Quantitative Stability Assessment of Landslides Following the 2024 Zixing Rainstorm Using Time-Series InSAR

Highlights

What are the main findings?

• Comprehensive InSAR monitoring revealed that nearly half (over 48%) of the rainfall-induced landslides in Zixing remained active or rapidly active during the year following the major disaster, with localized ground displacement rate exceeding 40 mm/year.
• The study successfully integrates the InSAR-derived displacement result with rainfall, geological, and optical imagery to assess the post-failure stability of a large landslide population, highlighting a correlation between accelerated landslide movement and precipitation.

What are the implications of the main findings?

• The high proportion of active landslides one year after the initial event underscores the prolonged hazard and significant risk to local communities, emphasizing the critical need for continuous assessment to inform early warning and disaster mitigation.
• The findings demonstrate the vital role of InSAR technology in moving beyond mere landslide inventory mapping to practical stability assessment.

Abstract

In July 2024, a major rainfall-induced landslide disaster occurred in Zixing county, Hunan Province, triggering more than 4000 landslides with a total area exceeding 21 km². The scale of this hazard underscores a critical need for long-term stability assessment of the affected slopes. While previous studies have primarily used optical remote sensing to map landslide distributions, quantitative evaluation of post-failure movement dynamics remains limited. This study developed an integrated monitoring framework that combines time-series SBAS-InSAR displacement measurements (using Sentinel-1 data from August 2024 to September 2025) with deep learning-based optical interpretation, rainfall analysis, and geological data. Our approach enables the quantitative, region-scale stability assessment of the Zixing landslide cluster one year after the initial event. Experimental results reveal sustained surface displacement with rates ranging from −30 to 30 mm/year, and localized displacements exceeding 40 mm/year. Notably, over 48% of the mapped landslides are classified as active or critically active, indicating widespread, ongoing instability. Correlation analysis further establishes precipitation as a key driver of accelerated movement. Beyond the Zixing case, this work provides a transferable methodology for assessing long-term post-disaster landslide behavior, offering direct value for regional hazard management and early-warning systems.

1. Introduction

As one of the most common and widespread geological hazards, landslides threaten human life and property safety, infrastructure security, and regional sustainable development [1,2]. China’s vast mountainous terrain and complex geology make it highly prone to landslides, which are often triggered by heavy rainfall, earthquakes, and human engineering activities [3]. In recent years, intensified global climate change has led to a marked rise in the frequency and intensity of extreme rainfall events. Consequently, rainfall-induced landslides have become a major geological hazard in many regions [4]. In July 2024, an extreme rainstorm struck Zixing City, Hunan Province, triggering multiple landslides. These events severely impacted local livelihoods, road transport, and the ecological environment [5,6,7]. Although post-disaster emergency investigations and mitigation measures have yielded initial results [8,9], the long-term stability of failed landslide masses and the associated risks of prolonged displacement or reactivation constitute a key scientific question that warrants further investigation. Therefore, continuous displacement monitoring and quantitative stability assessment of these landslide masses are crucial for long-term risk management.

Traditional methods for assessing landslide surface movement and stability primarily rely on techniques such as GNSS monitoring, UAV-LIDAR, leveling surveys, and manual surface measurements [10,11,12,13]. These methods are capable of providing high-precision, point-specific displacement monitoring data. However, their application is often constrained by high costs, limited spatial coverage, and operational difficulties in rugged mountainous terrain [14,15]. Particularly in areas where landslides have already occurred, fragmented terrain and difficult access pose significant challenges to implementing traditional monitoring approaches [16]. Interferometric Synthetic Aperture Radar (InSAR) technology, especially time-series InSAR, has becoming a revolutionary tool for surface displacement monitoring due to its millimeter-to-centimeter-level accuracy, wide-area coverage capability, and independence from weather and daylight conditions [17,18,19,20]. By processing synthetic aperture radar (SAR) images of the same area acquired at different times, InSAR can detect subtle surface displacement along the radar line of sight (LOS), thereby revealing the dynamic evolution of slow surface displacement processes, such as landslides, land subsidence, and glacier movement [21,22,23]. Currently, InSAR technology has been widely applied to landslide hazard identification, dynamic monitoring, and mechanism analysis, yielding substantial research outcomes [24,25,26].

With the support of machine learning, researchers can now integrate displacement result, topography, geology, land use, and other multi-source datasets. Consequently, landslide susceptibility mapping and early warning have emerged as a key area of investigation [27,28,29]. While these efforts largely aim to predict future risks, studies addressing the core question of post-disaster assessment remain relatively limited. In 2024, one year after the landslide in Zixing City, Hunan Province, deformation had not fully ceased. The landslide mass appears to have entered a prolonged, slow creep adjustment phase, with certain local blocks possibly remaining unstable. There remains a risk of reactivation or secondary sliding under subsequent rainfall conditions [8,9]. A summary of existing studies on landslides in Zixing is provided in

Table 1. Comparative Study of Landslides in Zixing.

Reference	Data	Method	Result
[5]	Gao Fen, Aerial images	Remote sensing analysis and field surveys	Three primary landslide clusters were identified, revealing the relationship between landslide distribution and environmental factors.
[6]	Planet Scope	Manual visual interpretation	Compiled an inventory of 23,513 landslides.
[9]	Gao Fen	Visual interpretation	A total of 19,764 landslides were identified.
[30]	Planet Scope	Deep Learning, machine Learning	A total of 16,120 landslides were identified.
[31]	Gao Fen, Sentinel-2 L2A	Visual interpretation, field investigations and UAV, deep learning	A database containing 19,403 shallow landslide was constructed.

[5,6,9,30,31]. By integrating remote sensing and machine learning approaches, these investigations have produced accurate landslide inventory maps, offering foundational scientific data for subsequent mitigation efforts. Nevertheless, research on post-rainfall surface displacement and stability assessment of landslide masses in the area remains scarce. Furthermore, the extent to which rainfall continues to affect the stability of these landslide bodies warrants sustained monitoring. In contrast, InSAR technology offers a direct and effective means to monitor post-disaster movement signals, providing essential data for evaluating current stability [32,33]. Analyzing the spatial distribution and temporal evolution of the displacement field, along with its correlation with rainfall, enables a scientific assessment of landslide stability and the identification of actively hazardous areas [34,35]. It should be noted, however, that zero or non-accelerating displacement may reflect force equilibrium in rigid geological materials rather than true stability; therefore, the interpretation of landslide activity status should consider geolithological context where available. This analysis provides critical support for evaluating mitigation measures, delineating monitoring zones, and developing targeted risk management strategies.

In this study, we aim to conduct refined displacement monitoring of the main landslide-affected areas based on the time-series InSAR method with Sentinel-1 SAR images magse acquired from July 2024 to September 2025. The specific objectives of this study are to derive high-precision displacement velocities and time-series data for the post-event period, and to extract and analyze the spatiotemporal displacement characteristics within the identified landslide boundaries. Furthermore, the study aims to explore the correlation between ongoing displacement and post-event rainfall to reveal post-disaster activity patterns. Finally, it seeks to assess the stability of different landslide masses based on persistent movement signals. This assessment will offer scientifically reliable support for local disaster prevention and risk management while also providing a transferable technical framework for post-disaster evaluation. Furthermore, it can serve as a methodological reference for the long-term monitoring of similar rainfall-induced landslides in other regions.

2. Study Area and Dataset

2.1. Study Area

The study area is located in Zixing City, in southeastern Hunan Province and eastern Chenzhou City, as shown in Figure 1a,b. Zixing lies in the upper reaches of the Leishui River basin, within a low- to medium-elevation mountainous and hilly zone at the southern end of the Luoxiao Mountains. The terrain primarily consists of mountains, along with hills, ridges, and plains. Topography generally slopes downward from southeast to northwest, with the highest point reaching about 2042 m at Bamian Mountain. Widespread mountainous and low-hill areas feature deep valleys and steep slopes, creating favorable conditions for geological hazards such as landslides, rockfalls, and debris flows [6,36]. The lithology in Zixing is dominated by granite and slate, as illustrated in Figure 1c. Climatically, Zixing has a subtropical monsoon humid climate with distinct seasons. The region features mild temperatures, high humidity, and abundant rainfall, with precipitation concentrated primarily from March to June and again in August. It experiences an average of 182 rainy days annually, marked by heavy rainfall in spring and summer, while autumn and winter remain relatively dry.

In late July 2024, Zixing City experienced sustained heavy convective rainfall associated with a typhoon periphery, which triggered widespread landslides, rockfalls, and debris flows. Field surveys confirmed significant slope instability across multiple mountainous areas (Figure 1d), with landslides blocking roads and damaging residential structures [8,9]. This extreme rainfall event significantly destabilized slopes across the region, forming the essential background for this study. Using SBAS-InSAR technology, this study monitors ground surface displacement and assesses post-event landslide stability.

In October 2024, a field survey was conducted in typical landslide-prone areas of Zixing City, Hunan Province. The investigation documented geomorphological features, geological conditions, surface vegetation, and damage to surrounding infrastructure. As shown in Figure 2, panel (a) displays a landslide adjacent to a mountain road. Panels (b), (c), and (d) show landslides near residential areas. In Figure 2b, the landslide mass has directly impacted residential building foundations, indicating that slope movement has already caused structural damage. The landslides in Figure 2c,d are relatively large in scale, with clearly visible surface movement and vegetation destruction. Their sliding direction points toward residential areas, highlighting a significant geological hazard and posing a serious threat to local residents and infrastructure.

2.2. Dataset

This study employs an integrated multi-source remote sensing datesets to assess the stabilities of landslide in Zixing City. The datasets include high-resolution optical imagery, synthetic aperture radar (SAR) data, and topographic data, supplemented by ground meteorological observations. High-resolution optical images were acquired from the China Centre for Resources Satellite Data and Application (CRESDA, https://data.cresda.cn/) and are derived from the GaoFen-6 (GF-6) satellite. With panchromatic and multispectral spatial resolutions of 2 m and 8 m respectively (

Table 2. Optical data information.

Data Type	Acquisition Date	Resolution	Source
GF-6	5 August 2024	2 m, 8 m	https://data.cresda.cn/
GF-6	11 May 2025	2 m, 8 m	https://data.cresda.cn/

), the GF-6 imagery enables detailed surface interpretation and accurate delineation of landslide boundaries. Sentinel-1 C-band SAR data from the European Space Agency (ESA) were used in this study [37]. A total of 27 Sentinel-1A images, acquired between August 2024 and September 2025, were downloaded from the Alaska Satellite Facility (ASF, https://search.asf.alaska.edu) for Small Baseline Subset Interferometric SAR (SBAS-InSAR) time-series analysis. The Sentinel-1 images have an approximate resolution of 5 m in range direction and 20 m in azimuth direction. Additionally, SRTM digital elevation model (DEM) data were incorporated to support the geometric correction and phase unwrapping steps in the time-series InSAR processing [38].

Daily precipitation data were obtained from 23 ground-based monitoring stations managed by the Hunan Meteorological Bureau. The locations of these stations are shown in Figure 1b. Figure 3 illustrates the daily rainfall precipitation recorded at two representative meteorological stations (P8495 and P8500). As shown, the majority of rainfall is concentrated between April and August, with the maximum daily precipitation exceeding 80 mm. The data from both stations exhibit strong consistency, confirming the regional representativeness of the rainfall pattern. These precipitation records provide a crucial temporal reference for analyzing the relationship between rainfall variability and landslide displacement dynamics.

3. Method

The methodological framework of this study consists of three main steps. First, the SBAS-InSAR method was applied to derive time-series surface movement results in the Zixing area. Next, high-resolution optical data and deep learning models were used to extract and analyze landslide boundaries and their changes. Finally, InSAR-derived displacement results, along with other environmental factors, were employed to assess landslide stability. The detailed workflow is illustrated in Figure 4.

3.1. InSAR Displacement Result Estimation

In mountainous landslide-prone areas, inter-annual vegetation changes often cause significant interferometric decorrelation (Figure 2), which can degrade the quality of conventional InSAR results. To address this issue, this study employs the Small Baseline Subset Interferometric SAR (SBAS-InSAR) method for surface displacement monitoring. This technique has been widely and successfully applied in landslide and geological hazard studies [25,39]. The main processing steps of SBAS-InSAR include image registration, interferogram generation, phase filtering, phase unwrapping, displacement parameter estimation, atmospheric phase correction, and time-series displacement estimation. Further technical details on the processing workflow can be found in [40].

In the SAR interferometric processing, the temporal baseline threshold was set to 60 days and the spatial baseline threshold to 150 m. A total of 107 interferometric pairs were generated for displacement inversion. During interferogram generation, multi-looking processing with 8 pixels in the range direction and 2 pixels in the azimuth direction was applied to suppress interferometric noise. An adaptive phase filtering algorithm was used to filter the interferograms. In mountainous area, the interpretation and analysis of surface displacement signals become more challenging due to the presence of atmospheric delays. To mitigate atmospheric delay effects, GACOS (General Atmospheric Correction Online Service) data were applied during processing. This step helps eliminate or reduce phase variations caused by atmospheric delay in the differential interferometric phase, leading to more accurate surface movement results in the study area [41]. The Minimum Cost Flow (MCF) algorithm was used for phase unwrapping. This study employed the average coherence coefficient method for selecting coherent points, where all points with an average coherence greater than 0.7 were selected as high-coherence points [42]. On this basis, the SVD (singular value decomposition) method was used to estimate the displacement rate and calculate the time-series displacement, resulting in a sequence of surface displacement outcomes along the line of sight. Subsequently, low-pass filtering in the spatial domain and high-pass filtering in the temporal domain were applied to obtain more precise final displacement results in the time series.

3.2. Landslide Mapping

Deep learning technology has demonstrated strong application capabilities in fields such as image processing, object detection, and recognition [43,44,45,46]. In this study, we utilize a deep learning-based approach with optical images to identify and map landslide boundary. High-resolution multispectral satellite data covering key geological hazard zones in Zixing City served as the primary input. The imagery first underwent standard preprocessing, including radiometric calibration, atmospheric correction, orthorectification, and spatial subsetting, to ensure spectral and geometric consistency across multi-temporal datasets. The landslide sample database was primarily constructed using multispectral imagery from the Gaofen-6 satellite with a spatial resolution of 2 m. High-resolution historical imagery from Google Earth provided essential visual verification to ensure labeling accuracy. The initial dataset comprised 1541 raw image patches of 256 by 256 pixels, all of which were meticulously annotated to maintain high label consistency. To improve model generalization and mitigate overfitting, the original dataset was expanded to 9246 patches through various data augmentation techniques, specifically random noise injection, Gaussian blurring, horizontal or vertical flipping, and brightness adjustment. The augmented dataset was randomly partitioned into training, validation, and testing sets according to a fixed ratio of 8 to 1 to 1. This study adopts the classic U-Net semantic segmentation architecture for landslide detection; its overall framework is shown in Figure 5. U-Net integrates an encoder–decoder structure to extract hierarchical features while reconstructing spatial details and uses skip connections to preserve fine-grained boundary information along with deep semantic representations. This design makes it well-suited for delineating landslides, which typically exhibit irregular shapes and multi-scale characteristics [47,48,49]. The model takes optical imagery as input and outputs pixel-level landslide delineations, enabling automatic identification and segmentation of landslide areas. Finally, vectorized landslide boundaries were generated to define the spatial extent of landslides within the study area. Based on optical data from August 2024 and May 2025, the landslide distribution was extracted. This inventory was then integrated with InSAR displacement results to evaluate the spatial distribution and stability of landslides in Zixing.

3.3. Landslide Assessment Using InSAR Results

To evaluate landslide stability and activity, this study utilized InSAR-derived deformation velocity data. After obtaining both the InSAR deformation results and the landslide boundary maps, deformation points within each landslide were extracted using the landslide inventory. The average deformation velocity for each landslide was then calculated to quantify its overall activity level. Subsequently, landslide stability was classified into four levels based on average displacement velocity, in accordance with established criteria from prior studies [50]. The specific classification thresholds are summarized in

Table 3. The classification of the landslide based on the InSAR results.

Classification	Rapidly Moving	Active	Potentially Unstable	Stable
InSAR velocity (mm/year)	≤−20	(−20, −10)	(−10, −5)	(−5, 0)

.

4. Experimental Results

4.1. InSAR Displacement Results

Using SBAS-InSAR method, 27 Sentinel-1A images acquired from August 2024 to September 2025 were collected and processed to derive the surface deformation of Zixing City (Figure 6). In the Zixing area, vegetation cover has led to a lack of coherent points in some parts of the InSAR results. However, measurable points remain sufficiently dense within the landslide areas. Overall, displacement rates across the study area show significant spatial variation, ranging from −90 mm/year to +40 mm/year. The regional average deformation rate is approximately −12.6 mm/year. The overall LOS deformation field suggests a general trend of slight displacement away from the satellite, while several local zones exhibit pronounced positive LOS signals. Given the dependency of LOS measurements on imaging geometry, these positive displacements may also indicate active or potentially unstable slopes, depending on the local slope aspect relative to the satellite’s viewing direction.

Three representative sub-regions (R1, R2, and R3) were selected for local movement analysis. As shown in the enlarged views (Figure 6b–d), all three areas display notable subsidence anomalies. This indicates relatively active localized ground instability and collapse. The results highlight the spatial variability of surface deformation in Zixing and underscore the need for ongoing monitoring of potential geohazards.

The time-series InSAR deformation in the typical subsidence zones of Zixing City is illustrated in Figure 7. From the overall spatial pattern, the study area is characterized by a general trend of slow subsidence during the observation period. Most observation points exhibit deformation values ranging from −30 to −10 mm, indicating that the regional surface deformation remains at a relatively low and stable level. However, several localized areas display pronounced negative displacements, reflecting the continuous deformation processes of unstable landslide bodies within the region.

Based on SBAS-InSAR time-series displacement results and corresponding rainfall data, three representative landslides (L1, L2, L3) were selected for temporal analysis (Figure 8, Figure 9 and Figure 10). Their displacement evolution shows clear temporal coupling with precipitation. The L1 landslide displays a consistently decreasing displacement trend over time, with an average velocity of −37.66 mm/year (Figure 8). It can be observed that when rainfall increased in April 2025, the displacement at point P1 accelerated significantly. The subsidence rate at point P increased from 35 mm/year to 72 mm/year.

As shown in Figure 9a,b, Landslide L2 exhibits significant subsidence across the entire area, with a maximum displacement rate exceeding −40 mm/year. Point P2, located within L2, shows a continuous subsidence rate of −31.17 mm/year during the observation period. Following several intense rainfall events, the displacement at Point P2 increased in short-term acceleration phases (Figure 9c), indicating high sensitivity to extreme precipitation.

In contrast, Landslide L3 shows a relatively small cumulative displacement (−9.09 mm/year) and an overall stable trend. Minor fluctuations occur during months of heavy rainfall, suggesting it may be in a potentially unstable or incipient displacement stage (Figure 10). The results indicate a strong positive correlation between rainfall intensity and landslide displacement velocity: more concentrated and intense precipitation leads to more pronounced landslide activity. This finding underscores the critical role of rainfall as a key trigger for slope instability and deformation dynamics in the study area.

4.2. Landslide Detection with Optical Images

Using high-resolution optical images acquired from August 2024 and May 2025, the landslide distributions in Zixing was obtained with the deep learning model described in Section 3.2 (Figure 11). Figure 12 shows the model’s loss and MIoU across training epochs, where loss clearly decreases, indicating effective convergence. The landslide distribution maps from both periods are highly consistent spatially. Previous studies reported 19,764 landslides in the Zixing region, with high densities concentrated in Bamianshan and Zhoumensi [9]. Although total landslide counts differ due to varying study areas, the number and spatial clustering patterns remain consistent within the overlapping Bamianshan and Zhoumensi zones. The landslides identified in our two periods number 4236 and 4352, respectively, showing only a minor change. A comparison with existing results confirms a similar spatial distribution [8,9]. It should be noted, however, that detecting landslide changes based solely on these two optical observations is challenging. Slow landslide movements are difficult to capture through texture changes in optical data and require integration with other sources, such as InSAR displacement results. In Section 4.2, clear displacement signals can indeed be observed for some landslides.

4.3. Landslide Assessment

Based on the long-term SBAS-InSAR-derived displacement velocity results and the landslide displacement rate classification criteria, the lanslides stability across the study area was systematically evaluated (Figure 13). The landslide kernel density map (Figure 13a), calculated using optically extracted landslide distribution data, reveals that the landslide areas are primarily concentrated in the northeastern part of Zixing County (the area within the white box in Figure 13a). Therefore, we extracted the InSAR displacement results for this corresponding region to assess each landslide.

According to the magnitude and temporal evolution of displacement velocity, landslide stability was classified into four levels: Stable—Potential—Active—Rapid, as shown in Figure 13b. Relatively obvious signs of movement have been observed in the northern and southern parts of the landslide area under study, as indicated by the red dots in Figure 13b. Statistical analysis identified a total of 4236 landslide points within the study area, of which 488 (11.5%) were classified as stable, 1680 (39.7%) as potentially unstable, 1568 (37.0%) as active, and 500 (11.8%) as rapidly deforming, as shown in Figure 14. These results indicate that over half of the landslides fall within the potential to active categories, suggesting generally low stability and widespread displacement activity. Overall, the landslide stability in Zixing City is relatively poor, with nearly 77% of the slopes exhibiting potential or active instability. The findings reveal a pronounced spatial concentration of unstable slopes, implying a high level of susceptibility to secondary geological hazards and underscoring the necessity for continuous displacement monitoring and early warning in key risk areas.

5. Discussion

5.1. Characteristics of the Landslides

Based on the 4236 identified landslide sample points in the study area, a multi-factor statistical analysis was conducted, incorporating topography, vegetation, slope aspect, and lithology (Figure 15). The results show that landslide activity is mainly concentrated in low- to mid-mountain regions, where terrain relief and geological–geomorphological conditions play a dominant role. Vegetation coverage analysis reveals that most landslide points are located in areas with NDVI > 0.75, indicating a spatial association with vegetated terrain. However, this may reflect landslides occurring in small clearings or disturbed patches within forested areas, rather than on densely forested slopes themselves. During heavy rainfall, such areas—where root reinforcement is weakened or soils are saturated—may be particularly susceptible to shallow sliding or surface collapse. Elevation statistics reveal that landslides are most frequent between 300 and 800 m, with peaks at 300–400 m and 600–700 m. These correspond to hilly and low-mountain terrain, where rainfall concentration and infiltration are enhanced. Slope gradient analysis shows landslides mainly occur on moderate to steep slopes (8–24°), with the highest frequency in the 12–20° range. Medium slopes often have thicker soil and more human disturbance (e.g., road cuts), reducing stability, while steeper slopes are prone to gravitational failure due to higher potential energy. Slope aspect displays a clear directional pattern, with landslides most common on south-, southeast-, and east-facing slopes, followed by northeast-facing aspects. Southeast-facing slopes receive dominant monsoonal rainfall, leading to greater infiltration and runoff during heavy precipitation, which increases instability. In terms of land cover, landslides are largely found in evergreen broadleaf forests and mixed coniferous–broadleaf forests (Figure 15e). This indicates that even dense vegetation cannot fully ensure slope stability under intense rainfall, as root reinforcement weakens when soil is saturated. Regarding lithology, most landslides are concentrated in granite areas (Figure 15f). Here, rocks are often highly weathered, fractured, and mechanically weak. These conditions, combined with concentrated runoff and elevated pore water pressure, favor sliding surface development and slope failure.

5.2. Impacts of Rainfall

Rainfall is the primary external driver of landslide displacement and evolution in the Zixing area, showing a strong spatiotemporal coupling with landslide activity. Daily rainfall observations were obtained from the Hunan Meteorological Bureau and aggregated into monthly totals for each station. Using station coordinates and monthly precipitation values, vector point datasets were created and spatially interpolated to produce continuous raster precipitation maps. Based on monthly rainfall from September 2024 to August 2025 (Figure 16), precipitation in the study area displays distinct seasonal variation. From October 2024 to February 2025, the region experienced a dry season with monthly precipitation generally below 100 mm. During this period, landslide displacement rates remained relatively stable. Starting in March 2025, rainfall increased rapidly. From April to June, widespread heavy rainfall occurred, with local monthly totals exceeding 350 mm. This significantly impacted slope stability. During these months, multiple monitoring points showed abrupt changes or acceleration in displacement, confirming rainfall as a direct trigger for slope displacement. Rainwater infiltration raises groundwater levels, increases pore water pressure, and reduces effective stress and shear strength within slopes, collectively accelerating deformation and potentially leading to failure. Overall, a positive correlation exists between rainfall intensity and displacement magnitude. Precipitation distribution governs the seasonal pattern of slope displacement, while extreme rainfall events are the main triggers for intensified landslide activity and potential instability.

5.3. Comparison with Optical Images

To validate the temporal consistency and reliability of landslide hazards, a comparative analysis was performed using multi-temporal optical imagery of Zixing City (Figure 17). The first row (a–c) shows landslide zones identified by InSAR displacement velocity. The second row (d–f) displays optical images from August 2024, and the third row (g–i) shows the same locations in May 2025. All three landslide sites show clear negative LOS movement signals, indicating subsidence or backward movement relative to the satellite. Their mean displacement velocities are −10.68 mm/year, −18.05 mm/year, and −8.75 mm/year, confirming ongoing settlement. Comparing landslide boundaries between the two periods reveals visible expansion and morphological changes. This demonstrates the high accuracy and practical applicability of InSAR-based displacement identification for engineering and hazard monitoring.

Previous studies have widely employed optical remote sensing, often combined with manual interpretation, machine learning, or deep learning, to map landslides (Table 1). For example, Huang et al. manually delineated 23,513 landslides using PlanetScope imagery [6], while Liu et al. automatically identified 16,120 landslides using deep learning and machine learning methods [30]. Overall, existing research has primarily focused on extracting and cataloging landslides from optical imagery using intelligent algorithms. Compared to earlier studies, the total number of landslides detected in this work is relatively limited. This difference is primarily due to our study’s focus on two severely affected areas—Bamianshan and Zhoumensi Town—where landslides are densely concentrated. Further complicating precise delineation, many failures occur in contiguous clusters, leading to aggregated counts. Crucially, this work extends beyond optical-based landslide detection by integrating InSAR analysis, thereby adding a new dimension to the existing foundation. Using Sentinel-1 radar imagery from August 2024 to September 2025, the study enables continuous monitoring of surface movement across the Zixing landslide area.

5.4. Limitation and Future Work

This study systematically analyzes landslide displacement characteristics and rainfall–landslide coupling mechanisms in Zixing by integrating SBAS-InSAR with multi-source data, such as rainfall, topography, vegetation, and lithology. The findings can provide valuable support for landslide monitoring and risk assessment. However, several limitations should be noted. First, the monitoring period is limited to about one year. This timeframe may not fully capture long-term landslide evolution, necessitating longer time-series radar data for future monitoring. For the displacement velocity threshold, the same value may carry different implications depending on lithology and failure mechanisms. Future work would incorporate material properties, detailed geotechnical, and hydrogeological parameters to develop more refined stability criteria. Second, the temporal resolution is constrained by Sentinel-1’s 12-day revisit cycle, which may miss rapid accelerations triggered by short-duration intense rainfall. Finally, the relationship between landslide displacement and rainfall remains largely qualitative. Although a positive correlation is observed, no quantitative hydro-mechanical model was developed to simulate pore-water pressure effects on slope stability.

Future work should focus on improving data quality and expanding ground validation to enhance our understanding of landslide evolution and rainfall impacts, ultimately strengthening geological hazard early-warning capabilities.

6. Conclusions

This study presents an integrated monitoring and assessment framework for the post-failure behavior of rainfall-induced landslide clusters. Focusing on the Zixing event triggered by extreme rainfall in July 2024, we combined time-series SBAS-InSAR displacement monitoring with multi-source data analysis to quantitatively evaluate the spatiotemporal evolution and stability of landslides during the first year after the disaster. The main conclusions are as follows:

(1) Time-series InSAR results clearly reveal the post-disaster surface displacement pattern in the study area. Displacement rates across most of Zixing range between –30 mm/year and 20 mm/year. Significant displacement is concentrated in high-landslide-density zones, where maximum rates exceed –30 mm/year, indicating continued settlement and movement after the initial failure.

(2) Based on high-precision InSAR displacement results, a quantitative stability assessment was performed for 4236 landslide points. The assessment shows that overall landslide stability in Zixing is poor. Potentially unstable and active landslides together account for up to 77% of the total, clearly indicating that most landslide masses remain unstable and continue to pose a substantial failure risk.

(3) Spatiotemporal correlation analysis between time series displacement and rainfall data confirms a strong linkage between landslide stability and precipitation. Statistics demonstrate a positive correlation between rainfall intensity and landslide displacement rate. In particular, intense rainfall events are identified as the primary external driver accelerating landslide displacement. Displacement often increase markedly following peak rainfall periods, suggesting a distinct time-lag effect.