Comparative Analysis of Finite Element and Discrete Element Methods for the Deformation and Failure of Embankment Slope

Abstract

The finite element method (FEM) and discrete element method (DEM) have been widely applied to analyze the deformation and failure processes of embankment slopes. Although both methods can produce promising results, the choice between them has long remained unresolved. In this study, a failure case of a granite residual soil (GRS) embankment was analyzed. FEM and DEM models were established to simulate the instability process of this embankment slope, and the applicability of both methods to GRS embankments was then evaluated. The main conclusions are as follows: (1) Geotechnical parameters of GRS were determined through laboratory testing, and FEM and DEM models were developed to reproduce the deformation and failure behavior of the embankment slope subjected to rainfall and vehicle loading. (2) Similar rainfall infiltration patterns were obtained from both FEM and DEM simulations; however, significant differences in deformation were observed. The FEM-predicted deformation was 0.075 m after rainfall, indicating that the embankment remained stable. In contrast, the DEM-predicted deformation reached 1.4 m, indicating that the embankment slope had already become unstable. (3) The DEM simulation closely reproduced the failure of the GRS embankment slope observed in the field. It realistically captures the process of particle disintegration in GRS caused by rainfall infiltration, as well as the subsequent slope collapse. Therefore, DEM can be regarded as the most appropriate approach for modeling the instability of GRS embankment slopes.

1. Introduction

Granite residual soil (GRS), formed by the long-term weathering of granite, represents a distinctive soil type featuring a loose fabric, high porosity, and densely developed fissures [1]. Owing to the low cementation and weak particle bonding of GRS, the soil is highly susceptible to disintegration when exposed to water, showing strong collapsibility and extremely poor water stability [2,3]. Consequently, the GRS exhibits limited resistance to erosion and scouring. Under the combined effects of rainfall infiltration and gravity, structural loosening and localized instability of GRS slopes are likely to occur [4]. Repeated wetting-drying cycles are found to intensify the leaching of cementing materials, resulting in reduced cohesion and enhanced crack propagation, thereby accelerating the deterioration and instability of slopes [5]. In southern China, particularly in Guangdong Province, GRS is widely distributed. Under intense rainfall, slopes commonly develop into “collapse gullies” [6], constituting a characteristic pattern of soil erosion and slope hazard.

The deformation and instability of slopes have remained key issues of interest to geotechnical engineers, and extensive research has been conducted in this field. In early studies, the Limit Equilibrium Method (LEM) was primarily adopted, where the factor of safety (FS) was determined through the equilibrium of forces or moments [7]. Because of its simplicity, clear physical meaning, and ease of parameter acquisition, LEM continues to serve as a standard stability evaluation method during engineering design [8]. However, the assessment of slope stability using LEM generally relies on several implicit assumptions or simplifications. Such as (1) the material along the potential slip surface is assumed to reach the limiting equilibrium simultaneously, In reality, slope failure typically develops progressively. (2) The shape of the slip surface must be predefined, although its actual location and geometry are often unknown prior to failure.

With the rapid development of computer technology and numerical algorithms, numerical analysis methods have gradually become essential tools for investigating slope stability. Among these, the finite element method (FEM) has been widely adopted for its capability to simultaneously determine the slip surface and the factor of safety. Compared with the traditional LEM, FEM does not require a predefined slip surface [9]. Instead, it constructs a continuum model that integrates the soil’s stress–strain relationship and elasto-plastic constitutive behavior to directly solve for the internal stress and strain distributions [10], enabling slope stability evaluation. This method can comprehensively account for complex factors such as stress redistribution, nonlinear deformation, and fluid-solid coupling [11], thereby providing a more realistic representation of geological processes.

In addition to FEM, the discrete element method (DEM) is another numerical simulation approach that treats soil as a discontinuous medium composed of numerous particles. By defining contact models, friction-slip laws, and bonding strengths, DEM enables direct computation of particle movement and failure [12]. Compared with FEM, DEM can more naturally simulate processes such as crack initiation, force-chain reconstruction, and particle sliding [13]. In recent years, extensive research on slope deformation has been conducted using DEM. For instance, Gao et al. [12] adopted the DEM to analyze the dynamic sliding process of a landslide in Baidian Township, Hunan Province, elucidating the structural characteristics, deformation behavior, and stability evolution of soil-rock mixture slopes. Hu and Lu [14] employed the DEM to investigate the deformation and failure mechanisms of soil-rock mixture slopes, revealing that rock content, shape, and spatial distribution influence slope bearing capacity, shear band evolution, and micro-mechanical behaviors such as force chains and anisotropy.

In summary, LEM, FEM, and DEM each have distinct characteristics and application scopes [15,16,17,18]. LEM is simple to compute but lacks process-based description; FEM can model continuous deformation but struggles to capture particle-level failure; DEM, though computationally intensive, can reveal the true evolution of discontinuous media. Both FEM and DEM have been widely applied to analyze the deformation and failure processes of embankment slopes, and both have produced favorable outcomes in engineering applications [19,20,21]. For example, Wang et al. [22] conducted a comparative evaluation of numerical techniques for modeling the fracture of brittle materials and found that FEM can capture the ductile deformation response, while DEM is suitable for representing brittle cracking and the mechanical behavior of interfaces, providing a more accurate description of the overall fracture process. Fattahi et al. [23] investigated the response of unreinforced masonry walls to dynamic loading induced by underground explosions and suggested that, although both FEM and DEM can model such highly disturbed geomaterial behavior, DEM provides more realistic predictions of displacement and strain. However, few studies have focused on comparing FEM and DEM for slope deformation, and the most suitable method for accurately representing slope instability remains controversial.

In the case of GRS slopes, the FEM has been predominantly used to model rainfall-induced deformation [24,25,26] (Bravo-Zapata et al., 2022; Yan et al., 2023; Xu et al., 2023). Nevertheless, hydro-erosional processes observed during rainfall indicate that GRS slope failure is inherently discontinuous. This raises substantial uncertainty regarding the ability of FEM to accurately reproduce the failure process. Therefore, a failure case of a GRS embankment slope is selected as the research subject. Typical geotechnical parameters were obtained through field investigation and laboratory testing, and FEM and DEM numerical models were constructed to simulate the deformation, seepage, and failure evolution of GRS embankment slope under rainfall. By analyzing the differences in deformation characteristics and failure modes predicted by the two methods under rainfall and comparing the simulation results with the field-observed failure features, the applicability and limitations of FEM and DEM for analyzing deformation and failure of GRS embankment slopes were evaluated.

2. Materials and Methods

2.1. Study Area

At approximately 01:57 a.m. on 1 May 2024, the Mei-Da embankment experienced a catastrophic collapse. The incident was triggered by prolonged rainfall, leading to the failure of a GRS embankment and resulting in 52 fatalities. The disaster occurred along the eastern extension of the Meizhou-Dapu Highway in Dapu County, Meizhou City, Guangdong Province (Figure 1), with the collapse center situated at 116°40′23.9″ E and 24°29′24.6″ N.

The geological configuration of the embankment was determined through examination of literature, on-site photographs, and Google terrain data. The embankment slope was approximately 30° with a height of around 30 m, and it was a typical semi-cut-and-fill embankment composed of GRS; in other words, a typical embankment built by partially cutting into the natural slope and partially adding fill material.

2.2. Direct Shear Test

The test soil samples were obtained from an area near Dapu County, Meizhou City, Guangdong Province. The sampling location is close to the landslide site, which can ensure the reliability of the simulation results. Highly weathered GRS with high clay content and a partially retained structure was collected through field investigation and sampling. The samples were collected using borehole drilling to a maximum depth of 22 m. To minimize disturbance, the samples were immediately sealed after extraction. The particle size of the tested soil ranged from 0.001 to 5 mm, and the particle size distribution is shown in Figure 2. The black dots in the figure represent the measured values of particle size distribution, while the white dots indicate the particle diameters corresponding to the cumulative mass percentages of 30%, 50%, and 60%. Two types of samples, undisturbed and remolded, were prepared. They represent the natural and fill soils of the Meida expressway embankment slope, respectively. The remolded soil was prepared using the compaction method, with the required water and dry soil mass calculated based on the moisture content and dry density of the natural soil.

The upper and lower bounds of water content for the regional residual soil were determined from the maximum and minimum values obtained from field samples. Thus, undisturbed and remolded specimens were prepared at water contents of 5%, 10%, 16.3% (natural), 20%, 25%, and 30%. For high-moisture samples, the natural soil was adjusted to the target water content using a quantitative wetting method. Direct shear tests were performed using a Zhilong Technology AZJ-4 automatic quadruple shear apparatus by Qilong Technology Company in Nanjing, China. The samples were placed in the shear box under normal stresses of 100, 200, 300, and 400 kPa. The shear rate was controlled at 0.8 mm/min according to the relevant Chinese standards, and the test was terminated when the horizontal displacement reached 6 mm.

2.3. Finite Element Method

In this study, the deformation and failure processes of the embankment slope were simulated using the FEM implemented in GeoStudio version 2018, which is based on the coupled saturated-unsaturated seepage field theory. The simulation process consisted of two main steps: (1) the SEEP/W module was used to simulate the variation in the slope seepage field during rainfall. Rainfall boundary conditions were applied, and the unsaturated hydraulic conductivity function and soil-water characteristic curve (SWCC) were assigned to the slope model. The Van Genuchten model was adopted for the SWCC to analyze changes in seepage and pore water pressure under rainfall infiltration. (2) Considering the influence of vehicular loads, the SIGMA/W module was used to analyze the stress and displacement fields of the embankment slope at different rainfall stages. The geotechnical parameters were obtained from previously published laboratory test results on granite residual soil [27,28,29], as summarized in

Table 1. Mechanical Parameters of Geomaterials in Embankment Slopes.

	Bulk Density (kN·m−3)	Cohesion (kPa)	Internal Friction Angle (°)	Elastic Modulus (Mpa)	Poisson’s Ratio	Permeability Coefficient (m/d)
Embankment fill	18	20	20	100	0.25	0.43
Natural slope	20	25	25	60	0.32	0.38

.

2.4. Discrete Element Method

The embankment slope model was constructed using the discrete element particle flow software PFC2D (version 5.0). First, the geometry of the original slope and the fill section was defined by creating node coordinates (geometry nodes), which were connected (geometry edges) to form the slope boundaries and generate the walls. Then, spherical particles (balls) were filled within the boundary walls to establish a complete mesoscopic particle flow model of the slope. In this study, particle diameters were set within the range of 0.1–0.15 m, resulting in a total of 23,830 generated particles. This selection was based on the scaling criterion proposed by Sun and Wang [30], which indicates that when the ratio between the characteristic length of the model and the particle radius exceeds 50, the influence of particle size on the macroscopic behavior becomes negligible. The micro-mechanical parameters of the particles are listed in

Table 2. Microscopic strength parameters of particle flow model.

	Water Content/%	Friction Coefficient	Normal Bond Strength/N	Tangential Bond Strength/N
Natural slope	16.3%	0.73	5.6 × 104	5.76 × 104
	20%	0.68	4.8 × 104	5.0 × 104
	25%	0.62	4.0 × 104	4.22 × 104
	30%	0.55	3.5 × 104	3.6 × 104
Embankment fill	16.3%	0.5	3.3 × 104	3.4 × 104
	20%	0.46	3.0 × 104	3.1 × 104
	25%	0.4	2.5 × 104	2.7 × 104
	30%	0.35	2 × 104	2.1 × 104

.

A linear contact bond model was adopted to simulate the soil behavior. Once the linear bond was applied to the particles, localized adhesive contacts were formed between adjacent particles, providing tensile bond strength and shear bond strength. When the inter-particle contact force exceeded the shear bond strength, the bond failed, and the resulting contact force was then controlled by friction. When the tensile force between particles surpassed the tensile bond strength, the bonding between particles was broken, the contact force dropped to zero, and the model degraded into a purely linear contact model. The DEM direct shear test model was constructed based on the shear stress–strain curves of natural and remolded granite residual soil (GRS) obtained from laboratory direct shear tests under varying moisture contents, ensuring consistency with the experimental scale (Figure 3). The micro-mechanical parameters were iteratively calibrated until the DEM specimen’s strength curve closely matched the laboratory results. The finalized parameters included elastic modulus, Poisson’s ratio, friction coefficient, and both normal and tangential bond strengths.

In this study, rainfall infiltration was simulated using a coupled PFC-FiPy approach. Irregular triangular meshes were generated using Gmsh2D version 2015, and the two-dimensional Richards equation (describing soil water movement) was solved using the Python version 2010-based FiPy solver. The Van Genuchten (V-G) model was incorporated to couple the hydraulic properties, enabling the computation of seepage field distributions under varying rainfall intensities and durations, thereby allowing the investigation of embankment slope deformation under rainfall.

2.5. Boundary Conditions of Loading and Seepage

Meteorological data collected in the month before the MeiDa accident indicate that the failure was triggered by heavy rainfall in the five days preceding the event, with a total precipitation of 560 mm. Therefore, a rainfall intensity of 4.68 mm/h over 5 days was adopted in the simulations.

In the FEM, the left, right, and bottom boundaries were set as impermeable, while the slope surface was assigned a unit flux boundary condition to allow rainfall infiltration. Horizontal displacement constraints were applied on both lateral boundaries to restrict lateral movement, whereas both horizontal and vertical displacements were fixed at the base to prevent any movement. The finite element mesh size was set to approximately 1 m. As a result, the embankment slope model established by the FEM consisted of 1434 elements and 1539 nodes (Figure 4).

In the DEM, rainfall was applied as a vertical infiltration across the slope surface, with the left, right, and bottom boundaries set as impermeable (Figure 5). Saturated hydraulic conductivity and Van Genuchten parameters were adopted from the study of Ma et al. [31]. These parameters were obtained by fitting the soil-water characteristic curves (SWCC) measured in laboratory tests on GRSs from Guangdong, with values summarized in

Table 3. Permeability coefficient and V-G model parameters.

Permeability Coefficient (m/s)	θs/%	θr/%	α	n	m
5 × 10−6	42.11	5.38	0.01854	1.3862	0.2786

.

Considering standard pavement configurations and vehicle loading, a tire contact pressure of 0.7 MPa was applied, concentrated at the center of the roadway. To simulate realistic vehicle operations, 3000 cyclic dynamic loads were imposed over a 120 h (5 day) rainfall period, applied concurrently with rainfall to reflect the embankment slope’s dynamic response under the combined influence of precipitation and traffic.

3. Results

3.1. Direct Shear Test Results Under Different Water Contents

As shown in Figure 6, the undisturbed GRS exhibits a typical strain-softening stress–strain response under a low confining pressure of 100 kPa, where the shear stress decreases progressively after reaching the peak value. This behavior is only slightly influenced by changes in water content. The stress–strain curves of remolded GRS with different water contents under various normal stresses are presented in Figure 7. Similarly to the undisturbed soil, the peak shear stress of the remolded samples increases with increasing confining pressure and decreases with higher water content. Nevertheless, distinct differences are observed between the two types of specimens. The remolded soil consistently demonstrates strain-hardening behavior, in contrast to the strain-softening response of the undisturbed samples. Moreover, at identical water contents and confining pressures, the peak shear strength of the remolded soil is substantially lower than that of the undisturbed soil, primarily due to the loss of inherent soil structure during the remolding process.

3.2. Seepage Field Variation in Embankment Slopes

Since pore water pressure is a field variable in a continuous medium, the FEM is suitable for describing the continuity of groundwater flow and can directly determine the distribution of pore pressure within the slope. In contrast, the DEM treats the soil as a system of discrete particles and cannot directly solve for a continuous pore pressure field. Instead, variations in saturation are commonly used to represent the seepage behavior and moisture migration induced by rainfall.

As shown in Figure 8, pore water pressure variations were obtained at 40 h, 60 h, 80 h, and 120 h after the onset of rainfall, revealing the following patterns. At 40 h, rainfall initially infiltrated uniformly in a direction perpendicular to the slope surface. Pore pressure changes were primarily concentrated in the shallow slope layers, with dense contour lines confined near the surface. At 60 h and 80 h, pore pressure contours gradually extended toward the slope midsection and toe, with the most pronounced increases occurring near the toe and shallow layers. This indicates that rainfall infiltration caused groundwater levels to rise and the saturated zone to expand, showing progressive penetration of water into deeper soil layers. At 120 h (after 5 days), the lower slope reached peak pore pressure over a wide area, with near-surface and toe regions nearly fully saturated, and the maximum infiltration depth was approximately 3.5 m.

As illustrated in Figure 9, saturation distributions obtained via the DEM show that the saturated zone (red area) gradually expanded from the surface to deeper layers and toward the slope toe as rainfall progressed, with overall saturation steadily increasing. During the early rainfall stage, the saturated area was limited, primarily confined to shallow slope layers. With continued infiltration, water gradually percolated downward and accumulated at the slope toe, forming highly saturated zones. Ultimately, by 120 h, continuous saturation bands were formed at the slope toe and mid-to-lower sections, demonstrating the cumulative effect of rainfall infiltration and localized water accumulation, with a maximum infiltration depth of approximately 3.8 m.

3.3. Deformation Field Variation in Embankment Slopes

Vertical deformation of the slope at 40 h, 60 h, 80 h, and 120 h after rainfall is presented in Figure 10, as determined by FEM. At 40 h, vertical displacements were minimal, with only slight settlement observed near the crest (approximately 0.004–0.012 m), indicating that rainfall infiltration had little impact and the embankment remained in a stable condition. At 60 h, progressive infiltration and the formation of local saturation zones led to a notable expansion of deformation in the upper and middle slope, with peak vertical displacement reaching 0.048 m (Figure 10). Comparing Figure 10c,d, it is evident that with prolonged rainfall, vertical deformation under combined rainfall and traffic loading gradually enlarged, spreading from the slope crest and toe to the midsection. At 120 h, the maximum vertical displacement reached 0.076 m, indicating that the embankment had not yet failed.

Displacement contours of the embankment slope under the combined effects of rainfall and dynamic loading, based on DEM simulations, are shown in Figure 11a–d. After 40 h, soil moisture in the embankment increased due to infiltration, and the upper fill particles first formed an arcuate slip surface (Figure 11a). At this stage, the slope remained near a critical state, with no large-scale particle movement. By 60 h, the slip surface gradually extended toward the front edge of the fill, yet no significant particle sliding was observed (Figure 11b). At 80 h, the slip surface through the fill became continuous, and surface particles began to slide under rainfall, marking the onset of pronounced slope deformation (Figure 11c). At this stage, local failure of particle contact force chains occurred, sliding progressed along the potential slip surface, and local displacements accumulated rapidly. After 120 h of rainfall, significant collapse occurred in the upper fill, with total slope displacement reaching approximately 3 m, indicating overall slope failure (Figure 11d). DEM simulations indicate that the slip surface fully developed through the slope, and particles collectively slid to form a landslide mass.

4. Discussion

4.1. Deformation Differences Between Two Methods

Because the rainfall conditions were kept consistent in both DEM and FEM simulations, the resulting slope seepage fields were largely similar. As rainfall duration increased, the infiltration depth gradually grew, with maximum infiltration depth nearly identical, ensuring that both methods analyzed deformation differences under equivalent hydrological conditions.

In this study, Point A (slope crest) and B (mid-slope) were selected for comparative deformation analysis (Figure 4 and Figure 5). Deformation differences between FEM and DEM are displayed in Figure 12. In DEM simulations, deformation was most significant at the crest (Point A). From 0 to 60 h, displacement remained minor (<0.3 m), followed by rapid growth after 60 h, reaching 1.4 m at 120 h, signaling slope instability. At Point B, deformation remained only 0.27 m at 80 h, indicating no overall slope failure yet, only local downward displacement. After 90 h, displacement increased sharply, almost vertically, suggesting that slope-top deformation propagated, leading to overall slope instability.

For FEM analyses, displacement at Points A and B was relatively gradual, with the maximum occurring at the slope crest (Point A) (Figure 12). At Point A, displacement grew steadily over 0–120 h, reaching 0.075 m, and at Point B, it was 0.040 m. FEM results indicated the slope remained stable and no failure occurred under the given rainfall. FEM treats soil as a continuous medium, using elastoplastic constitutive models to describe stress–strain relationships, and can simulate overall slope deformation and pore pressure distribution. However, FEM has limitations in capturing discontinuous mechanisms such as crack initiation, particle sliding, and local collapse. In contrast, DEM treats soil as a discrete particle system, directly simulating particle movement, force chain rearrangement, and crack propagation via contact models and bond strengths, naturally capturing local failure and rapid slope instability. Due to these fundamental model differences, DEM can capture local sliding and global failure of GRS slopes under rainfall, while FEM generally produces gradual displacement without clear signs of instability.

Considering the loose, porous, fissured, and highly collapsible nature of GRS with low water stability, slope failure is governed by local particle sliding and crack development. Thus, DEM is preferred over FEM for investigating the deformation and instability of such embankment slopes, providing a more realistic representation of local and global failure progression.

4.2. Field Case Study

At approximately 01:57 on 1 May 2024, the Meida embankment experienced a collapse caused by continuous rainfall on GRS, leading to 52 deaths. According to official reports, extreme rainfall triggered the incident; the total precipitation prior to the accident was 3.75 times the historical average, accumulating over 560 mm. Based on weather data from the month prior to the Meida accident, the collapse was attributed to five days of heavy rainfall totaling 560 mm. Consequently, a rainfall scenario of 4.68 mm/h for 5 days was adopted in this study.

Continuous precipitation caused sustained groundwater rise and prolonged saturation at the embankment base, softening the foundation, reducing shear resistance, and ultimately inducing collapse of the upper fill, resulting in a highway embankment failure. Hence, the failure accident of the Meida embankment can be classified as a rainfall-triggered failure of GRS. As shown in Figure 13, half of the roadway was damaged, likely due to the sudden collapse of the upper fill caused by rainfall infiltration into the embankment. This study compared embankment slope deformation and failure processes under the same rainfall infiltration conditions using FEM and DEM. From the perspective of failure evolution, DEM successfully reproduced the process of crest deformation, through-going sliding band formation, and overall embankment collapse. After 120 h of rainfall, the upper slope deformation reached 1.4 m, closely matching the characteristics observed in field investigations and photographs (Figure 13).

In contrast, FEM responses showed gradually accumulated, minor continuous deformations (Point A ~0.075 m, Point B ~0.040 m), failing to capture sudden brittle failure of the embankment slope, differing significantly from the observed failure scale and pattern. Therefore, for investigating instability mechanisms of GRS embankment slopes, DEM is undoubtedly the most suitable approach. From a physical mechanism perspective, GRS is a discontinuous, collapsible particle–cement system; rainfall causes cement dissolution and contact state changes, with destabilization of particle force chains and initiation and propagation of cracks as the primary failure mechanisms. DEM models the soil as an assembly of particles with bonded contacts, directly capturing local contact force rearrangements, bond breakage, and crack coalescence, thereby naturally producing local instability that propagates to global failure. In contrast, FEM assumes a continuous medium and employs elastoplastic constitutive relations. While it can model pore pressure and overall strain, it cannot directly replicate sudden particle-scale failures, abrupt slope sliding, or through-going instability induced by rainfall infiltration in GRS.

4.3. Limitations of This Work

Although a comparative analysis of the FEM and the DEM for simulating the deformation and failure of GRS embankment slopes was conducted in this study, several limitations remain, which indicate meaningful directions for future research. First, although particle-scale failure processes and chain-type sliding behavior can be realistically reproduced by DEM, its high computational cost limits its applicability to large-scale slopes or long-term simulations. Therefore, it is suggested that future efforts be directed toward the development of multi-scale or hybrid modeling approaches, in which the strengths of FEM and DEM can be integrated to ensure computational efficiency while accurately capturing discontinuous failure mechanisms.

Second, the findings obtained in this work are based solely on rainfall-induced instability of GRS slopes and thus may not be fully generalizable to other soil types or to slopes subjected to different triggering factors. External influences such as seismic loading, rapid drawdown of groundwater levels, and anthropogenic disturbances were not considered. It is recommended that future studies extend FEM and DEM simulations to multi-hazard coupled scenarios so that a more comprehensive understanding of slope stability under realistic engineering conditions can be achieved.

In summary, future studies should emphasize multi-scale modeling, hybrid numerical approaches, and simulation of slope responses under complex loading conditions, thereby advancing the understanding and predictive capability of instability mechanisms in GRS slopes.

5. Conclusions

This study focuses on a case of embankment slope failure in GRS. By simulating slope deformation and failure processes using both FEM and DEM, the applicability and limitations of these two numerical methods for analyzing deformation and failure of GRS embankments were evaluated. The main conclusions are summarized as follows:

(1)

Rainfall infiltration leads to progressive expansion of the internal saturated zone of the embankment slope. With prolonged rainfall, the saturation zone spreads from the surface downward toward the slope toe. FEM and DEM produce consistent predictions of seepage evolution, with maximum infiltration depths of about 3.5 m and 3.8 m, respectively, suggesting that both approaches adequately represent soil seepage behavior under rainfall conditions.

(2)

Regarding deformation response, DEM effectively reproduces the full sequence of slope failure progress under rainfall, from crest deformation to sliding band penetration and eventual overall collapse. Crest deformation reaches 1.4 m in the DEM simulation after the rainfall, indicating clear slope failure. In contrast, FEM predicts relatively slow deformation, with a maximum crest displacement of only 0.075 m over the same period, indicating that the embankment remains stable.

(3)

By comparing the FEM and DEM results with field survey data and site photographs, it is evident that DEM best replicates the observed failure patterns, including crack initiation, soil detachment, and progressive landslide development. In contrast, traditional FEM fails to capture the complex failure process of GRS embankments. Thus, DEM is clearly the most appropriate approach for analyzing the instability mechanisms of GRS slopes.