Multi-field responses and failure mechanisms of loess slopes under engineering disturbance and extreme rainfall: implications for sustainable slope management

Abstract

Introduction:

In the context of climate change, elucidating the stability of loess slopes affected by engineering interventions and extreme rainfall is crucial for sustainable slope management in loess terrains.

Methods:

This research employs a 1:20 large-scale physical model to systematically examine the multi-field responses and failure mechanisms of loess slopes subjected to combined surcharge, excavation, and sustained rainfall, with continuous monitoring of stress, volumetric water content, pore-water pressure, and deformation dynamics.

Results:

The findings demonstrate: (1) Engineering activities induce significant stress concentrations that are further intensified and driven downward by rainfall infiltration; in the late stage, peak vertical stress surpassed 150 kPa, indicating pronounced stress redistribution. (2) Rainfall infiltration is characterized by pronounced spatial and temporal variability, with the shallow soil layer rapidly reaching saturation, while deeper strata exhibit delayed water migration and a gradual buildup of pore-water pressure. After approximately 15 h of rainfall, a sharp increase in pore-water pressure was observed, particularly in the mid-to-lower slope toe, which considerably diminished effective stress. (3) The progression of slope failure follows the sequence of “shallow softening → shallow mud-induced sliding → toe-shear failure → flow-plastic/liquefied sliding,” with shallow failure events preceding deep-seated instabilities.

Discussion:

These insights elucidate the underlying mechanisms by which engineering disturbances and rainfall infiltration interact to govern loess slope instability, providing a scientific basis for slope management, early warning systems, and risk mitigation strategies in loess regions under extreme rainfall conditions.

1 Introduction

Climate change constitutes one of the most formidable threats facing society today. According to the World Meteorological Organization (WMO), there is persistent global temperature escalation, accompanied by accelerated glacial ablation and progressive sea-level rise, all highlighting the intensifying global warming trajectory (Hoegh-Guldberg et al., 2019). In this context, the incidence of extreme meteorological phenomena—such as intense precipitation, prolonged heatwaves, and alternating sequences of drought and flooding—continues to surge worldwide (Schiermeier, 2011; Krebich et al., 2022). Projections derived from climatological analyses indicate a substantial increase in the occurrence of extreme rainfall events in the coming decades (Zhang et al., 2011). On the Loess Plateau, precipitation regimes have been profoundly reshaped by climatic shifts; the frequency and magnitude of extreme precipitation—characterized by WMO and local geo-hazard benchmarks (e.g., 1-h rainfall ≥12 mm, 24-h rainfall ≥50 mm, or cumulative rainfall over three consecutive days ≥100 mm)—have risen, often surpassing critical hydrometeorological thresholds (Huang et al., 2014; Fan et al., 2025; Shi et al., 2021; Wu et al., 2022). These intense rain events frequently precipitate cascading disaster chains, including landslides, slope failures, and sudden flash floods, posing significant and compound risks to lives and property throughout the Loess region (Zhang et al., 2019).

Meanwhile, with the ongoing advancement of national urbanization, the Loess Plateau has experienced an increasing frequency of large-scale civil engineering activities, such as road construction and urban expansion. These anthropogenic disturbances have disrupted the original stress equilibrium and structural integrity of loess slopes, resulting in the formation of preferential pathways for rainfall infiltration. Additionally, rainfall-induced softening of the soil and reduction in shear strength have significantly heightened the risk of slope failure, rendering them the primary triggers for loess landslides (Ma et al., 2023). Under extreme rainfall events, loess landslides exhibit distinct characteristics compared to normal conditions: (1) widespread distribution and high destructiveness, often erupting en masse in areas recently subjected to excavation or overloading; (2) abrupt onset and challenging monitoring, where engineering-induced fissure networks expedite rapid water infiltration into slope interiors, causing swift internal degradation with little surface deformation, frequently leading to large-scale instability within hours; (3) pronounced chain-reactive effects and formidable prevention challenges, as the low permeability and high water sensitivity of loess, when coupled with extreme precipitation and anthropogenic perturbations, readily trigger shallow slides, debris flows, and deep-seated landslides, substantially diminishing the efficacy of single mitigation measures. Consequently, elucidating the mechanisms controlling loess slope stability under the synergistic influence of engineering activities and extreme rainfall has emerged as a central objective in engineering geology, holding significant implications for the sustainable development of loess regions.

Significant progress has been achieved in research on loess slope instability under mono- or dual-factor influence, with studies primarily exploring the interrelations among “rainfall, anthropogenic disturbance, and slope response.” A systematic review and comprehensive analysis can be conducted across three core dimensions: seepage field evolution, stress field adjustment, and deformation-failure mechanisms, elaborated as follows:

1.1 Rainfall infiltration and seepage field characteristics

Research in this domain focuses on the migratory dynamics of rainfall within loess media and the spatiotemporal evolution of seepage fields, culminating in three principal findings. First, the coupled control of rainfall intensity and duration on infiltration processes has been substantiated; Ng and Shi (1998) demonstrated via numerical simulation that high-intensity rainfall accelerates infiltration, and rainfall exceeding a critical duration markedly increases infiltration depth. These two variables jointly regulate the extent of slope saturation and the development rate of the seepage field. Second, the heterogeneous nature of infiltration modes is now a consensus. Zhao et al. (2020), utilizing time-domain resistivity tomography, revealed that rainwater does not infiltrate vertically and uniformly, but rather preferentially migrates along macro-pores and structural joints in the loess, creating localized saturated zones. Third, the spatiotemporal heterogeneity of volumetric water content has been quantified; indoor experiments by Wu et al. (2017) indicated that shallower soil layers (0–30 cm) reach 35%–40% saturation within 10–15 h, whereas deeper layers (below 60 cm) respond more slowly, ultimately achieving saturation increments only 1/2–1/3 those of the surface, reflecting a “shallow-priority saturation and deep-layer gradient increase” pattern (Hu et al., 2024). However, current research exhibits marked limitations: most studies focus on natural slopes or single rainfall events, lacking targeted investigations into the seepage characteristics under stress redistribution induced by engineering activities (e.g., crest overloading, toe excavation). These engineering operations modify the pore structure and hydraulic gradients of slopes, but due to insufficient empirical data and theoretical support, the evolution of seepage fields under “combined anthropogenic disturbance–rainfall” regimes remains inadequately understood (Lin et al., 2024).

1.2 Stress field response of slopes under rainfall and engineering disturbance

Research in this field centers on elucidating “the regulatory effect of external disturbances on slope stress distribution,” with two main advancements. On one hand, the influence of permeability on slope surface stress stability has been clarified; Pradel and Raad (1993) indicated that low-permeability, cohesive soil slopes can retain rainwater at the surface, causing self-weight increments and stress concentration, whereas high-permeability sandy slopes, due to rapid rainfall infiltration, experience more uniform stress distribution. On the other hand, the quantification of stress effects under anthropogenic disturbances (overloading, excavation) has been advanced. Sun et al. (2021), using physical model tests, showed that crest overloading induces stress concentration zones in the lower-mid portions of the slope, with both magnitude and intensity of concentration positively correlated with the load—vertical stress rises from <20 kPa to >50 kPa as overloading increases from 0 to 100 kPa, with the affected zone expanding toward the toe. Liu et al. (2025a) observed that excavation disrupts the initial stress equilibrium at the loess-mudstone interface, and steeper excavation angles lead to more pronounced stress concentration; when the excavation angle increases from 45° to 60°, peak shear stress increases by 20%–30%, with concentration occurring in the lower-mid interface (potential sliding origination zones). The primary research gap lies in the lack of clear definition of the initial stress field formed by “crest overloading–toe excavation,” and an insufficient understanding of the nuanced stress redistribution during soil softening induced by rainfall infiltration (Fustos et al., 2020). Furthermore, the dynamic interplay among these three factors and the nonlinear impact of stress evolution on slope stability have not been systematically elucidated (Zong et al., 2024).

1.3 Study of slope deformation, failure modes, and mechanisms

Current research has identified various failure modes and mechanisms in loess slopes triggered by single or dual factors. Peng et al. (2015), integrating field surveys and numerical modeling, revealed the chain disaster mechanism of the “landslide–debris flow” for loess-mudstone slopes—intense rainfall leads to ponding at the loess-mudstone interface, reduced shear strength initiates bedding-controlled landslides, and the landslide mass amalgamates with colluvial materials to form high-mobility debris flows. Scholars have also investigated shallow rainfall-induced landslides (Hou et al., 2024), freeze–thaw and rainfall combined failures (Zhang et al., 2024; Lü et al., 2023), and joint-controlled instability modes (Liu et al., 2025b; Wang et al., 2025), illuminating certain mechanisms under single or interactive triggers (Zhang and Wang, 2018). However, research exhibits obvious contextual limitations: studies focusing on the combined conditions of “crest overloading–toe excavation–rainfall” are scarce, with key scientific issues unresolved. Specifically: How do engineering disturbances and rainfall synergistically lead to rapid soil liquefaction and fluidization? Which critical factors control the initiation, propagation, and breakthrough of the slip surface? What governs the transformation from shallow, sliding-type to deep shear failure? (Zhuang et al., 2024) The lack of resolution of these questions has hindered the development of effective disaster mitigation strategies for complex engineering scenarios.

In summary, existing studies have preliminarily clarified the characteristics of seepage, stress, and failure in loess slopes under mono- and dual-factor actions. However, a significant disconnect persists between research scenarios and practical engineering contexts, where slope failures are predominantly triggered by the combined effects of “anthropogenic disturbance (overloading/excavation) and rainfall.” Current research lacks systematic investigation into the synergistic interactions of multiple factors, particularly regarding the dynamic relationship between seepage, stress, and deformation, and the evolution of composite failure mechanisms under multifactor scenarios. There is not yet a comprehensive theoretical framework or technical system for such conditions. This gap underscores the core research focus of the present study: to investigate loess slope instability under the combined action of three factors, elucidate the synergistic mechanisms, and develop key preventive technologies for composite hazard scenarios.

This research seeks to address the theoretical deficit in three-factor synergy studies by supplying empirical evidence and theoretical grounding for targeted mitigation and control of slope hazards in loess engineering zones, thereby advancing sustainable slope management practices amid climate change-driven extreme precipitation events.

2 Large-scale physical model testing design

2.1 Model design

The model was constructed to replicate the typical dimensions and morphology of Malan loess landslides documented in the Southern Jingyang Loess Plateau. Based on field investigations and geological surveys of the study area, the prototype slopes are characterized by a height of 30–90 m (average 60 m), a natural slope angle of 40°–80°, and a dominant failure layer within the surface Malan loess (Q₃) with a thickness of 5–14 m. Utilizing similarity theory as a framework (geometric scaling factor 1:20) and considering constraints such as site conditions, experimental timeline, and available funding, the model was scaled to reflect the prototype’s key features: the model container has dimensions of 3.3 m (length) × 2.4 m (width) × 1.8 m (height), corresponding to a prototype length of 66 m, width of 48 m, and height of 36 m—consistent with the scale of small-to-medium landslides in the region.

2.2 Loading and rainfall systems

The loading mechanism exerts force at the slope crest via three hydraulic actuators, each delivering load through an 80 cm × 80 cm × 2 cm steel load-distribution plate. These loads are transferred to a high-tensile-strength steel reaction beam that is securely welded to the structural model frame. Four vertical cylindrical steel restraining columns are strategically positioned to mitigate tilting of the load plates during loading, thereby maintaining their horizontal alignment. The surcharge regime (0–100 kPa) is informed by prevailing geotechnical engineering standards: typical surcharges produced by infrastructure such as roadways, building foundations, and earth embankments on the Loess Plateau range from 50 to 100 kPa, with 100 kPa representing the upper threshold for routine anthropogenic disturbance—effectively modeling worst-case stress scenarios.

Rainfall infiltration is simulated using the Groundwater Explorer XHZ-23 artificial rainfall apparatus, which deploys vertically oriented nozzles to produce conical water droplets under precisely regulated hydraulic pressure. The nozzle configuration follows a diamond array (two by two), delivering rainfall intensities from 8 to 200 mm/h across a uniformly covered test area of 12 m² from a vertical release elevation of 4 m. Prior to experimentation, the system underwent calibration with a rain gauge, determining that a pump setting of 150 kPa yields a rainfall rate of 15 mm/h. This calibrated intensity was adopted as the standardized rainfall input for the experimental investigations. This rainfall intensity aligns with the hydrological characteristics of extreme precipitation events observed in the southern Jingyang region of the Loess Plateau. Regional field investigations and meteorological data indicate that the Loess Plateau is prone to short-duration, high-intensity extreme rainfall, with the southern sector experiencing a relatively high annual frequency of such events (0.6–0.8 events/year) (Hoegh-Guldberg et al., 2019). According to the World Meteorological Organization’s classification criteria for extreme rainfall in loess areas, precipitation exceeding 12 mm/h within 1 hour is categorized as “landslide-triggering extreme rainfall” (Schiermeier, 2011). The 15 mm/h rainfall intensity selected for this experiment not only corresponds to the historically observed 5-year return period value for extreme rainstorms in the region, but also falls within the typical intensity range for landslide-inducing extreme rainfall events (10–30 mm/h) (Shi et al., 2021).

2.3 Monitoring system

A comprehensive monitoring system was implemented to track stress, moisture, and deformation. Stresses within the slope are measured in three orthogonal directions: Profile 1 for σxx, Profile 3 for σyy, and Profile 5 for σzz. 48 BX-1 type resistive earth pressure sensors (accuracy error ≤0.5%) are installed across these profiles. These sensors were zeroed after soil filling to monitor additional stress changes during loading and rainfall.

Volumetric water content is measured at Profile 4 using EC-5 sensors (±2% accuracy, 0.1% resolution). Pore water pressure is monitored at Profile 2 using Donghua Instrument DMKY-type sensors (range: 0–50 kPa, accuracy ≤0.3%). Displacement at the slope crest is recorded using Liyang Jincheng YWD-50 displacement meters (accuracy: 0.01 mm). The deformation process is visually documented by a video camera recording the slope face and timed photographs taken from the side (Zhou et al., 2025).

2.4 Soil preparation and experimental procedure

The soil utilized in this study is Malan loess obtained from Zhaitou Village, situated in the southern sector of the Jingyang Loess Plateau, thereby aligning with the primary stratigraphic layer of the prototype site. Comprehensive laboratory analyses determined the following geotechnical parameters: natural moisture content of 12.26%, bulk density of 1,480 kg/m³, plastic limit of 18.46%, liquid limit of 30.37%, cohesion of 27.59 kPa, internal friction angle of 29.3°, and a permeability coefficient of 2.0 × 10⁻⁶ m/s. These values corroborate previously reported data for Malan loess within this geographical region.

The 10 cm × 10 cm grids and slope reduction lines were marked on the sidewalls before filling. The soil was sieved through a 0.5 cm mesh. For each 10 cm thick layer, the required soil mass was calculated based on the target moisture content and density. After weighing, the soil was spread, compacted, smoothed, and roughened, with a white marker layer placed between successive layers. Golf tees were inserted at gridline intersections on the sidewalls as reference points, and sensors were installed at predetermined locations. Upon completion of filling, the slope was cut to an initial angle of 45°, and 20 cm × 20 cm marker lines were drawn on the slope face with nails buried at the intersections.

The experimental procedure consisted of three stages: Surcharge Staged: A load was applied in 10 kPa increments to the slope top within the area x = 0.8–1.6 m, y = 0–2.4 m. The model was allowed to stabilize after each increment until a total load of 100 kPa was reached. Excavation Staged: The slope angle was reduced in 5° increments after load stabilization. The model was stabilized after each cut until a final slope angle of 60° was achieved. Rainfall Simulation: Continuous rainfall was applied at a constant 15 mm/h rate over the entire model surface. Volumetric water content and pore water pressure were continuously monitored during this phase. The experiment was terminated upon the occurrence of either a significant deep-seated or shallow landslide.

The experimental setup and sensor layout are detailed in Figure 1.

FIGURE 1

3 Stress-field response characteristics of slopes under the combined effects of surcharge, excavation, and rainfall

This section examines the development of the stress field in loess slopes experiencing surcharge loading, excavation activities, and precipitation influence. The analysis focuses on variations in stress distribution and temporal response characteristics under these operational scenarios, drawing upon data obtained from physical model experiments. The principal findings are summarized as follows.

3.1 Surcharge phase

As illustrated in Figure 2a, applying a surcharge load results in distinct load-controlled stress distributions within the slope. When the surcharge is small, stress concentration is minimal. As the load increases, a stress concentration zone forms, with the maximum stress σzz reaching approximately 30 kPa at P = 40 kPa. Further increases in the surcharge result in a broader stress concentration zone, with the maximum stress exceeding 40 kPa (when P = 70 kPa) and 50 kPa (when P = 100 kPa), respectively. These results indicate that an increase in surcharge load significantly intensifies stress concentration within the slope, and the stress field becomes increasingly inhomogeneous as the load level increases.

FIGURE 2

3.2 Excavation phase

Figure 2b illustrates the variation of excavation angles on the stress σzz field. When the excavation angle is mild (β = 45°), a noticeable stress concentration zone forms within the slope, with the maximum stress reaching approximately 50 kPa. As the excavation angle increases (β = 50°, 55°, and 60°), both the range of the stress concentration zone and the maximum stress increase, with the maximum stress approaching 60 kPa when β = 60°. These findings suggest that steeper excavation angles exacerbate unloading and stress redistribution in the slope, thereby increasing the concentration of σzz.

3.3 Rainfall phase

Based on the results presented in Figure 2c, there are pronounced temporal variations in the relationship between rainfall duration (t) and vertical stress σzz. At the initial stage of precipitation (t = 2.0 h), the stress concentration zone is relatively limited, with the peak stress at approximately 80 kPa. As the rainfall event progresses (t = 7.5 h, 15 h, 21 h), both the scope and magnitude of stress concentration markedly increase, with the maximum stress exceeding 150 kPa at t = 21 h. Additionally, the soil within the rainwater infiltration region undergoes mechanical softening, which diminishes local stress magnitudes. Conversely, vertical stress in the deeper portion of the infiltration zone rises considerably as a consequence of stress redistribution. These changes are chiefly attributed to increased soil moisture and the consequent reduction in soil mechanical strength, resulting in ongoing stress adjustment and concentration within the slope. Thus, stress concentration intensifies with prolonged rainfall duration.

For the monitoring of x and y stresses, some sensors have missing or unreliable data, so it is impossible to draw contour lines. We can only select some data to draw time series curves.

3.4 Time-history characteristics of stress at representative points

The stress time-history profiles for selected monitoring locations, as depicted in Figures 2d–f, elucidate the influence of varying operational scenarios on the temporal evolution of stress. Notable findings are as follows:

Surcharge Phase: At each instrumented location (e.g., S-B-Z-2, S-E-Z-4), stress exhibits stepped increments as successive loads are imposed. The magnitude of stress augmentation aligns with surcharge intensities, indicating progressive stress accumulation and redistribution within the soil mass during loading.

Excavation Phase: For designated positions (e.g., S-E-X-3, S-C-X-4), stress registers discrete changes, with variance corresponding to increases in excavation angle. These observations underscore the perturbation and subsequent reorganization of the antecedent stress regime induced by excavation activities.

Rainfall Phase: Under rainfall, stress readings at specific points generally reveal pronounced volatility, characterized by rapid acceleration (e.g., z-direction stress at S-B-Z-2 surpasses 150 kPa within a short interval), followed by a subsequent decline. This trajectory evidences the dynamic disruption rainfall imparts on the stress landscape, signifying preliminary indicators of incipient instability as soil strength deteriorates and deformation intensifies.

Collectively, these results confirm that the stress distribution within loessial slopes is distinctly governed by surcharge, excavation, and rainfall variables. Amplified surcharge loads, greater excavation angles, and extended rainfall exposure collectively intensify stress concentrations within the slope mass. The temporal stress data at representative monitoring points delineate a phased evolution of the stress field, encapsulated as “surcharge accumulation—excavation disturbance—rainfall-induced weakening.” These insights furnish a vital empirical foundation for elucidating the instability mechanisms of loess slopes subjected to multifaceted, in situ boundary conditions.

4 Laws of water migration and pore water pressure variation

During rainfall, water migration and pore-water pressure responses in the slope are key factors that induce slope instability. Based on monitoring data from large-scale physical model tests, this study systematically analyzes the laws of internal soil water migration and the characteristics of pore water pressure variation under rainfall conditions, providing an experimental basis for research on the instability mechanism of rainfall-induced slopes.

4.1 Characteristics and laws of soil water migration

4.1.1 Temporal evolution of spatial distribution of volumetric water content

As illustrated in the volumetric water content contour maps in Figure 3a, prolonged rainfall duration leads to a progressive downward and inward migration of high-moisture zones within the slope soil profile, advancing from the surface layer toward deeper strata and from the slope crest toward its interior. In the early phase of precipitation, zones of elevated water content are predominantly confined to the near-surface layer, with negligible gradients observed within the slope. Upon reaching 7.5 h of rainfall, the saturated zones expand into the mid- and lower slope regions, penetrating deeper layers and prompting a more pronounced water content gradient throughout the soil matrix. After 15 h of sustained rainfall, these zones further enlarge, resulting in the formation of contiguous preferential flow pathways within the slope. By the late rainfall stage at 21 h, the high-moisture areas envelop the upper to mid-sections of the slope, signifying extensive internal water redistribution.

FIGURE 3

This spatiotemporal evolution in soil moisture distribution principally results from water flux migrating along soil pore networks, driven by the hydraulic gradient from superficial to deeper horizons under rainfall infiltration. The slope crest and surface function as primary recharge zones for subsurface flow due to direct rainfall input. The presence of a waterproof bearing plate induces a localized low-moisture anomaly beneath it, demonstrated by peak-like isohyets. Rainfall infiltrating laterally from both sides of the bearing plate is redirected by matric suction toward the subplate soil. When the precipitation event exceeds 15 h, the soil moisture content beneath the bearing plate commonly approaches the liquid limit.

4.1.2 Temporal variation laws of volumetric water content at typical monitoring points

Overall, the moisture content at all sampling points demonstrates a generalized ‘increase–stabilize’ trajectory, with the terminal equilibrium values strongly correlated with each monitoring point’s spatial coordinates (including soil depth and slope position). This underscores the pronounced spatial heterogeneity in subsurface water redistribution mechanisms during rainfall infiltration episodes.

4.2 Characteristics and laws of pore water pressure variation

It can be seen from the temporal variation curves of pore water pressure at typical monitoring points in Figure 3c that the variation of pore water pressure presents complex characteristics of “lag-mutation-fluctuation”, and the response timing and amplitude of different monitoring points are significantly different:

Lag in response timing: For most monitoring points (e.g., P-E−1, P-D-1), the pore water pressure is almost 0 in the early stage of rainfall (first 15 h) and only begins to increase significantly in the later stage (after 15 h), which indicates that the increase of pore water pressure lags behind the water migration process, and pore water pressure begins to accumulate only when the soil water content reaches a certain threshold (saturation or near-saturation).

Mutation of pressure increase: After the pore water pressure starts to respond, its amplitude increases rapidly (e.g., P-C-1 and P-B-1 quickly exceed 1.0 kPa at around 20 h). This mutation arises from the “breakthrough” migration of water within the slope. When water accumulates in the pores enough to overcome the soil skeleton’s resistance, the pore water pressure rises rapidly.

Fluctuation characteristics in the later stage: After reaching the peak value, the pore water pressure does not remain stable but exhibits pronounced fluctuations (e.g., the P-D-1 and OA2 curves), which reflect the coupling between water migration and soil structure deformation within the slope during the later stage of rainfall. The unsteady water migration and micro-displacement of soil particles together lead to dynamic changes in pore water pressure.

In addition, the peak pore water pressure is closely related to the spatial positions of the monitoring points. The peak pore water pressure of monitoring points near the slope toe in the middle-lower part of the slope (e.g., P-C-1, P-B-1) is significantly higher than that of points at the slope edge or surface layer, indicating that the middle-lower part of the slope is the central accumulation area of pore water pressure, which is also consistent with the spatial law of water migration mentioned earlier—the convergence of water near the slope toe in the middle-lower part of the slope directly leads to the significant increase of pore water pressure in this area.

Under rainfall conditions, soil water migration along the slope is the primary driver of changes in pore water pressure. The change in pore water pressure, in turn, affects the migration path and rate of water, showing a strong coupling between the two. In the initial stage of rainfall, water accumulates in the surface layer through infiltration, and pore water pressure does not increase significantly. As water migrates to the deep layer, the pores within the slope are gradually filled with water. When the water content reaches saturation, the pore water pressure starts to rise rapidly. The rise in pore water pressure will further reduce the soil’s effective stress, potentially triggering redistribution of soil particles, altering the pore structure, and affecting subsequent water migration efficiency, ultimately reflected in fluctuations in pore water pressure.

This coupling relationship reveals the intrinsic mechanism of rainfall-induced slope instability: water migration leads to an accumulation of pore water pressure, which, in turn, weakens the soil’s shear strength. When the coupling effect of the two exceeds the anti-sliding capacity of the slope, slope instability will occur.

In conclusion, results from the large-scale physical model test demonstrate that, under rainfall conditions, soil moisture migration within the slope exhibits a spatial pattern of “preferential movement along the surface, followed by progressive infiltration into deeper strata,” and a temporal evolution characterized by “heterogeneous onset and gradual stabilization.” The variation in pore water pressure follows a trend of “delayed response, abrupt escalation, and subsequent fluctuation.” These processes collectively influence slope stability through a coupled “hydro-mechanical” effect, providing crucial empirical evidence for the early warning and mitigation of rainfall-triggered slope failures.

4.3 Characteristics of the wetting front in excavated loess slopes induced by rainfall

This experiment systematically documented the evolution of the wetting front (delineated by the blue “infiltration boundary” zone) and the dynamic changes within the slope soil throughout the rainfall process by capturing sequential on-site images at 3-min intervals (see Figure 4). The total rainfall duration for Test 1 was 1,488 min, and Figure 4 distinctly illustrates the lateral distribution of the wetting front and corresponding slope conditions at critical time points:

FIGURE 4

At 15 min, initial records show localized saturation first occurring at the slope toe (the surface at the toe being covered by the blue infiltration zone), while the slope shoulder remains completely dry—a direct result of concentrated runoff accumulation at the toe. By 33 min, the infiltration front advances slowly from the toe towards the interior of the slope, yet the main slope body is not yet wetted, limiting the infiltration zone to the shallow toe region. During the early rainfall phase, both the slope shoulder and toe become saturated first. At 156 min, the blue infiltration area has extended to the middle and lower sections of the slope, with full saturation of both the surface and the crest (blue zone covering the entire slope face and crest edge), and a superficial infiltration depth of approximately 5 cm. By 597 min, the blue infiltration region entirely occupies the internal toe area, indicating complete saturation of the toe; at this point, the expansion rate of the blue region within the slope slows, aligning with a decline in the overall infiltration rate. After 657 min, localized slippage occurs at the slope shoulder (grey region indicating collapsed soil), exposing a fresh failure surface directly to rainfall. With the original soil barrier removed, infiltration rates in this area increase significantly due to direct rainfall impact. From 822 min onward, photographic observations reveal the emergence of scarps and rills on the slope flanks, which, with continuing rainfall, gradually develop into gullies. This shifts erosion and sediment transport to dominate on the side slopes, resulting in the wetting front (blue zone) advancing more rapidly along gully walls into the slope interior, thereby expanding the infiltration region adjacent to the gully. At the conclusion of the experiment (1,488 min), the measured maximum infiltration depths in each zone correlate with the extent of the blue infiltration area depicted: the slope crest (beneath the non-perforated load plate) exhibits an infiltration depth of approximately 83 cm (with the blue region extending deeply beneath the crest), the infiltration depth directly under the load plate is only 40 cm (the narrow blue area illustrates the limited direct rainfall infiltration), and the slope surface zone records an infiltration depth of 56.6 cm (the coverage of the blue region matches this depth).

Variations in infiltration rate are also critical for investigating the dynamics of soil moisture migration. The following equation is used to quantify the infiltration rate at various depths along the slope wetting front.

In the above equations, represents the infiltration rate, denotes the infiltration depth at the wetting front, and refers to the time required for the wetting front to reach the measurement point. and indicate the difference in infiltration depth and the elapsed time between two measurement points, respectively. Plot the variation curves of vertical infiltration rate with depth for the upslope region and lateral infiltration rate with depth for the midslope region, as illustrated in Figure 5.

FIGURE 5

Figure 5 comprises three subfigures that illustrate the infiltration rate variations across distinct zones at the slope crest and slope face, clearly delineating the relationships between vertical/lateral infiltration rates and infiltration depth at different locations:

As illustrated in Figure 5 (three subplots corresponding to infiltration rate curves for different slope regions), in the unconfined zone at the slope crest (Figure 5a: vertical infiltration rate curve for the unsaturated zone above the water table), the initial vertical infiltration rate increases with infiltration depth during early rainfall: the curve exhibits a pronounced ascending trend within the 0–20 cm depth interval, with peak values (∼12 × 10⁻² cm/min) occurring near 20 cm depth. However, beyond 20 cm depth, infiltration rates decrease according to a power-law relationship with increasing depth, and the curve gradually flattens—consistent with increased seepage resistance as soil pores become water-saturated.

Conversely, in the confined zone beneath the impermeable layer at the slope crest (Figure 5b: vertical infiltration rate curve below the confining layer), vertical infiltration rates decline with increasing depth from the initial stage (∼5 cm depth), exhibiting no rapid infiltration phase. Additionally, peak infiltration rates in this zone (∼10 × 10⁻² cm/min) remain lower than those in the unconfined zone, with the entire curve consistently positioned below Figure 5a curve. This disparity stems from the confining layer’s coverage restricting direct rainfall infiltration, resulting in lower initial soil moisture content and more impeded seepage pathways.

For lateral infiltration on the slope surface (Figure 5c: lateral infiltration rate curve for slope surface), infiltration rates exhibit an overall pattern of “initial rapid fluctuation followed by sustained decline”: during the early phase (0–10 cm depth), infiltration rates rapidly ascend to peak values (∼14 × 10⁻² cm/min); subsequently, influenced by localized slope displacement (such as the shoulder collapse observed previously), infiltration rates display minor fluctuations; ultimately declining progressively with increasing infiltration depth. These fluctuation characteristics directly reflect the exposure of new interfaces and altered infiltration pathways under rainfall erosion.

Analysis of wetting front morphology and infiltration rate variations during experimentation reveals that rainfall infiltration follows consistent patterns: rainwater initially saturates the slope shoulder and accumulates at the toe, surface layers of both slope surfaces and crest regions achieve saturation first, followed by infiltration advancing into the slope interior; upon soil saturation, infiltration rates decelerate (corresponding to the flattened segments in the three subplot curves). With continued rainfall, localized displacement occurs first at the slope crest and gradually extends downward, with unsaturated soil subjected to direct rainfall erosion, subsequently enhancing lateral infiltration rates and depths on the slope surface (corresponding to the fluctuation and peak maintenance phases in Figure 5c), until reaching new equilibrium where infiltration ceases to advance.

5 Slope deformation response and failure modes

5.1 Vertical displacement response of the slope crest

Figure 6 presents the time-history curves of vertical displacement at the slope crest. As shown in the figure, ①Surcharge stage: The vertical displacement at the slope crest slowly increases from 0 to approximately 30 mm. In the early surcharge stage, the curves of different displacement gauges basically overlap, indicating that surcharge enables uniform preloading compaction of the soil at the slope crest, with no obvious differential deformation. In the late surcharge stage, the values of displacement gauges WY-2 and WY-4 (located close to the slope shoulder) are slightly larger, and the bearing plate exhibits a slight inclination at this time. ②Excavation stage: The displacement curves generally maintain the loading phase’s rate of variation. Displacement gauge WY-2 records the maximum displacement; when the slope is excavated from 45° to 60°, the vertical displacement monitored by this gauge is about 10 mm. The displacement of gauge WY-1 first rises slightly and then drops, indicating that the excavation stage exacerbates the inclination of the bearing plate; subsequently, the bearing plate moves downward as a whole under the influence of the initial excavation. The displacement curves show that the impact of excavation on slope deformation is similar to that of slope crest loading—both will induce slope crest deformation. In contrast, the degree of deformation and failure is related to the magnitude of the excavation slope angle. ③Rainfall stage: The displacement curves increase sharply. After 24 h of rainfall, the displacement exceeds 95 mm, and the displacement differences among the four gauges increase slightly (the maximum difference is about 10 mm), reflecting differences in soil softening due to rainfall infiltration. The displacement of WY-4 (close to the slope shoulder) increases rapidly, which is much larger than that of the other three gauges. The curves tend to stabilize after 24 h because the experimental error is caused by the loading device reaching its maximum displacement.

FIGURE 6

5.2 Deformation field of the slope

Figure 7 presents the horizontal (x-direction) and vertical (z-direction) displacement fields of the slope in this test, plotted from monitoring results of relative displacement between pre-embedded golf tees in the soil and the model frame grid. These fields cover the displacement distribution characteristics under multiple working conditions (surcharge, excavation, and rainfall), and the detailed description is as follows:

FIGURE 7

Horizontal displacement is primarily governed by the extrusion deformation of the slope directed toward the excavation free face, with its spatial distribution closely associated with distinct construction phases. (1) Surcharge and post-excavation initial state: Following the formation of the 60° free face through excavation, the overall x-direction displacement remains minor (maximum value <0.5 cm), with only modest displacement concentration observed proximate to the free face (x ≈ 100–200 cm, z ≈ 120–180 cm). This suggests that excavation-induced unloading predominantly results in minimal horizontal extrusion concentrated near the slope shoulder. (2) 8-h rainfall phase: The x-direction displacement concentration zone extends to the upper-middle section of the slope, chiefly located within x ≈ 100–220 cm and z ≈ 60–180 cm. Maximum displacement rises to 0.5–1.0 cm, with a marginal increase in displacement gradient adjacent to the free face, indicating that early rainfall infiltration leads to softening of the surface soil and initiates horizontal deformation toward the free face. (3) 24-h rainfall phase: The x-direction displacement concentration further widens to encompass x ≈ 80–240 cm and z ≈ 40–180 cm, with maximum displacement surpassing 1.0 cm. Displacement contours propagate deeper into the slope (toward decreasing x), demonstrating that sustained rainfall infiltration facilitates deformation progression from shallow to deeper soil layers. Consequently, horizontal extrusion in the upper-middle slope intensifies notably, and deformation gradually interacts with the geometric constraints imposed by the excavation free face.

The sliding surface (Figure 8) is derived from time-series monitoring and mapping results of the soil failure interface on the slope side, under the surcharge-excavation-rainfall coupling effect of the loess slope. It intuitively reflects the progressive evolution characteristics of the shear failure interface driven by rainfall, and the detailed description is as follows:

FIGURE 8

The development of the sliding surface follows the process of “local initiation → progressive expansion → through-going instability”. The sliding surface first emerges in the local area of the slope shoulder, forming an included angle of approximately 15°–20° with the excavation free face; the failure zone (the shaded area) is relatively small. As rainfall persists, the sliding surface expands toward the interior of the slope, with its distribution range expanding, and its shape remains approximately slightly curved. Subsequently, the failure surface under rainfall further develops from the slope shoulder to the rear edge and the deep part of the slope; the stepped shape observed in the latter two stages of the visualization is formed by local slope collapse.

The progressive evolution of this sliding surface essentially corresponds to the disruption of dynamic equilibrium (of “shear stress accumulation - soil strength weakening”) driven by rainfall infiltration. Its shape and through-going characteristics serve as typical indicators of shear instability for loess slopes under the surcharge-excavation-rainfall coupling effect.

This section (Figure 9) describes the real-time observation results of the loess slope surface, focusing on the shallow sliding failure process during the rainfall period of 1,250′47''∼1,490′26''(corresponding to the shallow soil water content on the slope surface increasing from 38% to the critical value of 42%). Its evolution exhibits the characteristics of “progressive initiation - multi-region synergy - liquefied flow slide”.

FIGURE 9

The temporal evolution process of Sliding Failure is as follows. ① Potential Failure Initiation Stage (Rainfall time: 1,250′47″): Rainfall infiltration initially weakens the strength of the shallow slope soil (thickness <10 cm), with no macroscopic sliding observed at this stage; the soil surface shows slight bulging deformation, and shear stress begins to accumulate in the shallow soil, marking the initiation of potential failure. ② Shallow Flow Slide Initiation Stage (Rainfall time: 1,250′48″): The potential failure zone at the side edge rapidly develops into a shallow flow slide mass: the slide mass moves downward along the slope in a “flow-plastic mud-like state”. At this stage, the slide mass has exhibited the “fluidization” characteristic of loess at high water content, a direct manifestation of shallow instability triggered by rainfall. ③ Slope Toe Shear Failure Stage (Rainfall time: 1,326'∼1,346′): The flow slide mass rushes down to the slope toe, triggering shear failure at the slope toe: local collapse occurs in the slope toe soil, forming a pit with a depth of approximately 20 cm, and a shear plane with a dip angle of 60° develops at the edge of the pit; meanwhile, the slope toe soil extrudes outward by 5–8 cm, further weakening the anti-sliding constraint of the lower slope and providing geometric conditions for deep failure. ④ Shallow Flow Slide Failure Stage (Rainfall time: 1,418′): Failure develops from the slope toe and the upper-middle part of the slope. Shallow flow slide is initiated in the upper-middle part of the slope, with the slide mass thickness increasing to 20–25 cm; internal shear cracks and tensile cracks connect to form a “reticulated crack system”. At the same time, shear cracks with a depth >25 cm appear in the side edge area, further expanding the instability range of the slope. ⑤ Dynamic Liquefaction and Stop Stage (Rainfall time: 1,490′25''∼1,490′26″): The slide mass becomes a liquefied flow slide mass: it is completely mudified, and its movement speed increases, embodying the typical “dynamic liquefaction” characteristic; after impacting the slope toe, “fluid surging” occurs, and then the slide stops within a short time, leaving a “rough sliding bed”. Finally, the slope toe area is covered by a thick mud-like deposit.

5.3 Failure modes of the slope

Based on the results of physical model tests, the main failure modes of loess slopes under surcharge-excavation-rainfall conditions are generalized as follows: shallow surface flow slide failure and preferential interface sliding failure (Figure 10).

FIGURE 10

5.3.1 Shallow surface flow slide failure

Rainfall infiltration is primarily characterized by homogeneous vertical percolation, with minimal influence from focused flow along discontinuities, and preferential pathways through fracture zones are not prominent. Precipitation uniformly permeates the surficial loess horizon, leading to rapid saturation of the shallow soil layer. Upon wetting, the uppermost loess matrix undergoes a loss of cohesion and a pronounced reduction in shear strength. Concurrently, surface runoff develops, resulting in hydraulic erosion and the removal of unconsolidated particles, thereby incising rill and gully features. Driven by gravitational forces, the saturated soil mass is prone to shallow planar failure manifesting as rapid, localized flows resembling mudflows. This failure mechanism is defined by its high velocity, modest spatial scale, and broad spatial occurrence. The principal triggering mechanisms include surface layer saturation and erosive runoff generated by short-duration, high-intensity rainfall events, acting in concert with intrinsic material properties of loess such as its friable structure, hydrophilic softening, and tendency towards structural breakdown. Such failures are frequently documented on the slopes of loess tablelands, ridge crests, and mounded topographies.

5.3.2 Preferential interface sliding failure

Rainwater rapidly percolates downward along structural discontinuities such as vertical joints and unloading fractures within the loess, generating preferential flow pathways that circumvent the low-permeability paleosol horizons. Flow restriction at the terminations of these discontinuities elevates pore-water pressure, which facilitates the propagation of existing joints. A perched aquifer forms at the upper boundary of the paleosol due to hydraulic conductivity contrasts, thereby decreasing effective stress. The slip surface evolves along zones of weakness, including joints, fractures, paleosol strata, or deep-seated shearing zones, leading to large-scale, high-intensity, destructive landslides. The development of the deep-seated slip surface is primarily governed by the synergistic influence of joints, rear-edge slope fractures, and internal stratigraphic interfaces with marked permeability differences, and it is more prevalent on high terraces and in regions with prominent paleosol development.

6 Discussion

The experimental results comprehensively delineate the progressive evolution of loess slopes under the representative engineering conditions of “overloading—excavation—rainfall,” capturing the complete transition from initial disturbance to eventual failure and offering several practical engineering insights. Integrating these results with existing research deepens understanding of the experiment’s outcomes and augments knowledge of landslide mechanisms.

6.1 Key characteristics of slope failure

6.1.1 Compound and multi-scale failure modes

The slope undergoes a distinct sequence of compound failures—initial shallow slip on the slope side, followed by basal shear failure, and ultimately coupled shallow-deep slip across the mid-to-upper slope. Shallow failures (thickness <25 cm) predominate during early-stage deformation, while deep shear failures (depth >25 cm) ultimately define the extent of instability, evidencing the “coupled deep and shallow layer, multi-scale interaction” inherent to loess slopes subjected to rainfall infiltration. This failure pattern mirrors the “deep-seated multi-stage rotational–translational” behavior observed in the Beishan landslides of Tianshui, where loess-mudstone contacts similarly demonstrate synchronized shallow and deep layer failure (Wang et al., 2023).

Stress field evolution reveals that engineering disturbances induced by overloading and excavation significantly alter the inherent stress distribution, leading to pronounced stress concentration zones. Rainfall intensifies stress redistribution and promotes its downward migration, with maximum stresses exceeding 150 kPa. Notably, loess mining area studies in western China indicate that volumetric deformation resulting from engineering disturbances modifies the self-weight loading interactions with bedrock, producing evident soil–rock coupling effects (Wang SM. et al., 2025), corroborating this study’s central finding of “engineering disturbance remolding the stress field.” Similarly, coal seam mining in the loess-covered Ordos Basin confirms that disturbances foster stress redistribution in overlying strata, inducing soil degradation and reinforcing the regulatory role of engineering disturbance on slope stress fields.

Through a multi-factor compound testing framework, this research demonstrates that pre-overloading creates an initial stress environment that sets the stage for post-excavation stress redistribution; subsequent rainfall further disrupts equilibrium by softening soil properties, collectively resulting in acute stress concentration at the slope toe. This refines the theoretical model of stress evolution for loess slopes under combined engineering disturbance and rainfall, highlighting the practical necessity of multi-factor coupling analysis in loess plateau engineering applications (Sun et al., 2025).

6.1.2 Rheological behavior and morphology of the landslide mass

When the shallow soil moisture content reaches approximately 42%, the material transitions into a flow-plastic state, and dynamic liquefaction ensues during movement, rapidly shifting from “muddying to high-velocity sliding.” This threshold aligns with infiltration and hazard genesis parameters previously identified in studies on the Jinghe River loess terrace landslides (Hong et al., 2019), reinforcing its regional applicability as a critical indicator for loess flow slides (Sun et al., 2025).

A systematic network of fractures develops: the slope side is characterized by tension cracks 1–2 cm wide, steeply dipping shear fractures permeate the landslide mass, and gently inclined deep shear cracks manifest at the base. Observed dynamic liquefaction can be attributed to the material’s granular, loosely structured nature; engineering disturbance disrupts bonding, while rainfall saturation fills pores and dissolves binders, drastically reducing intergranular cohesion. Under shallow slip-induced agitation, effective stress rapidly dissipates, triggering liquefaction (Hong et al., 2019). Raindrop impact further causes aggregate breakdown and alters pore connectivity, directly affecting infiltration dynamics and mechanical behavior (Zhang et al., 2024)—a micro-scale process that supports the liquefaction mechanism.

This process is consistent with the “pore water pressure-mediated liquefaction in loess” paradigm articulated by Li Dianqing’s group (Wang R. et al., 2025). When critical initial pore water pressure accumulates, the soil undergoes a phase shift from stable creep to catastrophic failure, with microscale shifts in particle contact modes accelerating liquefaction, providing essential quantification for liquefaction triggers in this study.

6.1.3 Spatiotemporal evolution of failure

Failure initiates at the less constrained slope side, propagating to the toe and upslope regions, in agreement with the principle of “preferential failure at stress concentration zones.” (Huo et al., 2020). The entire sliding event, from onset to cessation, lasts only 1–2 min, with the rapid sliding phase lasting less than 1.0 s. These findings underscore the abruptness and rapidity of loess slope failures at elevated moisture contents, consistent with prior rainfall–induced deformation rate studies on unsaturated loess slopes (Sun et al., 2021) showing that intense rainfall considerably shortens the time to failure initiation.

This study advances a cascade model—“shallow softening → mudded shallow sliding → toe shear failure → flow-plastic/liqefied sliding”—that augments and extends previous research. This cooperative mechanism aligns with the “multi-factor nonlinear synergistic effect” revealed by interpretable machine learning models (Sun et al., 2025), wherein the interplay of moderate slope, high rainfall, and elevated soil erodibility markedly increases landslide risk, affirming the generalizability of multifactorial coupling. Furthermore, the cascade process shares developmental features with the multi-phase landslides of Beishan, Tianshui (Wang et al., 2023), suggesting that phased evolutionary behavior is ubiquitous in both natural and engineered settings.

Slope instability does not arise from a single cause but follows a staged process: “overloading induces stress concentration → excavation causes stress redistribution → rainfall triggers softening and pore water pressure buildup.” Engineering disturbance governs proximity to critical instability, while rainfall infiltration emerges as the ultimate trigger. This conclusion is corroborated by findings from LRT construction disturbance sites in Lanzhou showing “early-stage disturbance significantly influences subsequent moisture migration and deformation response in loess” (Liang et al., 2018), emphasizing the primacy of engineering disturbance in the chain of slope failure.

6.2 Multi-field evolution mechanisms in greater depth

6.2.1 Stress field evolution and correlation with existing research

Experimental observations of stress concentration migration and multi-phase response characteristics provide new empirical evidence to enhance loess slope stress evolution theory. Wang Shuangming et al. (Wang SM. et al., 2025) demonstrated that engineering disturbances in the Northern Shaanxi loess mining area significantly alter stress transmission pathways via soil-rock coupling effects—findings that complement this study’s insights on the dynamic adjustment of stress under sequential “overloading-excavation-precipitation” conditions. Furthermore, site investigations of the Tianshui Beishan landslide cluster reveal that stress concentration zones in bedding slope formations typically develop along the loess-mudstone interface (Wang et al., 2023), a pattern consistent with the observed stress accumulation at the slope toe in this study, thereby validating the regional consistency of stress concentration distribution.

6.2.2 Mechanisms of seepage field evolution and pore water pressure variations

The experiments reveal that rainfall infiltration first induces saturation within shallow soil layers, with pronounced lag effects in deep moisture migration and pore water pressure accumulation; 15 h post-rainfall, pore water pressure surges sharply and becomes concentrated in the lower-middle toe of the slope. This corresponds with findings from research on the Lanzhou urban rail disturbance zone (Liang et al., 2018), where disturbed loess under water infiltration exhibited a “slow growth-sudden spike-stabilization” deformation trend, and pre-existing disturbance-induced low void ratios substantially influenced water migration and deformation responses. Rainfall simulation tests on unsaturated loess slopes further confirm variation in infiltration depth under differing rainfall intensities, with water content and soil pressure changes in the slope showing broadly consistent patterns (Sun et al., 2021), thereby substantiating this study’s conclusions on seepage field evolution.

Hong Bo et al. (Hong et al., 2019) revealed that loess permeability is governed by pore structure, and engineering disturbances can modify pore size distribution and permeability anisotropy. This finding explains the observed phenomenon in this study wherein excavation-induced fissuring and overload-induced changes in soil pore structure together create preferential pathways for rainfall infiltration. Rain splash experiments further indicate that surface soil pore connectivity decreases with increasing rainfall intensity (Zhang et al., 2024), a microscale process that influences shallow layer infiltration rates and consequently regulates the “lag-surge” pore water pressure pattern.

Moreover, research by the China Geological Survey underscores that rainfall infiltration leads to a rapid reduction in loess matrix suction, which, in synchrony with accumulating pore water pressure (Zhang et al., 2024), promotes joint propagation and slope instability. The observed shearing crack propagation in slope toe pore water pressure concentration zones in this study concretely manifests this mechanism. Characteristics such as the observed “lag-surge-fluctuation” evolution pattern of pore water pressure and the “shallow-to-deep” extension of slip surfaces furnish valuable references for refining unsaturated soil constitutive and rainfall infiltration models. Experimentally derived parameters, including moisture migration rates and critical water content thresholds, can be directly employed for the calibration and validation of numerical simulations.

6.3 Key engineering insights

6.4 Limitations and future perspectives

Despite its contributions, this study has several limitations. Artificial rainfall experiments cannot fully replicate the short-duration, high-intensity erosive effects observed in natural extreme rainfall events, nor can they adequately simulate the spatial heterogeneity of natural precipitation. The use of model chambers inevitably restricts the pathways of moisture migration and stress distribution, possibly resulting in discrepancies between the patterns of deeper stress states and water movement compared to in-situ conditions. Moreover, natural weak structural features—such as paleosol layers and joint networks—are challenging to accurately reproduce in scaled models, yet these features are widely distributed in natural slopes and play a critical role in slope failure processes (Wang et al., 2023). Additionally, this study did not account for the destructive impact of raindrop splash on the surface soil structure (Zhang et al., 2024), which can influence near-surface infiltration rates and the initiation timing of shallow failure.

Future research should consider larger-scale physical models and incorporate simulations of natural weak structural features to enhance the engineering applicability of the findings. Integrating numerical simulation methods can also help to quantitatively assess the coupled effects of engineering disturbance and rainfall, offering more comprehensive technical support for precise slope hazard prevention and mitigation. Furthermore, the interpretable machine learning framework proposed by Sun Ping et al. (Sun et al., 2025) could be adopted to quantify the contributions and cooperative thresholds of different driving factors, providing new approaches for accurate landslide risk zoning.

7 Conclusion

In conclusion, this study provides critical experimental validation for advancing sustainable land development, enhancing infrastructure robustness, and formulating effective disaster risk reduction frameworks in loess landscapes. The results highlight the imperative to integrate engineering disturbance management with tailored rainfall adaptation strategies within sustainable slope management protocols.

References

• 1

  FanW.DengL. S.YuB.WeiY. N.et al (2025). Failure mechanism and risk mitigation of loess disaster under extreme rainfall. J. Earth Sci. Environ.47 (3), 285–312. 10.19814/j.jese.2025.04023

  • CrossRef
  • Google Scholar

• 2

  FustosI.Abarca-del-RíoR.MardonesM.GonzálezL.ArayaL. R. (2020). Rainfall-induced landslide identification using numerical modelling: a southern Chile case. J. S Am. Earth Sci. 10.1016/j.jsames.2020.102587

  • CrossRef
  • Google Scholar

• 3

  Hoegh-GuldbergO.JacobD.TaylorM.Guillén BolañosT.BindiM.BrownS.et al (2019). The human imperative of stabilizing global climate change at 1.5°C. Science365, eaaw6974. 10.1126/science.aaw6974

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 4

  HongB.DuS. S.LiX. A.WangL.WangS. L.ZhangH. Bet al (2019). Characteristics of permeability and hazard-pregnant mechanism of loess in the southern tableland of jinghe river. Bull. Soil Water Conservation39 (3), 72–77. 10.13961/j.cnki.stbctb.2019.03.013

  • CrossRef
  • Google Scholar

• 5

  HouT. S.JiangX. D.ChenY. (2024). Mechanism of rainfall-induced toppling in loess collapses. Earth Surf. Process Landforms49, 2825–2839. 10.1002/esp.5860

  • CrossRef
  • Google Scholar

• 6

  HuM. L.ZhangX. L.XuW. H.WangZ. W.ChenH. (2024). Influence of dry density and wetting-drying on hydraulic characteristics of compacted loess. J. Chang. River Sci. Res. Inst.41 (8), 128–134. 10.11988/ckyyb.20230377

  • CrossRef
  • Google Scholar

• 7

  HuangY. H.FengW.LiZ. G. (2014). Characteristics and geological disaster mode of the rainstorm happened on July 3, 2013 in Yan'an area of Shaanxi Province. J. Catastrophol29 (2), 54–59. 10.3969/j.issn.1000-811X.2014.02.012

  • CrossRef
  • Google Scholar

• 8

  HuoA.YangL.PengJ. B.ChengY. X.JiangC. (2020). Spatial characteristics of the rainfall induced landslides in the Chinese Loess Plateau. Hum. Ecol. Risk Assess.26 (1), 1–16. 10.1080/10807039.2020.1728517

  • CrossRef
  • Google Scholar

• 9

  KrebichH.Van LoonA. F.SchröterK.WardP. J.MazzoleniM.SairamN.et al (2022). The challenge of unprecedented floods and droughts in risk management. Nature608, 80–86. 10.1038/s41586-022-04917-5

  • CrossRef
  • Google Scholar

• 10

  LiangQ. G.FangJ.ZhangJ. D.ZhangY. J.PuJ. J.WangF.et al (2018). In-Situ soaking test on the disturbed loess site at Lanzhou Metro, Gansu Province, China. J. Geomechanics24 (6), 803–812. 10.12090/j.issn.1006-6616.2018.24.06.083

  • CrossRef
  • Google Scholar

• 11

  LinG. C.LiuW.SuX. (2024). Unraveling the hydraulic properties of loess for landslide prediction: a study on variations in loess landslides in Lanzhou, Dingxi, and Tianshui, China. China Geol.7(2), 291–302. 10.31035/cg2024006

  • CrossRef
  • Google Scholar

• 12

  LiuJ. W.ShenZ. L.HuB.ZhangY. J.OuX.CongK.et al (2025a). Experimental analysis of rainfall-induced shallow landslides: a case study of a loess slope in gaolan County, China. Front. Earth Sci.13, 1613118. 10.3389/feart.2025.161311

  • CrossRef
  • Google Scholar

• 13

  LiuX.DongJ. H.TangC. Q.PanY.ZhaoJ. J.WeiZ. X. (2025b). Instability mechanism of loess-mudstone landslides under rainfall infiltration conditions. Sci. Rep.15, 17591. 10.1038/s41598-025-01887-2

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 14

  LüQ. L.ZhangY. Y.ZhangT. D.ChengL. L.CaiS. D.LiuC. X. (2023). Change of mechanical strength of loess in Ili region under different freeze-thaw cycles and moisture contents. J. Eng. Geol.31 (4), 1269–1281. 10.13544/j.cnki.jeg.2023-0233

  • CrossRef
  • Google Scholar

• 15

  MaS. Y.ShaoX. Y.XuC. (2023). Physically-based rainfall-induced landslide thresholds for the Tianshui area of loess Plateau, China by TRIGRS model. Catena233, 107499. 10.1016/j.catena.2023.107499

  • CrossRef
  • Google Scholar

• 16

  NgC. W. W.ShiQ. (1998). Influence of rainfall intensity and duration on slope stability in unsaturated soils. Q. J. Eng. Geol.31, 105–113. 10.1144/gsl.qjeg.1998.031.p2.04

  • CrossRef
  • Google Scholar

• 17

  PengJ. B.FanZ.WuD.ZhuangJ. Q.DaiF. C.ChenW.et al (2015). Heavy rainfall triggered loess-mudstone landslide and subsequent debris flow in Tianshui, China. Eng. Geol.186, 79–90. 10.1016/j.enggeo.2014.08.015

  • CrossRef
  • Google Scholar

• 18

  PradelD.RaadG. (1993). Effect of permeability on surficial stability of homogeneous slopes. J. Geotech. Eng. Div. ASCE119 (2), 315–332. 10.1061/(ASCE)0733-9410(1993)119:2(315)

  • CrossRef
  • Google Scholar

• 19

  SchiermeierQ. (2011). Climate and weather: extreme measures. Nature477 (7363), 148–149. 10.1038/477148a

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 20

  ShiW. R.LiX.ZengM. J.ZhangB.WangH. B.ZhuK. F.et al (2021). Multi-model comparison and high-resolution regional model forecast analysis for the “7・20” Zhengzhou severe heavy rain. Trans. Atmos. Sci.44 (5), 688–702. 10.13878/j.cnki.dqkxxb.20210823001

  • CrossRef
  • Google Scholar

• 21

  SunP.WangH. J.WangG.LiR. J.ZhangZ.HuoX. T. (2021). Field model experiments and numerical analysis of rainfall-induced shallow loess landslides. Eng. Geol.295. 10.1016/j.enggeo.2021.106411

  • CrossRef
  • Google Scholar

• 22

  SunP.ZhangS.KeC. Y.LiR.SangK. Y.LiK.et al (2025). Evaluation of landslide susceptibility and contribution analysis of key driving factors on the Loess Plateau. J. Geomechanics31 (5), 972–989. 10.12090/j.issn.1006-6616.2025088

  • CrossRef
  • Google Scholar

• 23

  WangH. J.SunP.ZhangS.WangT.RenJ. (2023). Characteristics and slope structure of the Beishan landslide group in Tianshui City. J. Geomechanics29 (2), 236–252. 10.12090/j.issn.1006-6616.2022052

  • CrossRef
  • Google Scholar

• 24

  WangJ. H.ZhangL.ZhaoS.LiG. J.GuoH. P. (2025). Stability analysis of loess slope under heavy rainfall considering joint Effect—Case Study of Jianxi landslide. China. Water17 (22), 3271. 10.3390/w17223271

  • CrossRef
  • Google Scholar

• 25

  WangR.WangS.LiD.KangX.XinP. (2025a). Elevated pore-water pressure regulating dynamic liquefaction of a flow-like landslide in loess. J. Geophys. Res. Earth Surf.130, e2025JF008318. 10.1029/2025JF008318

  • CrossRef
  • Google Scholar

• 26

  WangS. M.SongS. J.WeiJ. B.LiuL.NiuR. LLiY. H.et al (2025b). Study on the soil damage effect of the layered structure of overlying rock series on coal seams in the loess-covered area of northern Shaanxi. J. China Coal Soc.50 (1), 23–42. 10.13225/j.cnki.jccs.YG24.1485

  • CrossRef
  • Google Scholar

• 27

  WuZ.ZhouY.SunP.ShiJ. S.LiuG. G.BaiL. Y. (2017). Laboratory characterization of rainfall-induced loess slope failure. Catena150, 1–8. 10.1016/j.catena.2016.11.002

  • CrossRef
  • Google Scholar

• 28

  WuW. W.XuT.GaoZ. Q.LiuB. (2022). Rainstorm characteristics and flood disaster analysis of “2021.10” in Jinzhong Area, Shanxi Province. China Flood Drought Manag.32 (12), 82–88. 10.16867/j.issn.1673-9264.2022083

  • CrossRef
  • Google Scholar

• 29

  ZhangF.WangG. (2018). Effect of irrigation-induced densification on the post-failure behavior of loess flowslides occurring on the Heifangtai area, Gansu, China. Eng. Geol.236, 111–118. 10.1016/j.enggeo.2017.07.010

  • CrossRef
  • Google Scholar

• 30

  ZhangX.AlexanderL.HegerlG. C.JonesP. D.TankA. K.PetersonT. C.et al (2011). Indices for monitoring changes in extremes based on daily temperature and precipitation data. Wiley Interdiscip. Rev. Clim. Change2 (6), 851–870. 10.1002/wcc.147

  • CrossRef
  • Google Scholar

• 31

  ZhangK.WangS.BaoH. J.ZhaoX. M. (2019). Characteristics and influencing factors of rainfall-induced landslide and debris flow hazards in Shaanxi Province, China. Nat. Hazards Earth Syst. Sci.19, 93–105. 10.5194/nhess-19-93-2019

  • CrossRef
  • Google Scholar

• 32

  ZhangZ. Y.BaoW. X.LiZ.ChenR.GuanX. L. (2024). Experimental study on the failure process of loess slopes in Ili under freeze-thaw and rainfall. KSCE J. Civ. Eng.29, 100100. 10.1016/j.kscej.2024.100100

  • CrossRef
  • Google Scholar

• 33

  ZhaoK.XuQ.LiuF.XiuD.RenX. (2020). Field monitoring of preferential infiltration in loess using time-lapse electrical resistivity tomography. J. Hydrol.591, 125278. 10.1016/j.jhydrol.2020.125278

  • CrossRef
  • Google Scholar

• 34

  ZhouZ. W.WangY. J.Yepes-BellverL.AlcaláJ.YepesV. (2025). Intelligent monitoring of loess landslides and research on multi-factor coupling damage. Geomech. Energy Environ.42, 100692. 10.1016/j.gete.2025.100692

  • CrossRef
  • Google Scholar

• 35

  ZhuangJ. Q.PengJ. B.DuC. H.ZhuY.KongJ. X. (2024). Shallow-Landslide stability evaluation in loess areas according to the revised Infinite slope model: a case Study of the 7.25 tianshui sliding-flow landslide events of 2013 in the Southwest of the Loess Plateau, China. Nat. Hazards Earth Syst. Sci.24, 2615–2631. 10.5194/nhess-24-2615-2024

  • CrossRef
  • Google Scholar

• 36

  ZongJ. M.ZhangC. J.LiuL. F.LiuL. L. (2024). Modeling rainfall impact on slope stability: computational insights into displacement and stress dynamics. Water16 (4), 554. 10.3390/w16040554

  • CrossRef
  • Google Scholar