-intensity rainfall, which leads to large-scale bedding plane slides and debris slides. In these cases, the slip surface is often located within deeper, weaker layers [34].

Huang et al. [35], using borehole inclinometer data, found that displacement in the Jinshanlong large deep-seated landslide in the Yalong River Basin was closely linked to seasonal rainfall. The sliding surface depth of this landslide was approximately 25–30 m, and the maximum monthly displacement lagged behind the peak monthly rainfall by 12 months. It was noted that most of the rainfall during the June–July rainy season was absorbed by the dry unsaturated zone. By September, the soil reached saturation, and rainwater continued to percolate downward [36], weakening the shear strength of the sliding surface. Moreover, the clay layers in the slip zone had low permeability, and it took time for the clay to become fully saturated, resulting in a 1–2 months delay in the deformation response to rainfall. Hilley et al. [32] used InSAR technology to monitor the deformation of a slow-moving landslide in the East Bay of San Francisco from 1992 to 2001, observing a 3-month lag between the onset of rainfall and the acceleration of landslide deformation.

In this study, daily rainfall data from October 2014 to October 2023 were collected to examine the correlation between accelerated deformation and cumulative rainfall over 15-day periods at typical monitoring points, Z2 in Accumulation Body II-2 and Z4 in Accumulation Body II-1 of the Zhongxinrong landslide (see Figure 12). The results showed that the Zhongxinrong landslide exhibits a clear rainfall lag in its deformation response. When the cumulative rainfall over a 15-day period exceeded 120 mm (15.93% of the annual average rainfall), the deformation rate increased with a lag of 2 to 28 days. The main composition of the Zhongxinrong landslide accumulation body is slate, which has low permeability, leading to very slow water infiltration. The soil requires a significant amount of time to reach saturation, which in turn reaches the critical condition for slope instability. Since the Zhongxinrong landslide consists of two accumulation bodies with different deformation patterns, the lag time varied between them. The higher-elevation accumulation body experienced deformation acceleration later than the traction-type accumulation body, with a time difference of 0 to 14 days. This time difference was influenced by the rainfall amounts on the preceding and following days. On July 24, 2021, a daily rainfall of 40.92 mm, significantly higher than on other days, led to a consistent deformation acceleration response in both accumulation bodies.

Additionally, by analyzing the stepwise characteristic of the cumulative deformation curve shown in Figure 10a, it was found that all five instances of lagged acceleration in deformation were associated with stepwise deformation. These instances of lagged acceleration occurred during Phases ①, ④, and ⑤.

5.2. Partitioned Deformation Mechanisms of Large Ancient Landslides Under Multi-Factor Influences

The deformation mechanisms of large deep-seated ancient landslides have been extensively studied by various scholars. For instance, Cui et al. [37] analyzed the Mahu ancient landslide group in Leibo County, Sichuan, which spans an area of 18.2 × 10⁴ m² and has a volume of approximately 20 × 10⁸ m³. Triggered by multiple earthquakes, this landslide group is characterized by large, high-speed, long-distance slides along steep interfaces between strata, reflecting the unique formation characteristics of ancient landslides in specific geological environments. Similarly, Huang et al. [38] studied the Dalixi landslide in the Three Gorges Reservoir Area and concluded that it is a traction-type landslide primarily triggered by engineering excavation, with rainfall serving as the activating factor. The deformation in this landslide progresses gradually along its longitudinal axis. Li et al. [39] examined landslides along the Longwu River in the Tibetan Plateau, identifying them as tensile deformations induced by erosion, which were reactivated due to rainfall and engineering activities.

In contrast to the typical deformation characteristics observed in ancient landslides, the Zhongxinrong landslide analyzed in this study exhibits the simultaneous occurrence of high-elevation deformation and tensile deformation. The Zhongxinrong landslide is classified as an accumulation landslide, and the deformation mechanisms must be examined separately for its two distinct accumulation bodies. Accumulation Body II-1, located at a higher elevation, is influenced by rainfall, bedrock weathering, and human engineering activities, whereas Accumulation Body II-2 is a tensile accumulation body impacted by groundwater, rainfall, and river erosion.

Human activities, such as construction projects, have altered the stress distribution and hydrological conditions in the landslide area, resulting in a deformation pattern distinct from the traditional ancient landslide deformation mechanisms driven by natural factors. This has led to the unique co-occurrence of high-elevation deformation and traction deformation in the Zhongxinrong landslide. This finding emphasizes the need to focus not only on areas where deformation has already occurred but also on how engineering activities, even in areas with less apparent deformation, can exacerbate overall landslide deformation. For the Zhongxinrong landslide, Accumulation Body II-1 developed after the formation and rapid deformation of Accumulation Body II-2, with influences from road construction and other human factors. These two accumulation bodies have a mutual influence on each other during deformation, with the tensile deformation in Accumulation Body II-2 causing cracks at the lower part of Accumulation Body II-1 and promoting its deformation.

Understanding the impact of human engineering activities and other anthropogenic factors is crucial for providing a comprehensive understanding of the complex deformation mechanisms in ancient landslides, ultimately offering more accurate guidance for their prevention and early warning systems.

The Zhongxinrong landslide is located in the Jinsha River Fault Zone (see Figure 1b), where several large deep-seated slow-moving landslides have developed. The evolution of these landslides is strongly influenced by tectonic uplift, fault activities, and periodic seismic events, which lead to the development of joint fractures and the fragmentation of rock structures. These geological conditions provide substantial internal driving forces and abundant material sources for landslide formation (see Figure 13a). Concurrently, long-term surface weathering, river erosion, and other processes have caused the slope in the source area to crack and deform. Once the deformation reaches a critical threshold, the slope becomes unstable and begins to slide. After the initial landslide, seismic events, rainfall, and river erosion continue to fragment the rock mass, causing surface cracks to expand and facilitating the infiltration of rainwater. This infiltration increases pore water pressure within the slope, further destabilizing it. Additionally, river erosion at the foot of the landslide increases the sliding force in this area, promoting slow deformation of the surface layers (see Figure 13b). In the high-elevation sections of the landslide, where bedrock is exposed and weathering is more severe, deformation begins to occur under the continuous infiltration of rainfall (see Figure 13c). As the landslide progresses, human activities, such as road construction and vehicle traffic, disturb the high-elevation slopes, exacerbating deformation and increasing the risk of instability and sliding (see Figure 13d).

The formation and evolution of large landslides in the Jinsha River Basin, a typical high-mountain canyon area, involve highly complex deformation mechanisms. Based on the research conducted, the formation and evolution of large deep-seated landslides in the Jinshajiang active tectonic zone can be summarized in four stages:

(1)

Complex regional tectonic activities provide substantial internal driving forces for landslide formation.

(2)

Instability in the landslide source area occurs due to the combined effects of internal and external forces.

(3)

Earthquakes, rainfall, surface weathering, and river erosion lead to slope sliding, blockage, and river diversion in the source area.

(4)

Human engineering activities, such as road construction and settlement development, combined with rainfall infiltration, exacerbate deformation in high-elevation areas, increasing the risk of instability and sliding.

6. Conclusions

This study investigates the deformation mechanisms and key influencing factors of large deep-seated ancient landslides on the Tibetan Plateau, using the Zhongxinrong landslide in the upper reaches of the Jinsha River as a case study. By employing SBAS-InSAR time-series analysis, UAV LiDAR, and field surveys, we analyzed the formation factors and slow deformation patterns of these landslides in the Jinsha River basin. The main conclusions drawn from this study are as follows:

First, a large ancient compound landslide was identified in the Batang section of the upper Jinsha River, named the Zhongxinrong landslide. The landslide stretches approximately 1220 m in length, with a vertical elevation difference of about 552 m. The area is approximately 9.9 × 10⁵ m², with an average thickness ranging from 15.0 to 35.0 m, and a volume between 1.48 × 10⁷ m³ and 3.46 × 10⁷ m³.

Second, the Zhongxinrong landslide exhibits creeping deformation. According to SBAS-InSAR descent deformation rate calculations, from 2014 to 2023, the maximum surface deformation rate reached −65.65 mm/a, with the maximum cumulative deformation of −600.18 mm, and characterized as the very slow landslide. The most significant deformation occurs in the upper part of the landslide, particularly in the western side of Accumulation Body II-1, in areas with secondary slip zones L02 and L04. Deformation is also notable near Zhongxinrong Township and the middle and rear sections of the landslide, which are classified as strong to extremely strong deformation zones.

Third, the Zhongxinrong landslide demonstrates a dual-accumulation deformation pattern in Accumulation Body II-1 and II-2. Deformation in Accumulation Body II-1 is predominantly high-elevation deformation, driven by rainfall and human activities such as road construction. In contrast, Accumulation Body II-2, located at the landslide’s front edge, experiences tensile deformation due to the combined effects of rainfall and river erosion.

Fourth, this study reveals a strong correlation between rainfall and landslide deformation. Following heavy rainfall events, landslide deformation intensifies, with a noticeable lag and stepwise pattern in response. When the 15-day cumulative rainfall exceeds 120 mm, the deformation rate increases with a lag of 2 to 28 days. Notably, the lag response for high-elevation deformation bodies is slower than for the tensile deformation in Accumulation Body II-2. A peak daily rainfall of 40.92 mm resulted in synchronized lag response times across both accumulation bodies. This suggests that prolonged rainfall significantly weakens the shear strength of the landslide’s shear zone, thereby triggering large-scale deformation of deep-seated landslides.

Last, this study has clarified the basic developmental characteristics and deformation mechanisms of the Zhongxinrong landslide using InSAR, UAV LiDAR, field surveys, and other methods. It proposes a deformation model for large deep-seated landslides in high-mountain canyon regions under the coupling of human engineering activities and rainfall, revealing the phenomenon of rainfall lag under specific conditions. The presence of residents in this landslide area necessitates appropriate monitoring to enable collective risk prevention and mitigation. With ongoing climate change and continuous technological advancements, the deformation mechanisms of landslides are expected to become increasingly complex. However, the introduction of new methods and technologies will likely lead to further progress in landslide mechanism research and risk prevention. The findings of this study provide a practical case for analyzing landslide deformation mechanisms under future climate change scenarios, and the rainfall lag model based on InSAR remote sensing offers valuable reference for subsequent research in this field. This research has important implications for related studies. However, as this study did not explore the sliding surface depth, landslide stability, and other aspects in depth, we recommend that future research combine remote sensing interpretation analysis with geophysical surveys, drilling, and numerical simulations to provide more accurate scientific data and theoretical support for landslide stability analysis and disaster prevention.