Exploring rainfall effects on the Xingfu Dayuan Landslide in Yecheng, Xinjiang, China using time-series InSAR technology and numerical simulation

Abstract

Rainfall-induced landslides are increasingly frequent in loess-covered regions of Xinjiang, China, posing a serious threat to residents and infrastructure. This study investigates the deformation behavior and stability response of the Xingfu Dayuan landslide in Yecheng County, aiming to clarify how rainfall characteristics control landslide development. Field investigations, time-series InSAR-based deformation monitoring, and numerical simulations were integrated to analyze landslide evolution from 2022 to 2024. Satellite observation results reveal that deformation is unevenly distributed across the slope and is most pronounced in the middle and lower sections, where average deformation rates reach approximately 15–18 mm per year, and cumulative displacement exceeds 50 mm, indicating that the landslide remains active. Deformation intensified from 2023 to 2024 and exhibited clear seasonal variations, with accelerated movement during periods of concentrated rainfall and higher temperatures. Numerical simulation results reveal that rainfall infiltration significantly increases internal water pressure, with infiltration depths extending from about 4 m under low-intensity rainfall to nearly 8 m during extreme rainfall events. As a result, slope stability decreases progressively, with the stability factor approaching or falling below the critical threshold. Notably, the minimum stability does not occur during rainfall but typically 12–24 h after rainfall ceases, revealing a pronounced delayed response of the loess slope to rainfall infiltration. These findings demonstrate that rainfall intensity, duration, and post-rainfall infiltration jointly control loess landslide deformation and stability, highlighting the importance of considering delayed instability in landslide hazard assessment and early warning.

Introduction

The Quaternary loess is a widespread sediment in arid and semi-arid regions. In China, it occurs mainly in the northwestern areas and the middle reaches of the Yellow River Basin, covering approximately 6.31 × 10⁵ km^21,2. In the Yecheng area of southern Xinjiang, this sediment exhibits distinctive material properties and engineering geological characteristics compared with loess from other regions of China^3,4. Influenced by the arid climate and strong westerlies, the loess in this region is predominantly Aeolian in origin, with a relatively loose structure, weak cementation, and high porosity³. Its pore system is characterized by large intergranular voids and vertical joints, resulting in pronounced water sensitivity and rapid strength reduction upon wetting^4,5. Yecheng County, located in the Xinjiang Uygur Autonomous Region, lies on the northern flank of the Karakoram Mountains (Fig. 1a). In this area, thin-bedded loess is widely distributed on slopes and tablelands, forming the primary geomaterial for shallow landslide development⁵. In recent years, influenced by climate change, Yecheng has experienced more frequent episodes of persistent rainfall and short-duration high-intensity rainfall, which readily trigger clustered loess landslides^6,7. The unique structural and hydraulic properties of local loess, combined with rainfall infiltration processes, provide the fundamental geological background for rainfall-induced instability in this region. In the early hours of 18 May 2018, the Second Pasture in Yecheng County experienced a rainfall event that began with moderate to heavy rainfall and later turned into light rainfall, persisting for approximately 12 h. The event triggered eight landslides, blocked approximately 10 kilometres of roads, and caused indirect economic losses amounting to 40 million yuan⁸. With global warming and intensified human activities, rainfall-related shallow loess landslides are exhibiting an increasing trend annually, posing severe threats to human life and property as well as engineering facilities^9–11. Consequently, it is of great significance to gain an in-depth understanding of the failure characteristics and formation mechanics of loess landslides by explicitly considering rainfall infiltration processes.

Fig. 1.

Location of the landslide at Xingfu Dayuan.

The impact of rainfall on the stability of loess landslides has been extensively studied through field investigations, laboratory model tests, and numerical analysis. Some researchers have suggested that, within loess-distributed regions, owing to the collapsible nature of loess, the weakening effect of water on slope stability is more pronounced than that of other factors. Water-induced degradation of the internal pore structure gradually reduces the mechanical properties of the soil at the macroscopic scale¹². When minor failure events reach a critical state, localized plastic damage zones develop within the soil mass and progressively coalesce, ultimately leading to slope instability^13,14. Moreover, the maintenance of the loess pore structure relies on matric cohesion. Rainfall infiltration increases the unit weight of the soil and the pore water pressure within the slope, thereby significantly reducing matric cohesion¹⁵. Numerous studies have linked seepage processes with slope stability, demonstrating that this coupling is conducive to elucidating the physical mechanisms of rainfall-induced shallow landslides^16–18. Therefore, it is necessary to consider variations in pore water pressure under different rainfall conditions for the same slope to gain a deeper understanding of the controlling role of rainfall in the formation and stability evolution of shallow loess landslides. However, studies focusing on actual landslide cases that couple seepage processes under different rainfall conditions with slope deformation characteristics remain limited.

Numerous researchers have employed hydrological models to investigate rainfall infiltration patterns on slopes and their disaster-inducing mechanisms. In recent years, a “physically based modeling” framework has been widely adopted, in which the unsaturated seepage process is described using the Richards equation (RE), while the van Genuchten (VG) model is employed to characterize the soil water retention curve (SWRC) and hydraulic conductivity functions. Within this framework, numerical methods are used to simulate rainfall-induced variations in water content, pore water pressure responses, and their corresponding effects on the stress field^19–21. With the advancement of seepage–mechanical coupling studies, research focus has gradually shifted from pure seepage calculations to investigating how the temporal evolution of pore water pressure drives the degradation of soil shear strength, and to incorporating seepage results into mechanical models to perform slope stability assessments based on the strength reduction method^22–24. Compared with generalised multiphysics solvers, researchers prefer to use geotechnical engineering–oriented numerical platforms, such as FLAC3D, because they provide well-established fluid–solid coupling strategies and a wide range of constitutive models tailored for geomaterials^25,26. In recent years, some scholars have further developed rainfall infiltration boundary treatment functions and real-time updating functions for unsaturated hydraulic conductivity using the FISH language on the basis of unsaturated seepage analysis, thereby enabling the simulation of seepage processes in both saturated and unsaturated zones of slopes under rainfall conditions, as well as the analysis of slope deformation characteristics^27–29. However, in complex slope stability analyses that account for variations in the seepage field, slope surface deformation and internal stress evolution exhibit significant differences in response to varying rainfall intensities and rainfall durations^30–32. Consequently, it is necessary to conduct a systematic comparison and evaluation of slope stability analysis results that consider seepage processes under different rainfall parameter conditions.

This study investigates the deformation evolution characteristics and formation mechanisms of the Xingfu Dayuan landslide in Yecheng County, Xinjiang Uygur Autonomous Region, China. Detailed field investigations, analysis of slope deformation monitoring data, and numerical simulations were conducted. Time-series deformation information was extracted from the study area using SBAS-InSAR technology to identify anomalous deformation zones and to characterize the spatiotemporal evolution of slope deformation. Based on saturated-unsaturated seepage theory, a numerical model incorporating the hydraulic properties of loess was established to simulate moisture migration within the slope, the evolution of pore water pressure, and the distribution of displacement. The stability of the landslide under different rainfall conditions was then analyzed, and the simulation results were validated using InSAR-derived deformation data. The findings contribute to an improved understanding of the mechanisms and processes of rainfall infiltration–induced landslide failure. They provide a basis for managing the Xingfu Dayuan landslide and are expected to offer technical support and theoretical reference for the precise prediction and scientific prevention of shallow loess landslides within the study area.

Overview of landslides

The landslide area is located in Xihexiu Township, Yecheng County, Kashgar Prefecture, Xinjiang Uygur Autonomous Region, China (76°41′0.67′′E, 36°58′11.90′′N), on the western bank of the Tizinafu River (Fig. 1). It lies within the interior of the Karakoram Mountains and is characterized by a tectonic denudation–dominated middle- to high-mountain geomorphic setting. The slope elevation ranges from approximately 2890–3430 m, with a relative height difference of about 540 m and an average slope angle of 40°. The site is located near the intersection of the Kugongna River Fault and the Mazha–Kangxiwa Fault, where geological structures are complex. The landslide mass mainly consists of residual colluvial silty soil from the Quaternary Holocene to Upper Pleistocene (Q₃–₄^dl), with an average thickness of 4.6 m and a maximum thickness of 7.8 m revealed by boreholes. The sliding bed comprises moderately weathered quartz schist (Pt₁), underlain by Lower Proterozoic quartz schist (Pt₁). The study area exhibits pronounced seasonal climatic variability. Meteorological records from 2022 to 2024 show that precipitation is mainly concentrated between May and September, accounting for most of the annual rainfall, with peak summer precipitation exceeding 200 mm. In contrast, winter precipitation is minimal. The region undergoes a freezing period from approximately November to March, during which temperatures remain generally below 0 °C, resulting in seasonal ground freezing. Such conditions—characterized by concentrated summer rainfall and winter freezing—directly affect rainfall infiltration, pore water pressure evolution, and slope stability. The Xingfu Dayuan landslide poses a serious threat to 37 households with 305 residents. Damage to livestock sheds, vegetable greenhouses, power lines, and roads could result in direct economic losses exceeding 8 million yuan.

Landslide deformation characteristics

Through field investigations, the deformation and failure characteristics of the Xingfu Dayuan landslide were identified. Several roads and buildings are distributed around the landslide area, with a village road located at the slope toe. The landslide exhibits an overall fan-shaped distribution, bounded by gullies on both lateral sides. The rear scarp extends to the junctions with the lateral gullies, while the front margin is defined by the Tizinafu River. The landslide has an approximate length of 730 m in the movement direction, an average width of about 820 m, a planimetric area of 3.5 × 10⁵ m², and a volume of 1.24 × 10⁶ m³, classifying it as a medium-sized landslide based on its scale. According to its deformation characteristics, it can be identified as a traction-type landslide. Traction landslides are typically characterized by progressive retrogressive deformation, tensile cracking in the rear part, and compressive bulging or extrusion at the slope toe. Field investigations reveal well-developed rear scarps and tensile fissures, together with localized soil bulging and deformation near the slope toe, which are consistent with the typical features of traction-type failure.

The deformation zone of the landslide can be subdivided into three distinct areas: HP1, HP2, and HP3, where multiple scarps and cracks are well developed on the surface, as illustrated in Fig. 2a. The average slope angles of the three zones are generally similar, all approximately 40°, indicating that large-scale topographic gradients do not differ significantly among the subzones. The exposed scarps and cracks are mainly composed of silty soil; crack widths range from 0.1 to 0.2 m, and scarp heights vary between 0.3 and 0.6 m. HP1 is characterized predominantly by scarps and is mainly distributed in the rear part of the landslide (Fig. 2b). In the HP2 area, scarps and cracks are developed primarily on the side adjacent to HP1 and are particularly pronounced at the slope toe (Fig. 2c and d). In the HP3 area, scarps and cracks are mainly distributed in the middle–upper part of the slope and at the slope toe, extending through nearly half of the area, with lengths of approximately 100–120 m. Although the slope gradients are comparable among the three zones, differences in deformation characteristics are closely related to variations in fissure development, local stress redistribution, and rainfall infiltration conditions. Multi-episode rainfall infiltration has progressively widened the steep scarps and fissures in the HP2 and HP3 areas, exhibiting lateral development characteristics. The expansion of fissures has severely compromised the integrity of the surrounding soil mass, causing significant damage to roads, buildings, and farmland.

Fig. 2.

Deformation and Failure Characteristics of the Xingfu Dayuan Landslide.

SBAS-InSAR-based deformation characteristics of the Xingfu Dayuan landslide

Analytical methods

The SBAS-InSAR technique, proposed by Berardino et al.³⁴, is a time-series ground deformation monitoring method based on synthetic aperture radar (SAR) data. By generating differential interferograms with short temporal and spatial baselines, SBAS effectively suppresses decorrelation and phase noise, thereby improving deformation retrieval accuracy. Owing to its capability for long-term and continuous deformation monitoring, this technique has been widely applied in the investigation of slow-moving landslides and regional surface subsidence³⁴.

In this study, a total of 86 Sentinel-1A SAR images were used, covering the period from January 2022 to December 2024. The images are evenly distributed over time, with 2–3 acquisitions per month, ensuring sufficient temporal coverage for deformation analysis. The selected monitoring period corresponds to recent years during which newly developed surface cracks were observed in the landslide area, making it necessary to conduct detailed deformation investigation. The dataset includes 28 scenes from 2022, 28 from 2023, and 30 from 2024. All images were acquired in VV polarization using the Interferometric Wide (IW) swath mode, providing a ground-range resolution of approximately 5 m and an azimuth resolution of about 20 m. Precise orbit ephemerides provided by the European Space Agency (ESA) were applied to minimize orbital phase contributions during interferometric processing³³.

For data processing, SAR images were first cropped according to the extent of the study area, and temporal and spatial baselines were calculated. The image acquired on 11 July 2023 was selected as the master image, and all other images were co-registered to it. Differential interferograms were generated using a maximum temporal baseline threshold of 20 days and a maximum perpendicular baseline threshold of 150 m to minimize temporal and geometric decorrelation. Multilooking was applied with a factor of 4 × 1 (range × azimuth) to improve the signal-to-noise ratio. Interferograms with coherence greater than 0.30 were retained for subsequent analysis. Phase unwrapping was performed using the Minimum Cost Flow (MCF) algorithm to convert the wrapped phase into continuous phase values, enabling quantitative deformation retrieval. A 30 m resolution DEM was used to remove the topographic phase component from the interferograms.

Given the arid climatic conditions of the study area, atmospheric water vapor variability is relatively limited compared with humid regions. Therefore, no external atmospheric correction model was applied. Atmospheric phase components were mitigated within the SBAS inversion framework using spatial low-pass and temporal high-pass filtering. Finally, time-series analysis was conducted to derive cumulative displacement and deformation rate maps for the study area³⁵. The overall technical workflow is illustrated in Fig. 3.

Fig. 3.

SBAS-InSAR Technical Process Flow.

Analysis of monitoring results

Based on the deformation monitoring results from 2022 to 2024, most areas within the landslide exhibited measurable deformation, with the most pronounced deformation occurring in the HP2 and HP3 zones, particularly in the central section, as illustrated in Fig. 4. All deformation values reported in this study are line-of-sight (LOS) displacements derived from ascending-track Sentinel-1 observations. In the InSAR results, negative deformation values represent ground movement away from the satellite along the radar LOS direction. Deformation in the HP1 area was relatively minor, consistent with the field investigations. The maximum LOS deformation rate and cumulative LOS displacement in HP1 are approximately − 10 mm/a and − 30 mm, respectively, whereas those in HP2 reach about − 13 mm/a and − 35 mm, and in HP3 about − 18 mm/a and − 54 mm, respectively. The overall trend suggests that the landslide remains active, particularly in the HP2 and HP3 areas, where LOS deformation rates are relatively high.

Fig. 4.

InSAR deformation results. a Average deformation rate diagram (mm/a), b Cumulative deformation diagram (mm).

Deformation characteristics vary among different zones of the landslide, with distinct trends in the upper, middle, and lower parts of the slope, as illustrated in Fig. 5. The central part of HP2 exhibits the largest LOS deformation, while the lower part of HP3 shows the most pronounced displacement signal. Feature points located at the outer margin of the slope toe indicate an apparent uplift tendency relative to the radar LOS, which may reflect the geometric projection of slope movement onto the satellite viewing direction.

Fig. 5.

Deformation Curve Diagram of Deformation Quantity at Characteristic Points at Different Locations of the Xingfu Dayuan Landslide. a Deformation Quantity Curve Diagram for HP1 Deformation Feature Points, b Deformation Quantity Curve Diagram for HP2 Deformation Feature Point, c Deformation Quantity Curve Diagram for HP3 Deformation Feature Point, d Deformation Quantity Curve Diagram for HP Outer Side Deformation Feature Point.

It should be noted that the reported deformation represents only the LOS component of ground motion from the ascending orbit and does not directly correspond to pure vertical or slope-parallel displacement. For landslide kinematics, slope-parallel movement is typically the most physically meaningful component; however, due to the availability of single-track data in this study, no ascending–descending decomposition was performed. Therefore, the measured LOS deformation may underestimate or overestimate the actual parallel displacement of the slope surface. Despite this geometric limitation, the spatial pattern and temporal evolution of LOS deformation remain effective indicators of relative activity within different zones of the landslide.

Triggering factors and formation stages of the Xingfu Dayuan landslide

The deformation results were analysed in conjunction with contemporaneous temperature and precipitation data. For each deformation zone, the displacement time series at the point exhibiting the maximum deformation was selected and compared with monthly rainfall and daily temperature records, as shown in Fig. 6. The data indicate that total rainfall in the landslide area in 2024 was markedly higher than that in the preceding two years, with precipitation primarily concentrated between May and September and a peak monthly rainfall of 232.5 mm. During this interval, an apparent increase in the deformation rate was observed, particularly within the HP3 zone, as highlighted by the red box in Fig. 6.

Fig. 6.

Relationship between cumulative displacement curves of characteristic landslide points and rainfall/temperature.

Regarding temperature effects, from November 2023 to March of the following year, air temperatures remained below 0 °C, leading to freezing of near-surface soil moisture and a relative reduction in deformation rates, as indicated by the yellow box in Fig. 6. With the onset of the thawing period and progressively rising temperatures, cumulative displacement exhibited a gradual acceleration. Overall, the temporal patterns of cumulative displacement generally align with periods of concentrated seasonal rainfall and fluctuations in temperature, suggesting that the interplay of intense rainfall events and pronounced seasonal temperature variations serves as the primary driver for the initiation and reactivation of this landslide.

A comprehensive analysis integrating InSAR-derived deformation data and meteorological records indicates that the Xingfu Dayuan landslide has undergone a distinct evolutionary process driven by the combined effects of freeze–thaw cycles and rainfall. Based on multi-source data acquisition, detailed field investigations, and InSAR observations, the landslide evolution can be classified into three successive stages. Freeze-thaw degradation phase (Phase I): During prolonged natural evolution, primary fissures and joints develop within the original slope body. Under low-temperature and snowfall conditions (Fig. 7b), pore water in the loess undergoes volumetric expansion upon freezing and subsequently thaws during periods of solar radiation or temperature recovery. Repeated temperature fluctuations induce persistent freeze–thaw cycling. As a result, the microstructure of the loess is progressively altered, leading to degradation of its physical and mechanical properties and a reduction in long-term strength^36,37. This process weakens the soil mass and promotes the propagation and widening of pre-existing fractures, ultimately forming more distinct slope cracks that act as early precursors to landslide initiation (Fig. 7c).

Fig. 7.

Diagram of Landslide Evolution Process.

Fracture Development Stage (Phase II): As temperatures gradually rise above 0 °C, the loess is no longer subject to freezing, and pre-existing fractures enter a mechanically weakened state. During periods of concentrated rainfall (Fig. 7d), rainwater infiltrates along the fissures and continuously replenishes pore water, leading to a progressive increase in pore water pressure. This process reduces loess cohesion and degrades the shear strength of the soil, thereby initiating shear deformation within the slope^38,39. Meanwhile, both the scale and depth of the fissures further increase, with new cracks forming upslope of the pre-existing ones and progressively propagating. This evolution enhances the efficiency of rainfall infiltration into the soil mass, further increasing the susceptibility of the slope to landslide occurrence⁴⁰.

The slope instability phase (Phase III): Under the combined effects of persistent rainfall and gravitational loading^41,42, the soil strength within the slope progressively diminishes, accompanied by intensified fissure development at the crown. As these fissures propagate and coalesce to a critical extent, the shear resistance of the soil mass becomes insufficient to counterbalance its self-weight. Consequently, partial or complete slope failure is initiated, ultimately triggering downslope movement and resulting in a landslide disaster (Fig. 7e).

Numerical simulation of landslide at Xingfu Dayuan

Methodology for unsaturated rainfall infiltration and slope stability analysis based on the FLAC3D platform

When performing saturated and unsaturated seepage calculations on the FLAC3D platform, changes in saturation are primarily determined by variations in the fluid volume at each node. The key to unsaturated seepage analysis lies in establishing the relationship between saturation and negative pore water pressure, from which the permeability coefficient of the unsaturated zone can be derived. Subsequently, by dynamically updating the element permeability during the solution process, the unsaturated seepage behavior can be effectively simulated^43,44. For unsaturated soils, van Genuchten proposed a widely used relationship between soil moisture content and negative pore water pressure ⁴⁵:

1

where denotes volumetric water content; denotes residual volumetric water content; denotes saturated volumetric water content; denotes negative pore water pressure; ,,denote fitting parameters of the van Genuchten model, respectively. Based on the relationship between volumetric water content and saturation , the relationship between saturation and negative pore water pressure can be derived:

2

where is Residual saturation. The parameters are a = 100 kPa, =1, =2 and =0.05.

The hydraulic conductivity of unsaturated soils varies continuously with moisture content, and the curve relating moisture content to matric suction constitutes the soil-water characteristic curve (SWCC). Consequently, the hydraulic conductivity can be described as a function of matric suction. When establishing the relationship among hydraulic conductivity, saturation, and matric suction, the SWCC and hydraulic conductivity are mutually derivable. According to Eyo et al. (2022), the relationship between hydraulic conductivity and matric suction presented in Eq. (3) has been widely recognized and adopted by many researchers ⁴⁶.

3

where denotes the hydraulic conductivity varying with matric suction; denotes the saturated hydraulic conductivity of the soil; represents the air pressure within the pores; denotes the pore water pressure; denotes matric suction; denotes the fluid density; and the remainder are fitting parameters.

Equation (4) integrates the effective stress principle with the Mohr-Coulomb failure criterion. Based on the pore water pressure distribution obtained from seepage analysis, it is used to calculate the shear strength of unsaturated soil under rainfall conditions. This equation effectively captures the nonlinear relationship between matric suction and shear strength, thereby improving the understanding of unsaturated soil behaviour. It has been widely applied in slope stability analyses under rainfall conditions^15,47.

4

where is the shear strength of unsaturated soil; denotes effective cohesion; denotes the total normal stress on the shear surface; denotes the pore air pressure, taken as atmospheric pressure in this study; denotes the net normal stress on the shear surface; denotes the effective internal friction angle; denotes the pore water pressure; denotes the matric suction.

Equation (5) represents the reduction coefficient calculated with the strength reduction method. By progressively reducing the shear strength of the material until the slope reaches a critical failure state, this coefficient is used to determine the safety factor of the slope.

5

where and represent the actual cohesion and internal friction angle of the slope body, respectively; and denote the reduced cohesion and internal friction angle, respectively.

The development framework of the slope stability analysis program considering saturated and unsaturated seepage is illustrated in Figure 8 and consists of three main modules:

1. Saturated–Unsaturated Seepage Analysis Module: This module determines the actual degree of saturation based on the relationship between negative pore water pressure and saturation within unsaturated seepage theory (Eq. (2)). It further applies real-time correction of the hydraulic conductivity using the relationship described in Eq. (3). A cyclic updating function is employed to dynamically modify the hydraulic conductivity at each computational time step.

2. Slope Stability Analysis Module: The variation in shear strength of each element during rainfall is calculated using the Mohr–Coulomb failure criterion (Eq. (4)), incorporating effective stress principles that account for slope moisture content, matric suction, internal friction angle, and cohesion. Subsequently, slope stability and cumulative displacement are evaluated using the strength reduction method (Eq. (5)).

3. Boundary Condition Evaluation Module: First, slope surface boundary elements are defined as fixed-flow boundaries, with the infiltration rate taken as the minimum of the real-time infiltration capacity of the slope surface layer, qs, and the rainfall intensity . When the degree of saturation S of a slope surface element reaches 1.0, saturation is assumed to be reached, indicating the onset of surface water accumulation and runoff generation. The corresponding slope surface boundary element is then converted to a fixed-pressure boundary.

Fig. 8.

Development Process for Slope Stability Analysis Programme Considering Saturated and Unsaturated Seepage.

Following this procedure, the Fish language is implemented within FLAC3D to numerically simulate unsaturated seepage processes, slope deformation evolution, and stability behavior under rainfall loading.

In this study, the strength reduction method (SRM) was adopted to evaluate slope stability under rainfall infiltration conditions. This method was selected because it effectively couples seepage-induced stress redistribution with progressive strength degradation within a fully numerical framework. Compared with traditional limit equilibrium methods (LEM), which require predefinition of potential slip surfaces and assume simplified interslice force distributions, the SRM implemented in FLAC3D does not require prior assumptions regarding the failure surface geometry. Instead, failure surfaces develop automatically as a result of stress–strain evolution during numerical calculation process.

This advantage is particularly important for rainfall-induced loess landslides, where infiltration leads to spatially heterogeneous pore water pressure redistribution, nonlinear matric suction reduction, and progressive weakening of shallow soil layers. Under such conditions, potential failure surfaces may evolve dynamically rather than follow predefined circular or planar geometries typically assumed in LEM. The SRM allows continuous updating of shear strength parameters within each element while simultaneously accounting for unsaturated–saturated seepage transition, making it more suitable for capturing the coupled hydro-mechanical behavior of loess slopes during rainfall events.

Calculation model condition settings

1. Geotechnical layer calculation parameters

Based on the geological cross-section A–A′ of the Xingfu Dayuan landslide, a 2D plane strain model embedded in FLAC3D was adopted for the numerical simulation, with the geometric thickness of the out-of-plane direction (Y-axis) set to 1 m. The slope body can be sequentially divided from top to bottom into Quaternary Holocene-Upper Pleistocene residual-colluvial silt () and moderately weathered quartz schist. Between the surficial silt layer and the underlying bedrock, a relatively thick weathered bedrock zone is developed, as shown in Fig. 9. On the basis of published geological survey reports and previous research conducted in the study area, the physical and mechanical parameters of the rock–soil mass were determined and are summarized in Table 1.

Fig. 9.

Stratigraphic Distribution of the A-A′ Section at the Xingfu Dayuan Landslide.

Table 1.

Geotechnical Parameter Values.

Materials	Density/ (kg·m-3)	Internal friction angle/ (°)	Cohesion /(kPa)	Permeability coefficient /(m·s-1)	Volumetric water content /(%)	Elastic modulus /kPa	Porosity /n	Poisson’s ratio
Loamy soil	1800	28	2.5 × 101	5 × 10–6	15	8 × 103	0.45	0.32
Moderately weathered quartz schist	2500	40	1 × 103	1 × 10–7	1.5	1 × 107	0.05	0.28
Quartz schist	2650	45	4 × 103	1 × 10–8	0.05	6 × 107	0.01	0.26

• (2)Initial and boundary condition settings

Given the arid climatic conditions of the study area, groundwater occurs at considerable depth, and the landslide body is relatively thin; therefore, groundwater has a negligible influence on slope stability. Rainfall is regarded as one of the dominant factors controlling slope stability, and variations in groundwater level were thus neglected in this study. Field measurements using tensiometers installed in three slope sub-zones (HP1–HP3) indicate a matric suction of approximately − 40 kPa at a depth of 3 m below the slope surface. The measurements were conducted three times in August, and the results showed no significant differences among the three sub-zones; therefore, the average value of −40 kPa was adopted as the initial pore pressure for the surficial silt layer and the weathered fractured zone. The underlying quartz schist contains an extremely limited amount of pore water, and the effect of hydraulic pressure on stability is therefore minimal; accordingly, an initial pore pressure of – 5 kPa was assigned to this layer.

The numerical model was constructed along cross-section A–A′ as a two-dimensional plane-strain model embedded in the FLAC3D platform, with a unit thickness assigned in the out-of-plane direction to ensure plane-strain conditions. Mechanical boundary constraints were applied as follows: the bottom boundary was fixed in both the horizontal and vertical directions, the two lateral boundaries were fixed in the horizontal direction while remaining free in the vertical direction, and the slope surface was left free. Boundary conditions for seepage were defined according to the workflow depicted in Fig. 8. When surface soil elements remain unsaturated, rainfall boundaries operate as fixed-flow boundaries, with the infiltration rate governed by rainfall intensity, and corresponding parameters were adjusted for different simulation scenarios. Once saturation is reached, the rainfall boundaries are converted to fixed-pressure boundaries, with the applied pressure related to the depth of surface water; in this study, the minimum value was adopted.

To monitor the seepage effects of rainfall at the Xingfu Dayuan landslide, a series of grid nodes at different depths were selected as monitoring points along a vertical slope surface at the lower part of the slope. Adjacent monitoring points were spaced at 1 m intervals, with the deepest point located 8 m below the slope surface. The detailed configuration of the monitoring points is illustrated in Fig. 9.

• (3)Rainfall scenario configuration

The classification of rainfall intensity defined by the National Meteorological Administration is shown in Table 2. According to meteorological records from Xihexiu Township, where the Xingfu Dayuan landslide is located, the annual average rainfall is approximately 391 mm. Over the past five years, short-duration heavy rainfall events exceeding 25 mm/h have been recorded, with the maximum 24-h cumulative rainfall reaching approximately 200 mm. In addition, the longest documented intermittent rainfall duration was approximately four consecutive days. These statistical characteristics provide the basis for the parameter selection in the numerical simulations. Considering that rainfall intensity, duration, and cumulative amount jointly influence the seepage field evolution and slope stability, two representative rainfall scenarios were established to systematically investigate the landslide response under different rainfall patterns.

Table 2.

Classification of Rainfall Levels.

Rainfall Category	Light rain	Moderate rain	Heavy rain	Intense rain	Very heavy rain	Extreme heavy rain
24-h rainfall/mm	≤ 9.9	10.0–24.9	25.0–49.9	50.0–99.9	100–249.9	≥ 250

Scenario 1 assumes a fixed rainfall duration of 48 h, with rainfall intensities of 20, 50, 100, 200, and 400 mm/d. The lower intensity values correspond to commonly observed heavy rainfall events based on historical records and the national rainfall classification standard, while the higher intensities are designed to represent extreme or hypothetical rainfall conditions. This configuration enables evaluation of the sensitivity of slope stability to rainfall intensity under a controlled duration. Scenario 2 assumes a constant cumulative rainfall of 200 mm, corresponding to the maximum recorded 24-h rainfall in recent years, distributed over durations of 4, 2, 1, and 0.5 days. This design allows examination of the influence of rainfall concentration on infiltration dynamics and slope stability evolution, while maintaining a consistent total water input.

These parameter settings are therefore derived from local meteorological statistics combined with standardized rainfall classifications and are intended to provide a comparative assessment of landslide response under both typical and extreme rainfall conditions. Furthermore, to account for potential delayed effects of rainfall infiltration, the seepage field and slope stability were continuously monitored for a period following rainfall cessation, in order to clarify the temporal coupling between rainfall processes and slope instability development.

Spatiotemporal evolution characteristics of seepage fields in slopes

The spatial distribution of pore water pressure provides an effective indicator of rainfall infiltration processes. Continuous 48-h simulations of the Xingfu Dayuan landslide under different rainfall intensities (20, 50, 100, 200, 400 mm/d) revealed the evolution of pore water pressure with depth, as depicted in Fig. 10. During rainfall (t = 0–48 h), pore water pressure at the slope surface rose rapidly, with the rate of increase significantly intensifying as rainfall intensity and duration increased. With increasing depth, the pore pressure response gradually weakened, exhibiting a pronounced attenuation pattern. Under the extreme rainfall condition of 400 mm/d, pore water pressure reached – 7.85 kPa within 12 h after the onset of rainfall and peaked at approximately 1 kPa by the end of rainfall (t = 48 h), indicating that the surface layer of the slope had reached or was close to saturation. Twenty-four hours after rainfall cessation, as infiltrated rainwater continued to migrate downward, pore water pressure at the slope surface decreased to – 9.24 kPa, indicating a transient process of positive pore pressure dissipation within the shallow slope profile.

Fig. 10.

Variation of pore water pressure over time at different depths within the lower section of the slope under Operating Condition 1.

As rainfall continues, the rise in pore water pressure gradually propagates from the slope surface towards the interior, with the infiltration depth exhibiting a significant positive correlation with rainfall intensity. Under low-intensity rainfall (RI = 20 mm/d), significant changes in soil pore water pressure were observed within approximately 2 m of the slope surface after 12 h of continuous precipitation, and extended to depths of about 4 m below the slope crest after 48 h. In contrast, under extreme rainfall conditions (RI = 400 mm/d), the pore water pressure response within the slope was both more rapid and penetrated substantially deeper into the slope. After 12 h of rainfall, changes in pore pressure had already affected depths of 3.4 m below the slope surface. With continued rainfall for 48 h, the influence extended to a maximum depth of 8 m. These results indicate that intense rainfall promotes faster infiltration and deeper penetration pathways within the slope.

Following the cessation of 48-hour continuous rainfall, pore water pressure within the slope did not immediately return to a stable state. Indeed, a significant adjustment process was observed along the depth profile within 24 h after rainfall, indicating a clear lag effect associated with rainfall infiltration. Within 48 h after rainfall cessation, the infiltration depth continued to increase under all rainfall intensities. Specifically, for rainfall intensities of 20, 50, 100, 200, and 400 mm d^–1, the infiltration depths increased from approximately 4, 4.5, 5.5, 6.5, and 7.5 m at the end of rainfall to about 5, 5.5, 6.5, 7.5, and 8 m, respectively. Taking the extreme rainfall of 400 mm/d as an example, at the end of rainfall (t = 48 h), the maximum pore water pressure occurred at the slope surface, reaching 1 kPa; 12 h after rainfall cessation, the position of maximum pore pressure shifted downwards to approximately 1 m below the slope surface, with the pressure decreasing to – 3.8 kPa. By 24 h after rainfall, the maximum pore pressure had further migrated to a depth of approximately 2 m, corresponding to a value of – 5.2 kPa. This process demonstrates that after rainfall ceases, infiltrated water continues to migrate downward within the slope, resulting in a spatial redistribution pattern where peak pore pressure transitions from shallow to deep zones.

Figure 11 illustrates the temporal evolution of pore water pressure along the A-A' profile of the Xingfu Dayuan landslide under a torrential rainfall intensity of 200 mm/d. Compared with the initial state, after 48 h of continuous rainfall, the pore pressure disturbance zone expanded to a depth of approximately 6.5 m below the slope crest. At the rear of the slope (approximately 75 m in the horizontal direction), the seepage front advanced further toward the bedrock interface, forming a distinct transient saturation zone characterized by positive pore water pressure in the upper slope layer. Within 24 h after rainfall cessation, the transient saturated zone near the slope surface gradually dissipated. However, the zone influenced by elevated pore pressure continued to propagate into the interior of the slope, indicating that the slope remained in a lagged response phase dominated by ongoing infiltration. Overall, rainfall-induced pore pressure redistribution exhibited significant hysteresis, resulting in the persistence of seepage disturbances long after precipitation ceased.

Fig. 11.

Spatial distribution of pore pressure within the slope at different time points under 200 mm/d rainfall intensity.

Infiltration simulations were conducted for the Xingfu Dayuan landslide under total rainfall of 200 mm, with rainfall intensities of 50, 100, 200, and 400 mm/d, corresponding to rainfall durations of 4, 2, 1, and 0.5 d, respectively. The evolution of pore water pressure with depth in the lower part of the slope is shown in Fig. 12. During the 0.5-day rainfall period, pore pressure at the slope surface increased rapidly under all rainfall intensity conditions, with higher intensities producing more pronounced increases. Taking the 400 mm/d scenario as an example, the surface pore water pressure rose sharply from an initial value of – 40 kPa to approximately – 7.8 kPa. By 1.5 d, when rainfall at 400 mm/d ceased for 1 d, the combined effects of rainfall cessation and continued downward infiltration caused the surface pore pressure to decline from its peak of – 7.8 kPa to – 19.5 kPa. The peak pore water pressure gradually migrated downwards to a depth of approximately 1 m, indicating a pronounced delayed response. Under the other rainfall scenarios with different intensities, the pore water pressure at the slope surface remained in an ascending phase, with its influence progressively extending to greater depths over time.

Fig. 12.

Variation of pore water pressure over time at different depths within the lower section of the slope under Operating Condition II.

It is noteworthy that when rainfall persisted for 2.5 days, the 200 mm/d rainfall scenario had already ended after 1.5 days. Although pore water pressure at the slope surface began to decline, pressures within the zone approximately 3.5 m below the slope surface remained higher than those under the 400 mm d^–1 scenario. This indicates that the 400 mm/d rainfall continued to influence the slope seepage field for nearly 2 days after rainfall cessation, further confirming the lag effect of rainfall. After 3.5 days, rainfall under all three higher-intensity scenarios (400, 200, and 100 mm d^–1) had ceased. The pore water pressure trends under these conditions gradually converged, characterized by a continuous decrease in surface pore pressure while the maximum pore pressure shifted to the interior of the slope. After 4 days, under the 50 mm/d rainfall scenario, surface pore water pressure increased slowly and stabilized at approximately −20 kPa, consistent with the tendency for hydraulic conductivity to decrease with increasing water content. The depth distribution of pore water pressure within the slope is controlled simultaneously by rainfall intensity and duration. During rainfall lasting one day, the depth affected by pore water pressure increases significantly with rising rainfall intensity. By the fourth day, pore water pressure profiles under the 400, 200, and 100 mm/d had largely converged, whereas the 50 mm/d scenario remained in the active rainfall phase. Notably, within the shallow layer approximately 2 m below the slope crest, the pore water pressure under the 50 mm/d condition remained higher than that under the other three rainfall intensities. This indicates that, for the same total rainfall, a longer rainfall duration sustains elevated pore water pressure in the near-surface layer for a longer period, thereby increasing the likelihood of shallow-layer instability.

Response patterns of slope deformation fields under rainfall

Figure 13a, c and e depict slope displacement under different rainfall intensities for an identical rainfall duration, clearly illustrating the deformation zones of the Xingfu Dayuan landslide. As rainfall intensity increased from 20 mm/24 h to 100 mm/24 h, the average displacement in the middle-lower deformation zone rose from 14.01 mm to 74.23 mm, while that in the upper deformation zone increased from 8.23 mm to 31.29 mm. The increase in rainfall intensity led to a significant rise in slope deformation, with displacement growth rates approaching 429% in the strongly deformed area near the slope toe.

Fig. 13.

Distribution of slope displacement and equivalent plastic strain under varying rainfall intensities for the same rainfall duration (a and b: RI = 20 mm/24 h, c and d: RI = 50 mm/24 h, e and f: RI = 100 mm/24 h).

Figure 13b, d, and f present the distributions of equivalent plastic strains under different rainfall intensities for the same rainfall duration, highlighting distinct potential failure zones. The numerical simulation results indicate that the critical slip plane lies at the interface between the surface silty soil and the underlying bedrock, demonstrating that the failure mechanism of the Xingfu Dayuan landslide is primarily controlled by the bedrock-soil interface. Under the rainfall intensity of 20 mm/24 h, the potential failure surface converged at a horizontal distance of approximately 240 m (Fig. 13b). In contrast, under a rainfall intensity of 100 mm/24 h, the potential failure surface propagated towards the slope toe, converging at a horizontal distance of 300 m (Fig. 13f). These comparisons reveal that increasing rainfall intensity significantly modifies the extent and scope of slope failure in the upper and middle sections. Notably, in the lower part of the slope, the potential slip surfaces exhibit similar extents under different rainfall intensities (Fig. 13b, d and f). This indicates that although rainfall reduces overall slope stability, rainfall infiltration exerts a limited influence on slip-surface geometry and failure patterns in this zone.

Figure 14a, c and e illustrate slope displacement under identical rainfall quantities but different rainfall durations, clearly demonstrating variations in the potential deformation range and failure severity. The simulated primary deformation zones are mainly concentrated within two horizontal ranges: 160–280 m and 380–520 m. Displacement near the slope toe is greater than that in the upper slope sections. Under the same total rainfall, shortening the rainfall duration from 4 days to 1 day decreased the average displacement in the middle-lower deformation zone from 72.71 mm to 22.93 mm, while the average displacement in the upper deformation zone decreased from 28.02 mm to 15.18 mm. The shorter rainfall duration limited water infiltration into the slope body, resulting in a substantial reduction in overall deformation, with the middle and lower deformation zones exhibiting a more pronounced response. Under a rainfall duration of 4 days, the failure zone propagated upwards from the potential slip surface, with some areas extending to the slope surface, as illustrated in Fig. 14b. Comparison of the equivalent plastic strain distributions in Fig. 14b, d and f indicate that, for the same total rainfall, longer rainfall duration promotes deeper and more extensive water infiltration into the slope body, leading to more severe damage in the middle and lower slope sections. Unlike the effects of rainfall intensity, an increase in rainfall duration simultaneously modifies both the extent of the potential slip surface and the degree of failure in the upper-middle slope and the areas near the toe.

Fig. 14.

Distribution of slope displacement and equivalent plastic strain under different rainfall duration conditions for the same rainfall intensity (a and b: RD = 4 d, c and d: RD = 2 d, e and f: RD = 1 d).

This study employs FLAC3D to conduct a transient rainfall-seepage analysis of the Xingfu Dayuan landslide. Based on this analysis, the deformation evolution process was simulated and compared with time-series InSAR monitoring results, with particular emphasis on the spatial distribution of significant deformation within the landslide. The simulated primary deformation zones are predominantly distributed within the horizontal ranges of 160–210 m and 380–520 m. Figure 15 presents deformation monitoring data along the A–A′ profile of the Xingfu Dayuan landslide. Significant displacement (>12 mm) mainly occurred within the horizontal ranges of 140–200 m and 420–520 m, whereas displacements at both the rear and frontal margins of the slope remained below 8 mm.

Fig. 15.

InSAR Monitoring Results Along the A-A' Profile of the Xingfu Dayuan Landslide.

Although a direct quantitative validation cannot be performed due to the absence of independent ground observations, the strong spatial correspondence between simulated deformation zones and InSAR-derived deformation patterns along the A–A′ profile indicates a high degree of consistency in both location and relative magnitude. The numerical model successfully reproduces the concentration of deformation in the central HP2 and lower HP3 zones while maintaining low deformation at the rear and frontal margins. This agreement in spatial distribution and relative deformation intensity supports the reliability of the coupled rainfall–seepage–stability modeling framework adopted in this study.

Slope stability analysis considering rainfall infiltration processes

Figure 16 illustrates the variations in landslide stability coefficients under rainfall intensities of 20, 50, 100, 200, and 400 mm/d. During the first 48 h, all five scenarios remain within the rainfall phase. For the same rainfall durations, higher rainfall intensity results in greater variations in pore pressure magnitude and range along the slope depth, thereby reducing landslide stability. Similarly, under identical rainfall intensities, longer durations promote deeper infiltration and greater pore pressure fluctuations, further decreasing slope stability. Notably, the stability coefficient does not reach its minimum immediately at the end of the 48-h rainfall period. Owing to the lag effect of rainfall infiltration, water continues to migrate into the slope after precipitation ceases, progressively deepening the seepage disturbance zone. Within 16 h after rainfall cessation, the stability coefficients for rainfall intensities of 20, 50, 100, and 200 mm/d decreased continuously from 1.118, 1.104, 1.082, and 1.044 to 1.098, 1.078, 1.049, and 1.011, respectively, as shown in Fig. 16a. Thereafter, as time progressed, the stability coefficients remained relatively stable, with the slope generally maintaining a condition ranging from marginal instability to basic stability. However, under the extreme rainfall intensity of 400 mm/d, the stability coefficient was approximately 1.01 at the end of rainfall, indicating a near-critical state. Following rainfall cessation, the stability coefficient continued to decline due to the delayed infiltration effect, dropping to 0.964 within 24 h, at which point the slope transitioned into an unstable condition.

Fig. 16.

Effect of rainfall on landslide stability.

When the total rainfall remains constant, rainfall intensity and duration still exert a certain influence on the landslide stability coefficient. With the total rainfall fixed at 200 mm, rainfall durations were set at 4, 2, 1, and 0.5 days, respectively. As shown by the blue curve in Fig. 16b, the slope stability coefficient under no rainfall conditions was 1.153, indicating a stable state. Once rainfall commences, the stability coefficient decreases rapidly. In general, the longer the rainfall duration, the lower the stability coefficient at the end of the rainfall event. The reasons can be explained as follows. During shorter rainfall periods, by the end of the rainfall, infiltrating water has not fully penetrated into the slope body. Instead, a substantial proportion of rainfall either accumulates on the slope surface or is discharged as surface runoff. Consequently, the reduction in the slope stability coefficient is limited. In contrast, under lower rainfall intensities combined with longer durations, rainfall can infiltrate sufficiently into the slope body when the rainfall intensity remains lower than the soil permeability. For instance, as illustrated in Fig. 11h, under a rainfall intensity of 50 mm/d, the infiltration influence extends to a depth of approximately 6 m below the slope surface, leading to a pronounced reduction in the slope stability coefficient.

However, as indicated by the red curve in Fig. 16b, by the end of the fourth day, when the total precipitation has largely infiltrated into the slope body, rainfall intensity continues to exert a significant weakening effect on the landslide stability coefficient. Nevertheless, the magnitude of this effect becomes relatively small. For example, the difference in stability coefficients between rainfall intensities of 100 mm/d and 400 mm/d is only about 0.008, which is far smaller than the difference of 0.035 observed at the end of the rainfall period. As depicted in Fig. 11h, which shows the distribution of pore water pressure within the slope at t = 4 days, the infiltration process is essentially complete under rainfall intensities of 100, 200, and 400 mm/d. The pore water pressure profiles with depth are broadly similar under these intensities, resulting in a convergence of the corresponding stability coefficients. In contrast, at a rainfall intensity of 50 mm/d, pore water pressure below approximately 2 m from the slope surface remains lower than that under the other intensities, indicating that rainwater infiltration continues immediately after rainfall cessation. Once infiltration is largely complete, the stability coefficient values for this rainfall intensity will largely converge with those observed under the other three intensities.

Analysis of slope stability considering rainfall infiltration processes indicates that rainfall-induced landslide responses typically exhibit a pronounced time-lag effect, whereby slope failure or accelerated displacement often occurs after the peak rainfall or cumulative precipitation has already passed. Previous studies suggest that this delay primarily arises from the combined effects of gradual pore water pressure accumulation, seepage propagation in unsaturated soils, and the inherent soil structure and initial moisture conditions. Through transient seepage and slope stability analyses, Li et al. (2024) demonstrated that infiltration duration and the delayed development of pore water pressure establishment are closely related to the initial moisture content of the soil⁴⁸. Bhadiyadra et al. (2024) further emphasised that the accumulation and lag of pore water pressure under different slope conditions directly influence the triggering time and failure process of landslides⁴⁹. Consequently, the lag effect of rainfall is governed by multiple factors, including rainfall characteristics, soil hydraulic properties, slope structure, and initial moisture conditions, holding significant implications for landslide prediction and early warning systems^50–52. In the present study of the Xingfu Dayuan landslide, pore water pressure within the slope was observed to propagate progressively from shallow layers to greater depths as rainfall duration increased. Notably, variations in pore water pressure along the slope profile persisted for up to 24 h after rainfall cessation. Under the influence of continued downward infiltration, the maximum pore water pressure was not located at the surface but instead developed at a certain depth within the slope, further illustrating the delayed response of the landslide system to rainfall infiltration.

Discussion

In future work, long-term continuous monitoring could be implemented to obtain higher-precision environmental data, which would facilitate quantitative analysis of the relationship between environmental factors and slope displacement. In addition, incorporating real measured rainfall sequences in simulations would allow for more accurate representation of natural rainfall variability and its impact on slope stability, complementing the current constant-intensity rainfall scenarios.

In recent years, significant progress has been made in understanding slope deformation and failure mechanisms by integrating advanced monitoring techniques with physical model tests. For excavation-induced instability, Fang et al. used scaled physical models with multi-field monitoring—including 2D surface displacement, internal strain, and soil pressure sensors—to analyze deformation evolution and failure patterns. Their study revealed the formation of yield zones, static zones, and triangular shear planes, providing experimental evidence for slope failure processes⁵³.

For rainfall-induced failure, Jia et al. conducted centrifuge tests on multi-step loess slopes under high acceleration (approximately 50 g) with intermittent heavy rainfall. Equipped with laser displacement sensors, high-resolution cameras, soil pressure, and volumetric water content sensors, these experiments captured the coupled effects of infiltration front migration and crack development on failure evolution, offering new insights into complex rainfall–crack–failure mechanisms⁵⁴.

Recent studies have also explored the influence of water–soil interactions and multi-physical coupling. Centrifuge experiments on reservoir slopes showed that rapid water-level changes combined with rainfall can significantly alter pore pressures and trigger local instabilities⁵⁵. These findings highlight that combining multi-field monitoring with physical and centrifuge modeling allows comprehensive characterization of slope responses under complex environmental conditions, enhancing our understanding of deformation and failure mechanisms and supporting stability assessment and early warning strategies.

Conclusions

Based on SBAS-InSAR technology and numerical simulation methods, deformation monitoring and stability analyses of the Xingfu Dayuan landslide were conducted from 2022 to 2024. The evolution of pore water pressure and slope stability under different rainfall intensities and durations was systematically investigated. The main conclusions are as follows:

1. Deformation within the Xingfu Dayuan landslide is most pronounced in the lower part of the HP3 zone, where both cumulative displacement and deformation rates are highest, indicating that the landslide is still actively deforming. Distinct deformation characteristics are observed in the upper, middle, and lower portions of the slope, and soil uplift has been identified along the outer margin of the slope toe. The lower HP3 zone is immediately adjacent to a residential community and lies within 100 m of nearby roads, indicating that continued deformation may pose direct risks to infrastructure and public safety. In addition, landslide deformation intensified from 2023 to 2024, with deformation rates accelerating during the summer and autumn seasons, which are characterized by higher rainfall and elevated temperatures.

2. Under identical rainfall duration conditions, higher rainfall intensity results in a more pronounced increase in pore water pressure within the slope, deeper infiltration, and, in some cases, temporary saturation at the slope surface. After rainfall cessation, pore water pressure does not dissipate immediately but exhibits a pronounced lag effect, continuing to influence the internal seepage field. For the same total rainfall amount, longer rainfall duration sustains elevated pore water pressure in the near-surface layer for a longer period, thereby increasing the likelihood of shallow-layer instability.

3. Landslide stability is strongly controlled by both rainfall intensity and duration. Under identical rainfall durations, higher rainfall intensity produces greater pore water pressure accumulation, deeper infiltration, and consequently lower slope stability. For a constant rainfall intensity, prolonging the rainfall duration enhances infiltration depth and exacerbates stability degradation, with the stability coefficient continuing to decrease for a certain period after rainfall ceases due to the lag effect of seepage. When the total rainfall amount is the same, longer-duration rainfall results in poorer overall stability, whereas shorter rainfall events tend to generate surface runoff, thereby limiting water infiltration into the slope body and maintaining relatively higher stability.

4. Rainfall and temperature variations are key factors controlling landslide deformation. Rainfall infiltration increases pore water pressure and reduces matric suction and soil shear strength, thereby promoting slope deformation. Rainfall-induced landslide responses exhibit a clear lag effect, the magnitude of which is closely related to rainfall characteristics, soil hydraulic properties, and slope structure. In addition, soil moisture fluctuations and porosity increases associated with freeze–thaw cycles significantly affect landslide stability, accelerating deformation during thawing periods.

5. Field observations and numerical simulation results indicate that shallow failure represents the dominant failure mode of loess landslides in the Yecheng area. Rainfall infiltration primarily affects the near-surface soil layer, where rapid strength reduction and pore water pressure accumulation promote shallow sliding rather than deep-seated instability. This characteristic reflects the structural and hydraulic properties of local loess and constitutes a key geological feature of landslides in southern Xinjiang.

Author contributions

Z.T. conceived the research, conducted experiments using software, carried out fieldwork, and drafted the manuscript. Z.Z. contributed to the research methodology design, provided resource support, supervised the study, and secured funding. Z.L. participated in the design and execution of the experiments. X.Y. conducted fieldwork and organized and analyzed data. K.S. organized and analyzed data. Y.Z. participated in data analysis and fieldwork. Y.P. conducted experiments using software and assisted with visualization. All authors reviewed and approved the final manuscript.

Funding

This research was generously funded by the Tianshan Talent Scientific Research Project of Xinjiang Uygur Autonomous Region (No. 2023TSYCCX0010), the Natural Science Foundation of Xinjiang Uygur Autonomous Region (No. 2025D01C254), and the National Natural Science Fund (No. 42367021).

Data availability

The datasets used and/or analysed during the current study available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors declare no competing interests.