Spatiotemporal Landslide Monitoring in Complex Environments Using Radiative Transfer Model and SBAS-InSAR Technology

Abstract

Landslides are among the most frequent geological hazards, often resulting in casualties and economic losses, particularly in alpine valley areas characterized by complex topography and dense vegetation. Landslides in these regions are distinguished by their high altitude, concealment, and sudden onset, which render traditional monitoring methods inefficient. This study proposes a landslide monitoring method for complex environments that leverages multi-source remote sensing data, incorporating the radiative transfer model and Small Baseline Subset-Interferometric Synthetic Aperture Radar (SBAS-InSAR) technology. The proposed method was implemented to monitor the instability of the Baige landslide in Tibet, China. The results show that the vegetation Canopy Water Content (CWC) estimated using the radiative transfer model indirectly reflects landslide susceptibility. Specifically, excessive soil moisture from rainfall reduces oxygen in plant roots, affecting growth and lowering canopy water content. The region with lower Canopy Water Content (CWC < 0.04) exhibited an increasing trend in the number of pixels, rising from 271 to 549 before the landslide event, indicating poorer vegetation conditions in the area. Additionally, the SBAS-InSAR technique was utilized to extract surface displacement, achieving a maximum displacement of 112 mm during the monitoring period. Ultimately, the spatial changes of the two monitoring signals exhibited a high consistency. This study enhances the reliability of landslide displacement monitoring in complex environments and provides substantial scientific support for future large-scale monitoring efforts.

1. Introduction

A landslide is a phenomenon in which a slope’s geotechnical body slides downwards or is scattered along a penetrating damaged surface under the action of factors such as the natural environment and human activities [1,2]. Landslide hazards are among the most frequent, widely distributed, and damaging geological hazards globally [3,4]. Each year, landslides result in economic losses exceeding hundreds of millions of dollars and many deaths and injuries worldwide [5], posing significant risks to human life, property, and national infrastructure security. Consequently, monitoring landslide hazards has become a fundamental requirement and a critical task in disaster risk reduction [6,7].

In recent years, landslide hazards have frequently occurred in the mountainous regions of southwest China, characterized by steep valley slopes and dense vegetation [8,9]. For instance, the Liangshui Village landslide in Zhenxiong County, Yunnan, in January 2024 [10], the Jichang landslide in Shuicheng County, Guizhou, in July 2019 [11], and the Sanxi Village landslide in Dujiangyan, Sichuan, in July 2013 [12], which caused significant loss of life, destruction of infrastructure, and displacement of local populations, were found to have the common characteristics of “high altitude, high vegetation coverage, and high concealment”, rendering field surveys, optical remote sensing, and drone photogrammetry challenging in these regions [13]. Statistics indicate that 80% of landslide hazards occur outside the identified scope of potential hazards, and 70% occur in southwestern mountainous regions of China that are challenging to access via field surveys [14,15]. Thus, effectively identifying the location of landslides is a crucial research direction for monitoring and prevention in the alpine valley area of Southwest China, where the environment is complex and highly concealed.

Remote sensing technology provides strong support for geological disaster risk warning and management due to the characteristics of long-distance, non-contact, and multi-dimensional macroscopic perspectives [16,17]. Among them, the Interferometric Synthetic Aperture Radar (InSAR) technology in active microwave remote sensing [18,19] has the advantage of obtaining continuous coverage of ground elevation over a large area, all day and all weather, accurately extracting small deformation information on the Earth’s surface, and the measurement accuracy can reach millimeter level, which has been widely applied in related fields such as surface deformation monitoring and disaster assessment [20,21,22,23,24,25]. Since 1995, the Differential Synthetic Aperture Radar Interferometry (D-InSAR) technique has been used for monitoring landslides, such as the La Clapière landslide in the French Alps [26]. Ferretti et al. [27] used Permanent Scatterer Synthetic Aperture Radar Interferometry (PS-InSAR) technology to monitor landslides in the Ancona region and confirmed that the deformation monitoring accuracy of Permanent Scatterer (PS) points can reach up to 1 mm. Motagh et al. [28] applied the Small Baseline Subset (SBAS) technique for large-scale landslide identification and monitoring in the Uzgen region of southwestern Kyrgyzstan. However, in the southwestern mountainous areas with deep valleys, large undulations, and high coverage, the inherent limitations of satellite side-view imaging geometry and the attenuation of microwave signals caused by vegetation canopies make InSAR generate unexpected results with geometric distortion and spatiotemporal incoherence constraints and phase disentanglement, which makes landslide deformation monitoring difficult [29]. Therefore, there are still essential challenges in accurately monitoring landslide disasters in complex environments [30].

Vegetation, as the main surface element of remote sensing observation, plays a crucial role in landslide research [31,32,33]. In addition to the direct surface displacement monitoring mentioned above, a few scholars have also studied landslides using vegetation with the help of optical remote sensing [34,35]. For example, Saito et al. [36] used multi-temporal high-resolution satellite images and unmanned aerial drone terrain data to explore the recovery of vegetation after the Aso volcano-induced landslides in Japan by using Normalized Difference Vegetation Index (NDVI) and ratio and found that the NDVI ratio decreased significantly after the occurrence of the landslides, and then it reached the same level of vegetation as that before the landslides 12 years later. Wang et al. [37] proposed a novel research perspective, noting a significant spatiotemporal correlation between vegetation anomalies and landslide displacement during the landslide creep stage. This study indirectly reflects the evolutionary process of gradual landslide destabilization, providing a new framework for identifying potential landslide hazards in areas with dense vegetation cover. By utilizing subtle indicators, such as vegetation changes around the slope before a landslide, to indicate landslide creep, the limitations of existing methods in complex environments are addressed, thereby effectively conserving manpower and material resources. However, it is important to note that while various factors contribute to abnormal vegetation changes, not all of them are associated with landslides. Factors such as pests, diseases, deforestation, and land use changes can also play a significant role.

This study develops a spatiotemporal monitoring method for landslides in complex environments utilizing the radiative transfer model and SBAS-InSAR technology. The process is applied to monitor the Baige landslide in Tibet, China, employing multi-source data, including multi-temporal optical remote sensing data, Synthetic Aperture Radar (SAR) data, and Digital Elevation Model (DEM), to derive Canopy Water Content (CWC) and surface displacement information for mutual comparative validation. The method analyzes the spatiotemporal evolution of landslide hazards and facilitates effective identification and monitoring of large-scale landslide events in complex areas. Landslides represent a complex process involving the interaction of multidisciplinary factors, and they cannot rely solely on a single technology to investigate and monitor landslide events. It is essential to effectively monitor landslides in the southwestern mountainous regions of complex environments through the integration of various technological approaches at different levels of comparative analysis [3,8].

2. Study Area

The Baige landslide occurred at 7:00 a.m. (GMT + 8) on October 11, 2018, obstructing the mainstream of the Jinsha River and forming a barrier lake (Figure 1e). The Baige landslide is situated in the upper reaches of the Jinsha River in Baige Village, Jiangda County, Changdu City, Tibet Autonomous Region, bordering Baiyu County, Ganzi Prefecture, Sichuan Province, with the coordinates of 98°42′17.98″ E, 31°04′56.41″ N, as shown in Figure 1a,b. As the Tibetan Plateau continues to uplift, the Jinsha River is progressively downcutting, resulting in a ‘V’-shaped, tectonically eroded alpine canyon characterized by steep slopes and significant elevation differences [38], with an elevation range of 2775 m to 5285 m (Figure 1d). The study area features a cold-temperate semi-humid climate, with an average annual precipitation of 627 mm and a maximum annual precipitation of 919 mm recorded in recent years. The ecological environment within the study area exhibits marked heterogeneity, shaped by both climatic and topographic factors. Vegetation types vary significantly with elevation and slope orientation, comprising alpine shrublands and meadows typical of the high-altitude Tibetan Plateau, as well as subtropical montane coniferous and broad-leaved forests at lower elevations [39]. As shown in Figure 1c, the regional geological setting is complex, primarily comprising a series of NW-trend folds and faults. The predominant base rocks in this area include phyllite, argillaceous slate, rock fragments, sandstone, marble, and gneiss. Quaternary loose accumulation bodies, resulting from collapses, floods, and landslide activity, are distributed at the base of the valley slope [40].

3. Materials and Methods

In the southwestern mountainous areas with complex geomorphological elements, a single technical method may be limited and restricted in its application. In this study, a novel landslide monitoring method combining the radiative transfer model with the SBAS-InSAR technique was developed. As shown in Figure 2, the study area was first divided into slope structural units, and the Baige landslide was extracted as a forward slope prone to landslide. Then, the Canopy Water Content (CWC) was estimated using the radiative transfer model. Finally, the surface displacement was obtained by the SBAS-InSAR technique. The three results were compared and verified with each other to obtain the landslide monitoring method for complex environments.

3.1. Data

3.1.1. Satellite Remote Sensing Data

This study utilized Sentinel-2 satellite images to acquire vegetation information, with data from the European Space Agency Copernicus Data Center (https://dataspace.copernicus.eu/, accessed on 4 February 2024). The basic parameters are presented in

Table 1. Basic parameters of sentinel images for landslide monitoring.

Image Parameter	Sentinel-2 for Monitoring Vegetation Changes	Sentinel-1 for Monitoring Landslide Displacement
Wave Band	Band1~7, Band8A, Band8(NIR), Band9~12	C
Imaging Mode	MSI	IW
Data Products	L1C, L2A	SLC
Polarisation Pattern	/	VV
Spatial Resolution/m	10, 20, 60	5 × 20
Revisit Cycle/d	10	12
Image Time	30 July 2016, 4 August 2017, 5 June 2018, 25 July 2018	2016–2018

. Sentinel-2 is a medium-resolution imaging satellite equipped with a Multi-Spectral Imager (MSI), operating at an altitude of 786 km and covering 13 spectral bands with a swath width of 290 km. The satellite captures different spatial resolutions across the visible, Near-Infrared (NIR), and Short-Wave Infrared (SWIR) spectral ranges. Sentinel-2 data uniquely include three bands in the red-edge range, making it highly effective for monitoring vegetation health information. In terms of temporal phase, four images of July 2016, August 2017, June 2018, and July 2018 were selected to minimize the interference of seasonal and natural growth factors on the abnormal features of vegetation.

Based on the coverage of SAR imagery in the study area, Sentinel-1 data were employed for monitoring and analyzing landslide time series displacement, with data sourced from the European Space Agency Copernicus Data Center (https://dataspace.copernicus.eu/, accessed on 4 February 2024). The basic parameters are presented in Table 1. Sentinel-1 is an Earth observation satellite launched by the European Space Agency, and the C-band SAR on board uses Terrain Observation by Progressive Scans (TOPS) mode imaging. It offers four imaging modes and three data products, with polarization divided into single- and dual-polarization imaging. Sentinel-1 possesses cloud penetration capability, enabling day and night imaging. It is less affected by clouds and rain, allowing for continuous observation of the Earth, which is advantageous for scientific research and application analysis. The SAR data utilized in this study spans from 17 March 2016 to 5 October 2018, encompassing 32 scenes of IW data, with a coverage range of 250 km.

3.1.2. Topographical and Geological Data

Topographical and geological data are utilized for the Classification of slope types. The Advanced Land Observing Satellite (ALOS) satellite launched by Japan in 2006 was selected to acquire ALOS 12.5 m Digital Elevation Model (DEM) data from the National Aeronautics and Space Administration (NASA, https://search.asf.alaska.edu, accessed on 4 February 2024). This satellite provides higher accuracy compared to other DEM datasets, such as the 30m SRTM DEM, and extracts elevation, slope, and slope direction information, making it widely applicable in various natural hazard surveys and mapping studies.

Geological data for the study area, including stratigraphic lithology, geological structure, and attitude, were primarily sourced from the National Geological Data Centre (https://www.ngac.cn/125cms/c/qggnew/index.htm, accessed on 4 February 2024) and the national 1:200,000 geological map of Dege (H-47-03). This data provides a preliminary reference for geological risk assessment and disaster prevention and control in the Baige landslide and surrounding areas.

3.1.3. Precipitation Data

Rainfall, especially sudden extreme weather, may cause softening of geotechnical parameters, reducing the shear strength of geotechnical bodies while increasing the sliding force and sliding moment, causing slope destabilization and damage, and forming geological hazards such as landslides. In this study, the month-by-month precipitation data of meteorological stations in the study area from 2016 to 2018 were obtained through the China Meteorological Data Network (http://data.cma.cn, accessed on 16 May 2024).

3.2. Classification of Slope Types

Slope structure determines the type and intensity of landslides and is influenced by the stratigraphic attitude and slope orientation, which explains the topographic and geological environment boundary features of landslides [42]. Due to different geological environments, the distribution of slope types varies from region to region. Therefore, the relationship between the combination of rock inclination (geological conditions) and slope orientation (topographic and geomorphological features) is introduced to classify slope structure types to provide a basis for the subsequent identification of landslide hazards. The slope structure type is determined based on the relationship between the angle of the inclination of rock strata and the slope direction, combined with the ‘Technical Requirements for Geological Hazard Risk Investigation and Evaluation issued by the China Geological Survey. When the angle between the inclination of rock strata and the direction of the slope is less than 30°, the slope is a dip slope; when the angle between the inclination of rock strata and the direction of the slope is between 30° and 60°, the slope is downslope; when the angle between the inclination of rock strata and the direction of the slope is between 60° and 120°, the slope is transverse slope; when the angle between the inclination of rock strata and the direction of the slope is between 120° and 150°, the slope is reverse slope; when the angle between the inclination of rock strata and the direction of slope is between 150° and 180°, the slope is counter-tilt slope. The distribution of slope types varies regionally based on distinct geological settings. Dip slopes are susceptible to large rocky landslides and accretionary layer movements, while reverse slopes are prone to accretionary layer slides; sloping and transverse slopes infrequently experience landslides [43].

3.3. Estimation of CWC Based on the Radiative Transfer Model

3.3.1. Landslide Monitoring Indicator

Vegetation is the natural “link” connecting land cover elements, including soil, atmosphere, and water. Its temporal variations reflect the dynamic changes within the regional ecological environment [44]. During the developmental stage of landslides, the vegetation on the slopes also experiences abnormal changes, such as wilting and death [45]. Consequently, vegetation change in remote sensing images is an important indicator for monitoring and quantitatively evaluating landslide activity. This study can effectively assess vegetation growth and soil environmental conditions through the abnormalities in vegetation Canopy Water Content (CWC) and observe the correlation between its changes and landslide creep.

CWC serves as a valuable indicator for landslide monitoring, characterizing the ecological water storage within the vegetation canopy. CWC refers to the amount of water contained in the leaves of vegetation canopy per unit surface area [46]. Ceccato [47] proposed a method for calculating the CWC of the vegetation in the global study of the Vegetation Water Index (VWI) as demonstrated in Formula (1):

$$\alpha =\beta \times \gamma$$

(1)

where α is CWC, β represents Equivalent Water Thickness (EWT), which is the amount of water contained in a plant leaf, and γ represents the Leaf Area Index (LAI), which is the ratio of the total leaf area to the unit ground area in a plant canopy.

3.3.2. PROSAIL Model

The PROSAIL model is a widely utilized vegetation radiative transfer model that simulates spectral characteristics and radiative reflectance of vegetation, establishing a clear physical relationship between vegetation biophysical parameters and remote sensing observations [48]. The model is a coupling of the PROSPECT leaf model and the SAILH canopy structure model. It reflects the water storage of the canopy as a whole rather than the properties of a single leaf, explains the spectral reflectance of the canopy, and expresses the optical properties of the vegetation canopy from the perspective of physiological traits [49].

The PROSPECT model [50] characterizes the optical properties of the plant leaf blade at the leaf scale across the 400 to 2500 nm range, utilizing parameters such as leaf structure (N), chlorophyll content (Cab), carotenoid content (Car), brown pigment content (Cbq), EWT, and dry matter content (Cm). The SAILH model [51] is an enhancement of the Suits model, using the leaf reflectance (${\rho }_i$) and transmittance (${\tau }_i$) output from the PROSPECT model, combined with the LAI, Average leaf inclination angle (ALIA), soil moisture ratio (Psoil), hot spot factor (Hspot), solar zenith angle (${\theta }_S$), and observed azimuth angle (${\theta }_V$) which are input into the SAILH model [52].

3.3.3. Sobol Global Sensitivity

The PROSAIL model has a large number of input parameters, and when using PROSAIL to simulate reflectance, its input parameters are mainly determined by the canopy biophysical data in the existing database. To better determine the type and number of parameters, the sensitivity of the model input parameters needs to be analyzed first when constructing the model, which can reduce the amount of arithmetic in the calculation process, improve the accuracy of the model inversion, and optimize the model building.

Sobol global sensitivity analysis is an application of the Monte Carlo sampling method, which effectively explores the influence of parameters on each other on model simulation results [53]. Sobol [54] approximates the decomposition of the model as a function of a single input parameter or a combination of multiple parameters with each other and then calculates the size of the contribution of each parameter to the model with the following Formula (2):

$$y=f\left(x\right)=f(x_1,x_2,\dots ,x_n)(x=x_1,x_2,\dots ,x_n)$$

(2)

where $x_n$ is the parameter vector and n is the number of parameters. The variance decomposition formula is expressed as

$$V={\sum }_{i=1}^n{V}_i+{\sum }_{i<j}^n{V}_{i,j}+V_{\mathrm{1,2},\dots ,n}$$

(3)

$$S_i=V_i/V$$

(4)

where $V$ is the total variance of the model, $V_i$ is the variance term for the action of the $i^{th}$ parameter $V_{i,j}$ is the variance term for the joint action of the $i^{th}$ and $j^{th}$ parameter, $V_{\mathrm{1,2},\dots ,n}$ is the variance term for the joint action of all the parameters, and $S_i$ is the sensitivity index of the ith parameter.

3.4. Acquisition of Surface Displacement Based on SBAS-InSAR Technology

The SBAS-InSAR technique [55,56] is a time-series InSAR technique based on the small baseline computation strategy of multi-reference imagery. It is usually considered the most effective detection method to identify surface displacement in complex mountainous areas. SBAS-InSAR technology uses the temporal and spatial baseline threshold method to combine all images to obtain a scene interferometric phase. The interferometric phase corresponding to the interferogram generated by each interferometric pair can be expressed as [57]

$$∆{\phi }_j\left(x,y\right)=∆{\phi }_B\left(x,y\right)-∆{\phi }_A\left(x,y\right)\approx {\displaystyle \frac{4\pi }{\lambda }}[d\left(T_B,x,y\right)-d\left(T_A,x,y\right)]$$

(5)

where $∆{\phi }_j\left(x,y\right)$ is the differential interferometry phase, $∆{\phi }_B\left(x,y\right)$ is the primary image phase, $∆{\phi }_A\left(x,y\right)$ is the secondary image phase, $T_B$ is the data acquisition time of the primary image, $T_A$ is the data acquisition time of the secondary image, λ is the radar wavelength, $x$ is the azimuth coordinate, and $y$ is the line-of-sight coordinate.

Subsequently, a series of M equations with N unknowns can be established as the following matrix [58]:

$$\left\{\begin{array}{c}\phi ={[\phi \left(t_1\right),\dots ,\phi \left(t_N\right)]}^T\\ \Delta \phi ={\left[\Delta \phi \left(t_1\right),\dots ,\Delta \phi \left(t_N\right)\right]}^T\\ A\phi =\Delta \phi \end{array}\right.$$

(6)

where φ is the N × 1 vector of unknown displacement phase values with measurement points, Δφ is the N × 1 vector of unwrapped phase values, and A represents the M × N coefficient matrix.

Based on constructing the displacement phase into a system of equations, the algorithm of singular value decomposition is used to obtain the linear displacement rate and elevation error phase in the study area. The residual phase in the interfering phase is separated and disentangled quadratically, and the atmospheric phase and noise phase are estimated and removed by spatial domain filtering. After each component of the interferometric phase is calculated, the original phase time series are compared with the calculated interferometric phases, and the linear and nonlinear displacement phases of the study area are finally obtained.

4. Results

4.1. Landslide Susceptibility

This study focuses on the inherent sensitivity of landslides; that is, the instability of slopes is determined only by their lithology and structural conditions, without considering the influence of external triggering factors such as rainfall or earthquakes. Based on the method in Section 3.1, the slope structure types in the region are classified into five types: dip slope (0–30°), down slope (30–60°), transverse slope (60–120°), reverse slope (120–150°), and counter-tilt slope (150–180°) (Figure 3). According to the literature [59], dip slopes are the least stable of the rocky slopes, with a significantly higher density of disaster points than other slope structures, and are classified as unstable slopes prone to hazards. Moreover, in a case study of Fuling District (Chongqing), the geomorpho-structural delineation method produced slope units whose landslide-susceptibility predictions achieved 82% accuracy, compared with 65% for traditional hydrological-analysis units [60]. Visual interpretation identifies the Baige landslide as a dip slope susceptible to landslides.

4.2. CWC Estimation Results

4.2.1. Model Parameter Sensitivity Analysis

In this study, the Sobel algorithm was employed to analyze the global sensitivity of the parameters in the PROSPECT and SAILH models, with the sensitivity curves displayed in Figure 4. Among them, the global sensitivity of N shows minimal variation and remains relatively stable within the range of 400 to 2500 nm. Cab primarily influences canopy reflectance from 450 to 780 nm, exhibiting a sensitivity peak near 580 nm, while its contribution is negligible in the range of 780 to 2400 nm. EWT primarily influences canopy reflectance from 1100 nm onward and has no effect on reflectance in the visible region, demonstrating high global sensitivity. The global sensitivity of Cm fluctuates between 0 and 0.8, with a noticeable decrease of around 1350 nm. LAI exhibits high global sensitivity, presenting a distinct peak and valley near 760 nm. These results indicate that parameters N and Cab, which show lower sensitivity, have relatively minor effects on reflectance data. Conversely, LAI, EWT, and Cm significantly influence the reflectance of Canopy Water Content (CWC).

As detailed in

Table 2. Recalibration of input parameters of the PROSAIL model for CWC results.

Parameter	Range	Step Size
Leaf structure (N)	1.5	-
Chlorophyll content (Cab)	40 μg/cm2	-
Equivalent Water Thickness (EWT)	0.007–0.05	0.002
Leaf Area Index (LAI)	2–6	0.02
Dry matter content (Cm)	0.018–0.02 g/cm2	0.002
Carotenoid content (Car)	20 g/cm2	-
Observed azimuth angle (${\theta }_v$)	0°	-
Solar zenith angle (${\theta }_s$)	20°	-
Soil moisture ratio (Psoil)	0.2	-
Hot spot factor (Hspot)	0.003	-
Average leaf inclination angle (ALIA)	40°	-

, all parameter combinations were enumerated with specified increments when inputting into the PROSAIL model. The step sizes were determined based on data representativeness and the magnitude of variations to enhance accuracy. Other parameters, including Hspot, ${\theta }_S$, ${\theta }_V$, and Psoil, were assigned fixed values based on relevant literature, resulting in a total of 11 calibrated parameters.

4.2.2. Spatiotemporal Characteristics of Landslide Creep and CWC

Canopy Water Content (CWC) was estimated using the PROSAIL model with Sentinel-2 image data collected over three periods: 30 July 2016, 4 August 2017, and 25 July 2018. The vegetation on the Baige landslide exhibited a distinct growth status compared to adjacent areas. The boundaries of the landslide were delineated based on changes in CWC, leading to the identification of four vegetation anomaly regions (A–D), as illustrated in Figure 5a. These anomaly areas are located on rocky slopes with steep slopes and sparse vegetation (LAI < 0.3), so the CWC inverted by PROSAIL mainly reflects the limited Canopy Water Content state rather than the large leaf biomass. From Figure 5b–d, it is observed that from July 2016 to July 2018, regions A and C transitioned from predominantly yellow to red. Region A expanded from the back edge toward the front edge of the landslide, while region C remained relatively stable. Region B progressively deepened from orange to red, gradually extending towards the center of the landslide area, demonstrating a north-south expansion trend. Region D transformed from green in 2016 to red in 2018, indicating a rapid downward expansion. Overall, as the time of potential landslide occurrence approaches, groundwater disturbance affects vegetation, leading to a gradual decrease in CWC. This observation suggests that the landslide is currently in a slow creep stage.

The growth status of vegetation exhibits a physical characteristic of interannual cyclic patterns: a continuous acceleration in spring with biomass initiation, peak chlorophyll content, and biomass in summer, a gradual decline in autumn, and reaching a minimum in winter. A comparative analysis of CWC results from 5 June 2018 to 25 July 2018 (Figure 6a,b) indicates a gradual increase in CWC in July, consistent with the observed growth pattern. Except for the landslide area, other regions transitioned from large areas of orange to green. The sections labeled ABCD shifted from orange to red, indicating that vegetation may be adversely affected by landslide creeping, resulting in decreased water content and impaired growth and development. Although CWC is considered a relative “early warning indicator” rather than mechanistic evidence of groundwater stress, this consistent downward trend may indicate that hydrological disturbances within the slope have been increasing before the catastrophic instability, supporting the judgment that the landslide has entered the slow creep stage during the three-year observation period.

4.2.3. Quantitative Analysis of Vegetation Anomaly Pixels

The study above is primarily based on the visual interpretation of CWC for analysis, and the trend of CWC change caused due to landslide creep necessitates further examination from a quantitative perspective. Considering the specific conditions of the Baige landslide, a low CWC threshold of 0.04 was established, categorizing areas with CWC below 0.04 as low CWC and those above as medium-high CWC. The study quantitatively assessed the number and percentage of low CWC pixels within regions A to D of the landslide area, as presented in

Table 3. Spatiotemporal variation of the number and percentage of CWC pixels at four landslide areas.

	30 July 2016		4 August 2017		25 July 2018
	Pixel	Percentage	Pixel	Percentage	Pixel	Percentage
A(CWC < 0.04)	64	32%	146	73%	153	76.5%
B(CWC < 0.04)	187	62.3%	194	64.7%	220	73.3%
C(CWC < 0.04)	8	6.8%	87	74.4%	95	81.2%
D(CWC < 0.04)	12	7.5%	68	42.5%	81	50.6%

. It is evident that from 2016 to 2018, vegetative growth anomalies were observed in the landslide area. For instance, the number of low CWC pixels in region A increased from 64 before the landslide to 153, with the corresponding percentage rising from 32% to 76.5%. Similar increases were noted in the other three regions. These findings corroborate the visual interpretation and align with observed landslide creep phenomena, thus reaffirming the correlation between CWC and landslide creep.

4.3. Surface Deformation Extraction Results

4.3.1. Surface Deformation Analysis

Figure 7a shows the surface deformation before the landslide and the deformation results along the line of sight (LOS) based on the SBAS-InSAR technique, with a maximum deformation rate of −85 mm/year. The Baige landslide exhibits significant surface deformation, with over half of the deformation rates exceeding −30 mm/year. Several locations in the central part of the landslide demonstrate deformation rates exceeding −60 mm/year, primarily concentrated in the upper-middle section of the landslide. Before the initial movement of the landslide, the area exhibited pronounced slope fragmentation. Ongoing collapses at the rear edge of the slope result in debris accumulation in the central-lower region during downward movement, leading to noticeable uplift phenomena.

4.3.2. Quantitative Analysis of Surface Deformation

To enable a more comprehensive quantitative analysis of the spatiotemporal evolution of the landslide, characteristic points P1–P4 were selected within anomaly zones A–D for correlation analysis (Figure 7a). Figure 7b combines the cumulative surface-displacement time series of P1–P4 with monthly precipitation: the displacement curves (right axis) illustrate movement at each point, while the bar chart (left axis) reflects rainfall intensity. All four monitoring locations exhibited an approximately linear increase in displacement before mid-2017, followed by a pronounced acceleration that coincides with a peak rainfall event (>190 mm in July 2017). Of these points, P3 (zone C) shows the greatest cumulative displacement (−112 mm), followed by P2 (−103 mm), P1 (−93 mm), and P4 (−84 mm), indicating spatial variability in deformation intensity across the slope. The fact that heavy rainfall precedes the acceleration phase underscores the critical coupling between hydrological forcing and slope creep behavior.

Figure 7b illustrates the displacement process of the Baige landslide, which can be divided into three distinct phases: the initial startup stage (Stage I, from November 2015 to April 2017), the constant velocity displacement stage (Stage II, from April 2017 to May 2018), and the accelerated displacement stage (Stage III, from May 2018 to October 2018). The cumulative displacement during Stage I is minimal, and the landslide exhibits stable displacement. In Stage II, the cumulative displacement is substantial, with a significant increase in the rate of landslide displacement. By Stage III, the cumulative displacement reaches its maximum. Analyzing the relationship between rainfall and displacement reveals that the rate of surface displacement of the landslide gradually accelerates during periods of frequent rainfall. The acceleration of displacement and displacement in these three stages is strongly correlated with rainfall. Consequently, particular attention should be directed toward slopes exhibiting significant displacement changes during heavy rainfall to mitigate the risk of landslide hazards.

4.4. Analysis of Spatial Variability of Landslide

A longitudinal profile from the leading edge to the trailing edge of the landslide was plotted (Figure 8a), taking into account the evolutionary pattern of CWC and surface displacement across the landslide. Figure 8b presents the water content trend along the profile from the slope foot to its top, while Figure 8c illustrates the surface displacement pattern from the base to the summit of the landslide. Analyzing the typical positions along the profile line reveals observable regularities in the data. For instance, Figure 8b shows a low peak at a transverse coordinate of 10, corresponding to a low surface displacement point. Similarly, points 20 and 30 exhibit low CWC levels and simultaneous low peaks in surface displacement. In contrast, at coordinates 40 and 50, both CWC and surface displacement show simultaneous peaks. When the point is at 60, both are at low peaks at the same time. Therefore, at the characteristic point, the trend changes of the two monitoring signals are consistent. When the CWC reduction is at a low value, the vegetation condition deteriorates; at the same time, the surface displacement reduction is also at a low peak, but its absolute value increases and landslide creep occurs.

Figure 9 shows the changing trend of the characteristic points along the longitudinal section from the front edge of the landslide (position 0) to the rear edge of the landslide (position 70). Between positions 0 and 30, the CWC values remain stable, ranging from 0.025 to 0.03, and the surface displacement is minimal (within 0.002). This limited variability indicates that the lower section of the landslide, near the toe, is relatively stable with minimal signs of active creep. From positions 30 to 40, the variations in CWC and surface displacement are not completely synchronous, implying that additional geological factors, such as local variations in lithology, jointing, or subsurface structures, may influence the deformation patterns independently of canopy water status. Between positions 40 and 60, located closer to the landslide’s head scarp, both variables exhibit significant fluctuations: the CWC first rises sharply from 0.021 to a peak of approximately 0.035 around position 50, likely reflecting enhanced water absorption and retention due to initial tensile cracking and structural disturbances within the slope. Subsequently, it sharply declines to about 0.012, indicating a deterioration in vegetation conditions possibly caused by increased root-zone saturation, oxygen depletion, and consequent vegetation stress. Concurrently, surface displacement shows a corresponding increase from about 0.002 to approximately 0.01, followed by a decrease. This area is located at the back edge of the landslide, and the trends of vegetation canopy water content and surface displacement are highly similar and fluctuating, which is a high-risk area for landslide occurrence. The above analyses show that there is a certain spatial correlation between CWC and surface displacement, and the change trends of the two are consistent.

5. Discussion

5.1. Contributions

In this study, a novel landslide monitoring framework, which integrates the radiative transfer model and SBAS-InSAR techniques, is proposed for detecting anomalous changes in vegetation CWC and surface displacements associated with landslide creep. The method successfully identifies the creep in front of the Baig landslide, which is consistent with the results of previous studies. Some researchers used 23 C-band Sentinel-1A ascending radar images and applied the SBAS-InSAR technique to obtain vertical surface deformation time-series data for the Baige landslide, analyzing its displacement changes [61].

Currently, landslide studies rely mainly on InSAR techniques. This study is based on the fact that the formation of sliding cracks changes the rock and soil structure of slopes, leading to an increase in poor stratification and disorder, which in turn affects the growth of overlying vegetation on landslides. Large amounts of precipitation are absorbed by the vegetation foliage, resulting in a temporary increase in CWC. This water is subsequently transported to the roots and soil through the vegetative system, thereby increasing soil moisture content [62]. However, excessive soil moisture, which occurs when soil moisture consistently surpasses field capacity, can lead to oxygen depletion in the root zone. This hypoxic condition adversely impacts root growth and reduces CWC. In addition, the weight of the soil on the slope and the infiltration of rainwater into the internal rock structure reduce the shear capacity of the slope and accelerate its displacement tendency [63,64]. Studies have also shown that under heavy rainfall conditions, the large pores formed by the roots of vegetation significantly increase the infiltration rate and capacity, causing the groundwater level in the slope to rise rapidly and expand the saturated zone; the resulting pore water pressure not only weakens the shear strength of the soil and triggers landslides, but also inhibits transpiration due to long-term saturation and hypoxia in the root zone, causing the CWC of vegetation to continue to decline [65,66]. Therefore, in complex environments, the introduction of CWC as an indirect indicator for landslide monitoring and the mutual combination of InSAR techniques can effectively overcome the shortcomings of a single method and provide more comprehensive and accurate landslide monitoring results.

5.2. CWC Spatiotemporal Change Impact Factors

The topographic conditions in the mountainous areas of southwest China are complex, with numerous factors influencing the water content of the vegetation canopy, which results in spatiotemporal changes in vegetation that are not necessarily attributable to landslide displacement. The effects of precipitation and topographic factors on vegetation dynamics were examined by analyzing the vegetation recovery in the NDVI dataset following the 2008 Wenchuan earthquake. Precipitation in 2010 and 2011 appeared to inhibit vegetation recovery, with the recovery being particularly weak on gentle slopes at altitudes of less than 1300 m or greater than 3500 m and slopes less than 35° [67]. Another Deijns et al. [68] used the cumulative difference (CD) between the NDVI and the fitted harmonic sine curve for semi-automated landslide monitoring in the Buckinghorse River region of Canada. They found the utility of anomalous changes in vegetation as an indicator of landslide creep, suggesting that riverbank erosion, deforestation, etc., have similar NDVI responses, which may confound landslide characteristics. These studies indicate that the combined effects of hydrological meteorology, topography, the physical and chemical properties of vegetation, land use, and other factors, along with spatial changes in vegetation due to natural phenomena and human activities, must be considered, as they may compromise the accuracy of early landslide identification.

In summary, given sufficient time series of remote sensing images and normal vegetation phenological characteristics, the multiple quantification of known landslide time series images is employed to train the detection of abnormal vegetation sequence characteristics in specific landslide areas. This approach effectively distinguishes landslides from non-dominant factors such as farmland and bare land, enhancing the accuracy of spatial distribution recognition of landslides across extensive regions, which represents a significant direction for future research on landslide hazard identification in complex environmental contexts.

6. Conclusions

Considering the limitations of the current single landslide monitoring technique in complex environments, this study proposes a retrospective spatiotemporal analysis framework based on the radiative transfer model and the SBAS-InSAR technique, which effectively correlates the CWC with the surface movement and reveals the potential patterns of the landslide behavior through the remotely sensed data. We conclude that vegetation information can indirectly reflect landslide instability. As the time of landslide occurrence approached, the proportion of vegetation with low CWC in the study area increased significantly; the number of image elements increased from the initial 64 to 153, and the proportion increased from 32% to 76.5%. Surface displacement, on the other hand, allowed direct monitoring of landslide creep, with deformation rates ranging from -85 mm/year to 32 mm/year. In addition, there is a certain spatial correlation between the changing pattern of CWC and surface displacement. When the CWC decreases, the surface displacement is at a low point, and the trend change is consistent. Therefore, combining multi-source data and focusing on the response of such weak information as vegetation to landslides provides new ideas for future monitoring and prevention of geological hazards in complex environments. While the PROSAIL model assumes uniform canopy structure and fixed leaf optical properties, the CWC estimation may be biased under the conditions of diverse vegetation types and densities and needs to be calibrated by introducing high-frequency satellite imagery and combining it with field observations in subsequent studies.