Accelerating Effect of Vegetation on the Instability of Rainfall-Induced Shallow Landslides

Abstract

Rainfall-induced shallow landslides are widespread throughout the world, and vegetation is frequently utilized to control them. However, in recent years, shallow landslides have continued to frequently occur during the rainy season on the vegetated slopes of the Loess Plateau in China. To better probe this phenomenon, we considered vegetation cover in the sensitivity analysis of landslide hazards and used the transient rainfall infiltration and grid-based regional slope stability (TRIGRS) model to quantitatively describe the impacts of different types of vegetation cover on slope stability. Based on the rainfall information for landslide events, the spatiotemporal distributions of the pore water pressure and the factor of safety of the vegetated slopes were inverted under the driving changes in the soil properties under different vegetation types, and the average prediction accuracy reached 79.88%. It was found that there was a strong positive correlation between the cumulative precipitation and the proportion of landslide-prone areas in woodland covered by tall trees, grassland covered by shrubs and grasses, and cultivated land. The highest landslide susceptibility, which has the greatest potential to hasten the occurrence of rainfall-induced landslides, is found in woodland with tall trees. Therefore, this paper proposes the promoting relationship between vegetation and landslide erosion, which provides a new scientific perspective on watershed management to prevent shallow landslide disasters and manage and develop watershed vegetation.

1. Introduction

Landslides, as one of the most catastrophic natural disasters around the world [1,2,3,4], cause serious casualties and huge property losses [5,6,7]. They were responsible for at least 17% of all fatalities from natural hazards worldwide in the period of 1994–2013 [8,9]. Statistics from the Geological Bulletin of China show that the number of landslides in China accounted for 48.9% of geological disasters throughout the country in 2021, which are more likely to be triggered during high-intensity rainfall or prolonged rainfall in the mountainous regions [10,11,12,13]. In recent decades, the increase in anomalous precipitation has significantly increased the probability of landslides [14,15]. Meanwhile, the research on rainfall-induced landslides has gradually become a focus and challenge in current disaster prevention and management schemes [16,17]. Additionally, soil erosion, vegetation reduction, and other ecological environmental problems due to landslide disasters induced by heavy rainfall are also increasingly prominent [18]. Overall, attention should be paid to landslide hazards in terms of both ecosystem service modes and climatic zones, and disaster prevention and mitigation measures should be developed and optimized periodically. Thus, it is of great significance to conduct susceptibility analysis and studies on the mechanisms of rainfall-induced landslides.

The occurrence of landslides is closely related to the infiltration of precipitation into the unsaturated zone [19]. The most intuitive internal embodiment of slope damage is the increase in the pore water pressure in the soil [20]. Natural factors such as rainfall, land cover, and the type of vegetation affect slope stability and promote landslide susceptibility changes [21,22,23,24,25]. Vegetation, as an important factor that affects slope stability [26], can control slope instability by changing the hydrological and mechanical properties of the slope body. Thus, it has a direct impact on landslide disasters [27,28,29]. The failure mechanism of a vegetation-covered slope is always a subject worthy of further study.

At present, the vegetation coverage on the Loess Plateau has significantly increased due to the implementation of several vegetation restoration plans [30], which has significantly reduced the possibility of soil erosion [31,32,33]. However, the slope with high vegetation coverage is still prone to triggering mass shallow landslides in the rainy season [34,35]. The traditional belief is that vegetation restoration can effectively slow down the infiltration of surface runoff, and the reinforcement effect of vegetation roots can increase slope stability, effectively curbing the occurrence of soil and water disasters and promoting the transformation of the ecological environment [36,37]. In spite of this, the effect of vegetation restoration on the reciprocal mechanism between ecology and landslides remains highly controversial [38,39,40]. Thus, the mechanism of landslide disasters and the relationships with the impact of vegetation are far from sufficiently understood. In view of this, quantifying the mutual relationship between soil and water disasters and vegetation and discussing the relationship between vegetation and landslide occurrence are necessary components of landslide prevention and control research.

The modeling of shallow landslide triggering, which is one of the primary means of landslide prevention and control research, has been always a process of continuous development and improvement. At present, although there are knowledge-driven models, data-driven models, and physical deterministic models that have been used in regional landslide evaluation modeling [41,42,43], most modeling of the probability of shallow landslides has mainly adopted the deterministic method [44,45]. The deterministic method refers to the use of the specific risk component as the average value, so as to output univariate results [44], for instance, the shallow slope stability (SHALSTAB) model [46], the transient rainfall infiltration and grid-based regional slope-stability (TRIGRS) model [47], and the shallow landslides in-stability prediction (SLIP) model [48]. The latter has been demonstrated to have a good effect in assessing the spatial heterogeneity of the soil saturation, while the SHALSTAB model cannot represent the spatial variability of parameters on a small scale [49]. In recent decades, the development of probabilistic models and statistical methods has improved the model performance of landslide probability quantification and interpretation [44]. Furthermore, statistical methods do not provide a clear interpretation of the physical processes [49], while probabilistic methods can consider the physical processes and the probability distribution of one or more input parameter values and can quantify the reliability of the results [50]. As a physics-based regional model, the TRIGRS model precisely exemplifies these characteristics and has a strong predictive ability when combined with actual landslide failure mechanics and geotechnical parameters [1,42]. The quantitative evaluation of single or multiple parameters leading to landslide events has significant advantages. Therefore, in recent years, the TRIGRS model, as a grid-based regional slope-stability physical deterministic model of transient rainfall infiltration, has been widely used in slope stability analysis [51,52,53,54,55,56].

In order to quantitatively analyze the effects of vegetation cover on landslides, in this study, the impacts of different types of vegetation cover on the soil were considered to predict and evaluate landslide susceptibility using a physical model of the regional slope stability (TRIGRS). It is generally believed that vegetation affects slope stability through the interception and evapotranspiration of tree crowns and the reinforcement of roots. During extreme storms, the former can be ignored while the latter can affect the stability of the slope by affecting the hydrological and mechanical properties of the soil [57,58]. Therefore, in this study, we hypothesized that vegetation may be conducive to slope failure during prolonged rainfall because of specific failure mechanisms. Moreover, we regarded the vegetated slope as a whole and used rainfall data, lithology, soil mechanics, and hydrological factors to analyze the stability of the vegetated slopes. Finally, through verification using data for real landslides, we explored the relationship between vegetation and landslides. The results of this study provide valuable insights for understanding the importance of vegetation factors in landslide disasters in the future.

2. Study Area and Data

The Loess Plateau spreads across the middle reaches of the Yellow River in China and spans approximately 640,000 km². The Haojiata landslide group is located in the southern part of the Loess Plateau, the northwestern suburb of Fuxian County, Shaanxi Province (Figure 1a). It lies in the cross-transition zone between the gully region and the hilly gully region of the Loess Plateau, and it is a part of the Ordos Platform of the North China Platform. The bedrock is composed of a Mesozoic sedimentary rock series and Cenozoic Neogene laterite, which is extensively covered by Upper Pleistocene Malan loess [59], and the groundwater type is mainly Cenozoic unconsolidated lithologic pore water [60]. The loess hills in the study area are mostly elongated beams with steep beam slopes and active gully erosion [61]. The loess has a small particle size and is dominated by silt loam. Its texture is homogeneous, loose, and porous, and it has a large permeability [62]. In addition, the loess has large porosity, and channels for water infiltration form easily, which causes the soil erosion rate on the Loess Plateau to be higher [63]. Moreover, rainfall is plentiful in this region, with an average annual precipitation of 500–600 mm. The rainfall is concentrated in July, August, and September, accounting for 70% of the annual precipitation. This uneven distribution generally aggravates the possibility of soil erosion. Relevant statistics show that in July 2013, there were more than 8135 shallow landslides triggered by extreme precipitation events in the Yan’an region, among which rainfall-induced geological disasters accounted for 50% of the total historical disasters [63].

The study area is 17.8 km². Figure 1b shows the location and digital elevation model (DEM) of the watershed, where the coverage of the forest and grassland is approximately 68.7%, the slope angles are mainly 10–30° with a maximum slope angle of 53°, and the relative height difference is 331 m. By comparing satellite images acquired before and after the heavy rainfall in July 2013 and field investigations, a total of 772 shallow landslides were interpreted (Figure 1b), all of which were mass geological disasters caused by the heavy rain throughout Fuxian County in July 2013. In the satellite image with a resolution of 0.5 m from the IKONOS Satellite, we observed that after experiencing this rainstorm, the slope was seriously unstable, several landslides were triggered, and the scars were clearly visible. (Figure 1c,d). Based on field investigations of typical landslides, it was concluded that the landslides in this area were mainly shallow landslides, which developed under the loess-covered slope section, and the landslides had obvious regional characteristics.

The DEM used in this study was derived from the United States Geological Survey (USGS) of Shuttle Radar Topography Mission (SRTM) data with a spatial resolution of 30 m × 30 m (Figure 1b). The historical image used for the extraction of typical landslide boundaries was taken by the IKONOS Satellite, which was the first commercial remote sensing satellite worldwide to provide high-resolution satellite imagery on 24 July 2013. The satellite image is extremely clear and has a maximum ground resolution of 1 m, while the cloud cover is only 8.0%. We projected the image to the WGS_1984_UTM_Zone_49N coordinate system for further precise identification of landslides using ArcGIS (Figure 1c,d).

3. Methodology

In recent years, the frequency of rainfall-induced landslides on the vegetated slopes of the Loess Plateau in China has increased. The TRIGRS model software, which takes into account rainfall-triggered events that determine slope stability conditions, as well as, environmental characteristics such as topography, lithology, soil mechanics, and hydrology, was used to assess the slope susceptibility. To probe the relationship between the types of vegetation cover and landslide susceptibility, we modeled and examined the stability of different vegetated slopes under extreme rainfall patterns.

3.1. TRIGRS Model

TRIGRS, the transient rainfall infiltration and grid-based regional slope-stability model for the timing and distribution of rainfall-induced shallow landslides [1], regards the soil slope as a two-layer system including the saturated zone with a capillary fringe zone and the unsaturated zone connecting to the soil surface (Figure 2) [47].

Complex rainfall processes can be effectively solved using the analytical solution for vertical penetration in the Richards Equation [64]. In addition, this model can be extended to the infiltration into the unsaturated layer above the water table [65,66]. When the infiltration seeping down into the water table exceeds the maximum amount that gravity can discharge, the excess water is compared with the water table or the available pore space on the edge of the capillaries for slope-stability analysis. The pore pressure of water after rainfall infiltration into the soil can be expressed as follows:

$$\begin{array}{l}\psi (Z,t)=\left(Z-d\right)\beta +2{\displaystyle \sum _{n=1}^N\frac{{}_{{{\displaystyle I}}_{nz}}}{{{\displaystyle K}}_s}}H\left(t-{{\displaystyle t}}_n\right){\left[{{\displaystyle D}}_1\left(t-{{\displaystyle t}}_n\right)\right]}^{\frac{1}{2}}{\displaystyle \sum _{m=1}^{\infty }\{ierfc}\begin{array}{l}\left[\frac{\left(2m-1\right){{\displaystyle d}}_{LZ}-\left({{\displaystyle d}}_{LZ}-Z\right)}{2{\left[{{\displaystyle D}}_1\left(\mathrm{t}-{{\displaystyle t}}_n\right)\right]}^{\frac{1}{2}}}\right]\\ +ierfc\left[\frac{\left(2m-1\right){{\displaystyle d}}_{LZ}-\left({{\displaystyle d}}_{LZ}-Z\right)}{2{\left[{{\displaystyle D}}_1\left(\mathrm{t}-{{\displaystyle t}}_n\right)\right]}^{\frac{1}{2}}}\right]\end{array}\}\\ -2{\displaystyle \sum _{n=1}^N\frac{{}_{{{\displaystyle I}}_{nz}}}{{{\displaystyle K}}_s}}H\left(t-{{\displaystyle t}}_{n+1}\right){\left[{{\displaystyle D}}_1\left(t-{{\displaystyle t}}_{n+1}\right)\right]}^{\frac{1}{2}}{\displaystyle \sum _{m=1}^{\infty }\{ierfc}\begin{array}{l}\left[\frac{\left(2m-1\right){{\displaystyle d}}_{LZ}-\left({{\displaystyle d}}_{LZ}-Z\right)}{2{\left[{{\displaystyle D}}_1\left(\mathrm{t}-{{\displaystyle t}}_{n+1}\right)\right]}^{\frac{1}{2}}}\right]+\\ ierfc\left[\frac{\left(2m-1\right){{\displaystyle d}}_{LZ}-\left({{\displaystyle d}}_{LZ}-Z\right)}{2{\left[{{\displaystyle D}}_1\left(\mathrm{t}-{{\displaystyle t}}_{n+1}\right)\right]}^{\frac{1}{2}}}\right]\end{array}\}\end{array}$$

(1)

where ψ is the pressure head; t is time; d is the steady-state groundwater level depth in the Z direction; β = cos2 δ (Iz/Kz) _LT; K_s is the hydraulic conductivity; I_nz is the initial infiltration rate; D₁ = D₀/cos2 δ; N is the total number of time periods; D₀ is the hydraulic diffusivity; ierfc(η) = 1/√π ∗ exp (−η²) − η erfe (η), and erfc (η) is the complementary error function.

The TRIGRS model combines the physical attributes with the spatial distribution to effectively reproduce the triggering process of shallow landslides induced by rainfall [47]. The slope stability and landslide susceptibility assessments of an area are evaluated by calculating the factor of safety (FS) of each grid element using Equation (2):

$$FS(Z,t)=\frac{\tan {\phi }^{\prime }}{\tan \delta }+\frac{c^{\prime }-\psi (Z,t){\gamma }_w\tan {\phi }^{\prime }}{{\gamma }_{s}Z\sin \delta \cos \delta }$$

(2)

where Z is the soil depth; t is the time; φ′ is the internal friction angle of the soil; δ is the angle of inclination; c′ is the available soil cohesion; ψ is the pressure head determined by time and depth; and γ_w and γ_s are the unit weights of the water and soil, respectively.

According to previous studies [67,68], the factor of safety can be divided into five levels: Unstable (FS < 0.75), not stable (0.75 < FS < 1), potentially unstable (1 < FS < 1.25), relatively stable (1.25 < FS < 1.5), and stable (FS > 1.5). An FS < 1 indicates landslide-prone areas while an FS > 1 indicates stable areas.

3.2. Parameters

The topography, hydraulic characteristics, and geotechnical properties are indispensable for discussing the relationship between surface vegetation cover and rainfall-induced landslides. First, the topography of the study area was described using the DEM, which provided the preliminary topographic data. Second, based on the 30 m resolution global land cover product (GlobeLand30) data derived from (http://www.globallandcover.com/ (accessed on 28 September 2021)), the types of land cover in the study area were obtained. Woodland was dominated by tall trees, grassland was a mixture of small shrubs and grasses, and cultivated land was dominated by artificial use (Figure 3a). Then, based on the premise of scientific and feasible principles, the layout of the field sample points for the slopes covered by different types of vegetation was designed, including tall trees, herbs mixed with small shrubs, and cultivated land samples. The soil sample collection and in situ infiltration tests were implemented. Three groups of in situ infiltration tests were conducted at each sample site, and three groups of soil samples were collected to ensure the validity of the parameters. The soil samples were sealed and transported to the laboratory (Figure 3b,c). The geotechnical and hydraulic parameters of the slope soil covered with different types of vegetation were obtained through these experiments. During the field investigation, it was discovered that there were obvious landslide scars in grassland and woodland (Figure 3d,e).

In addition, the soil thickness parameter, which refers to the thickness of the soil cover, is a particularly important piece of information in the simulation. Multiple methods have been proposed to estimate the spatial pattern of the soil thickness [69]. It is undoubtedly reasonable to simulate the soil thickness as a constant value [68,70], but due to the spatial heterogeneity of the soil thickness, this method often has limitations [71,72]. In order to avoid the restriction of a single soil layer thickness [1], a linear relationship between the slope angle and the soil layer thickness [73] was adopted to describe the soil layer thickness distribution, which can be expressed as Equation (3):

$$H=-0.0045L+4.5$$

(3)

where L is the slope angle and H is the soil depth corresponding to the different slope angles.

Finally, the saturated hydraulic diffusivity (D₀) is a multiple of the hydraulic conductivity, i.e., between 2 and 500 times the Ks [24]. In this study, the value of the saturated hydraulic diffusivity was set as D₀ = 10 Ks. As is shown in

Table 1. Input parameters of the TRIGRS model.

Parameters	Symbol	Woodland	Grassland	Cultivated Land
Soil cohesion	c′ (kPa)	14.35	5.77	7.77
Soil friction angle	ϕ′ (°)	3.72	3.09	2.05
Hydraulic conductivity	KS (m/s)	6.4 × 10−4	4.7 × 10−5	4.6 × 10−5
Hydraulic diffusivity	D0 (m/s)	6.4 × 10−3	4.7 × 10−4	4.6 × 10−4
Unit weight of soil	γs (kN/m3)	25.5	21.5	23.2
Saturated volumetric water content	θs (m3/m3)	0.12	0.11	0.12
Residual volumetric water content	θr (m3/m3)	0.07	0.09	0.07

, the average saturated conductivities of woodland, grassland, and cultivated land were 6.4 × 10⁻⁴ m/s, 4.7 × 10⁻⁵ m/s, and 4.6 × 10⁻⁵ m/s, respectively, according to the results of the in situ permeability experiments. The hydraulic diffusivities of woodland, grassland, and cultivated land were 6.4 × 10⁻³ m/s, 4.7 × 10⁻⁴ m/s, and 4.6 × 10⁻⁴ m/s, respectively. In addition, the initial depth of the groundwater level was described according to the correlation between the thickness of the soil layer (Equation (3)) [43], and its input form was the spatial distribution. Moreover, the soil bulk density was 19.5–25.5 kN/m³, 21.5–24.3 kN/m³, and 20.7–23.2 kN/m³ in the woodland, grassland, and cultivated land, respectively, with mean values of 22.3 kN/m³, 22.6 kN/m³, and 22.5 kN/m³. Furthermore, Table 1 also shows the additional parameters required for the simulation.

3.3. Rainfall Data

Rainfall is the main external trigger of landslides [2]. Slope instability usually occurs during the rainfall period, and transient rainfall has a great influence on the change in the pore water pressure perpendicular to the slope surface [42,50]. In July 2013, Fuxian County experienced significant rainfall compared with previous years, and the total amount reached 422.9 mm, breaking the historical extreme maximum monthly precipitation value. On July 22 alone, the precipitation reached 151.9 mm (County Meteorological Station), and by 12:00, the accumulated rainfall had reached 149.8 mm (Figure 4a). The rainfall data from 24:00 on July 21 to 12:00 on July 22 was selected as the transient rainfall for the simulation (Figure 4b).

3.4. Evaluation of Model Accuracy

The landslide sensitivity map generated using the TRIGRS model was used to assess the response of the stabilities of slopes with different types of vegetation cover to rainfall. The landslide inventory created via the visual interpretation of satellite images and field investigations was combined to comprehensively determine the grid level of the landslide points and evaluate the reliability of the simulation results. The reliability of the prediction model is usually assessed by comparing the position of the actual landslide with the predicted results [42,50], and the use of the composite index LR_class [74] to evaluate the accuracy of the model simulation results is relatively reliable [75,76]. The landslide ratio for each predicted FS class was based on the relationship of each FS category to the actual total number of landslide points, and it can be defined by Equations (4) and (5) according to the predicted percentage of area in each FS category.

$${{\displaystyle LR}}_{class}=\frac{\%\mathrm{of}\mathrm{contained}\mathrm{slope}\mathrm{failure}\mathrm{locations}\mathrm{in}\mathrm{each}FS\mathrm{class}}{\%\mathrm{of}\mathrm{predicted}\mathrm{areas}\mathrm{in}\mathrm{each}FS\mathrm{class}}$$

(4)

$$\%L{R}_{class}=\frac{L{R}_{class}^i}{{\displaystyle {\sum }_{i=1}^{n}L{R}_{class}^i}}$$

(5)

The LR_class index considers the possibility of landslide occurrence in the stable zone and avoids excessive prediction of the landslide trend. Therefore, it is relatively reliable for evaluating the results of the numerical simulation.

4. Results

In this study, the physical model was used to simulate and predict the stability and landslide susceptibility of slopes covered by different types of vegetation cover. Through satellite images and field investigations, it was found that the vegetation coverage in the study area was high, and the main type of vegetation cover was woodland with tall trees. In addition, grassland with small shrubs and grasses and cultivated land were present. Thus, how the slopes with different types of vegetation cover responded to rainstorms and what role vegetation played in landslide disasters are questions worth exploring. The TRIGRS landslide susceptibility evaluation model was applied to predict the stability and susceptibility of vegetation-covered slopes under long-term rainfall conditions.

4.1. Evaluation of Pore Water Pressure in the Slopes Covered with Vegetation

The pressure state of the slope was determined as the internal embodiment of the landslide failure [25], and the stability-driven change in the soil slopes with different types of vegetation cover were evaluated based on the changes in the matric suction. The main reason for the occurrence of rainfall-induced landslides is the reduction or even loss of suction (or negative pore water pressure) in the soil, which leads to the reduction of the shear strength of the soil [77,78]. Suction changes in the soils can take three forms: (A) Enhancement from positive to higher positive (excess pore water pressure), (B) from negative to zero or positive (matric suction loss), or (C) increasing but remaining negative (suction change) [79].

Using the TRIGRS model, the variations in the pore water pressure (PWP) were obtained based on the hourly rainfall intensity and the geotechnical parameters in the study area. Figure 5 shows the variations in the suction at different depths in three plant-covered slopes. The pore water pressure distribution diagram shows that the negative pore water pressure changed to a positive pore water pressure and increased continuously, which was manifested as a loss of matrix suction and the appearance of excess pore pressure.

Figure 5a shows the spatial distribution of the positive pore pressure at different soil depths in woodland-covered slope soils (hereinafter referred to as PWP). In the initial state (at 24:00 on the 21st), the minimum PWP value was −3.5 m and the PWP in the study area was mainly concentrated between 0.4 and 0.6, indicating a mildly dry state. At this time, the slope soils were relatively stable. At the end of the simulation (at 12:00 on the 22nd), the PWP in the woodland slope in the red area with negative pore pressure rapidly changed from a negative value to a positive value and the maximum value reached 1.3 m. This indicates that the continuous rainfall resulted in changes in the pressure state of the slope, which was manifested as a decrease in the shear strength of the soil, the gradual disappearance of the matric suction (increased pore water pressure), the soil becoming nearly saturated, and a decrease in the stability of the slope, under which a landslide could easily occur. Most of the PWP values were 0.4–0.7 m. At this moment, the regional pore water pressure increased significantly compared with the initial value, and the excess pore water pressure caused by rainfall infiltration was significant.

As shown in Figure 5b, the positive pore water pressure of the soil covered by grassland was concentrated within the range of 0.6–0.7 m after the simulation of the real rainfall conditions. Compared with the slope soil covered by woodland, the possibility of the occurrence of a landslide was higher due to the weakening of the shear strength of the soil. Compared with the slope covered by grassland and woodland, the pore water pressure in most of the cultivated areas (Figure 5c) was in the range of 0.7–1.3 m at 12:00 on the 22nd, and a few red areas still exhibited suction changes at the beginning and end of the simulation. There are two explanations for this phenomenon. One is that the suction loss caused by the seepage was relatively insignificant and still resulted in a relatively stable state due to the small slope of the cultivated land, the deep soil layer, and the low initial groundwater level. Second, the rainfall in this region mostly formed surface runoff that could not infiltrate, so the soil was not sensitive to the expected disturbance of the rainfall.

4.2. Cumulative Changes in Slope Instability on the Rainfall Time Scale

The contribution of the types of vegetation cover to the shear strength of the soil is very important to the formation of potential landslide disasters. Analysis of the relationship between rainfall and landslide instability probability is helpful to clarify the landslide susceptibility of vegetation-covered slopes on the rainfall time scale. The solid line shows the change in the soil stability under different vegetation types. The band represents the total prediction band, which shows the fitting performance. As shown in Figure 6, the predicted landslide ratio was positively correlated with the rainfall for all three types of vegetation cover. In other words, the probability of the occurrence of a shallow landslide in all vegetation-covered soils increased as the accumulated rainfall increased. Furthermore, the probability of landslide occurrence in the slopes covered by different types of vegetation exhibited different characteristics under the premise of long-term rainfall infiltration.

In Figure 6, under the simulated real rainfall pattern, the soil covered by woodland (represented by c) exhibited obvious changes. Under the low-rainfall conditions, the landslide-prone area was smaller for grassland-covered slopes (b and c), which may have been because the rainfall infiltration depth was smaller and the dense vegetation roots played a certain role in anchoring the soil. The slope maintained high shear strength until 9 h after the rainfall began, and the landslide susceptibility increased slowly, especially in the woodland with tall trees. However, the stability of the slope significantly decreased as the rainfall accumulation increased (Figure 6). There was an inflection point after 9 h of rainfall, at which time the instability speed of the slope in the woodland with tall trees accelerated and the probability of landslide occurrence increased. At the end of the simulation, the cumulative area of the landslide in the woodland was the highest at approximately 6.3%. This indirectly indicates that the accumulated rainfall at this time was the time when vegetation contributed the most to the shear strength of the soil, and rainfall was the leading factor inducing the landslides before this time. To a certain extent, this can be considered the maximum threshold for rainfall-induced landslides. After exceeding this threshold, the vegetation gradually becomes the key factor leading to the occurrence of landslides in a large area, and the vegetation accelerates the occurrence of landslides. At this moment, the area of the landslide-prone area increased the fastest, reaching 0.21 km²/h, which was 1.5 times the landslide-prone area of the grassland and 5 times the landslide-prone area of the cultivated land. This also indicates that woodland area with better vegetation coverage is more likely to promote the occurrence of shallow landslides under heavy rainfall. Figure 6 also shows the relationship between the area prone to landslides and the rainfall in the cultivated land. It is predicted that the landslide area will increase slowly and linearly at a rate of 0.04 km²/h, accounting for only approximately 1.2% in the end, making it the most stable. In summary, it is concluded that within a certain range, the better the vegetation cover is, the easier it is to promote the occurrence of a landslide.

4.3. Spatial and Temporal Evolution of Landslide Hazards

Figure 7 shows the slope stability output of the TRIGRS model. The slopes with different types of vegetation cover are disturbed by prolonged rainfall to different degrees. Figure 7a shows the spatial and temporal distributions of the FS in the woodland with tall trees under the initial and final rainfall conditions. FS < 1 indicates the unstable slope areas and FS < 1.25 indicates potentially unstable slope areas where landslides are most likely to occur. The unstable area increased from 0.74 km² to 1.09 km², i.e., an increase of 0.35 km². Figure 7b shows the stability changes in the slope covered by grass. After the simulation, the landslide-prone area was also concentrated in the slope area, and the area of the landslide was 0.82–1.06 km², i.e., an increment of 0.24 km². Compared with the former, the increment of the potential landslide area was the smallest for the cultivated land under the conditions of long-term rainfall, only 0.05 km², as shown in Figure 7c. Generally, the vegetation root system of the woodland was quite deep, but the grid number of the landslide sites increased the fastest and landslide disasters were more likely to occur in woodland areas.

In conclusion, when the small watershed after vegetation restoration is subjected to high-intensity rainfall or long-duration rainfall, the vegetation slope with better vegetation coverage is more likely to trigger large-scale mass shallow landslides.

4.4. Verification

In this study, the LR_class index was used to analyze the reliability of the landslide susceptibility simulation results. Figure 8 shows the relationship between FS and the cumulative % of LR_class, where FS = 1 is the threshold between instability and stability. The simulation results showed that when FS = 1, the degree of coincidence between actual landslide and predicted landslide locations is 74.66%, 80.09%, and 84.77% for grassland, woodland, and cultivated land, respectively, and the average degree of coincidence is 79.88%, which is sufficient to show that the simulation results are reliable. That is, when vegetation-covered slopes are affected by heavy rainfall, the high-vegetation-cover areas are more prone to the occurrence of landslides. This is essentially consistent with the results of Wang et al. [80].

5. Discussion

Vegetation cover plays an undisputed role in slope stabilization; however, there is a negative correlation between rainfall intensity and vegetation interception under extreme rainfall: The greater the rainfall intensity is, the weaker (or even nonexistent) the interception effect of the vegetation [5,57]. Similarly, during extreme storms, vegetation evapotranspiration is very weak due to the very wet weather conditions [58]. In light of this, we ignored the interception and evapotranspiration of the vegetation and mainly considered the mechanical function of the vegetation in influencing the development of shallow landslides rather than the hydrological function, which can be ignored during extreme storms [58]. Based on the modeling and back analysis of the slope’s landslide sensitivity after heavy rainfall, the effects of vegetation cover on slope stability during extreme storms were evaluated based on a combination of laboratory tests and field data.

As one of the important causes of shallow landslides, forests can balance soil moisture and enhance soil stability [81,82]. Several previous studies have also reported the important contribution of vegetation roots to slope stability [83,84]. Moreover, the patches with higher internal friction angles and cohesion seem to have a higher shear-strength-to-shear-stress ratio, and the higher shear strength improves the slope stability [37,85,86]. This is similar to the results obtained by Keijsers [87] using the TRIGRS model in the Taiwan district [87]. In general, vegetation roots provide extra soil cohesion and help to prevent material from falling [81]. In view of this, we paid special attention to the cohesiveness of the root–soil composite layer, which is closely related to the mechanical function of trees and has a great influence on the sensitivity analysis [88]. As shown in Table 1, the soil cohesion values of the woodland and grassland were 14.35 kPa and 5.77 kPa, respectively, indicating that the shear strength of the woodland was higher than that of the grassland and its stability was relatively strong, so the reinforcing effect of the root system of the vegetation played a certain role. During the first 9 h of rainfall, the vegetation roots stabilized the soil by strengthening the shear strength of the soil, and the slope covered by woodland was less prone to landslides than grassland. At this moment, the mechanical effect of the vegetation still played a role in slowing down the occurrence of landslide disasters.

In contrast, it was found that in a forest containing many large trees with very stout old roots in the study area, the overland flow becomes turbulent at the bases of the stems and buttresses, which may increase the probability of landslides. After 9 h, the contribution of the woodland vegetation roots to the shear strength of the soil reached the maximum value, and the increase in the landslide-prone area accelerated (Figure 7). It can be concluded that under continuous rainfall, woodland with a better root system is more likely to promote the occurrence of landslides, The lateral root depth may play a dominant role in the landslide erosion intensity [5]. Due to the limited depth of root reinforcement, its contribution to the shear strength of the soil will gradually decrease or even disappear completely [89]. It can be considered that the vegetation and its roots will aggravate the occurrence of landslide erosion when precipitation reaches or exceeds the maximum critical intensity to induce landslide disasters. In addition, the soil permeability coefficient of the woodland was the largest (K_s = 6.4 × 10⁻⁴ m/s) (Table 1), which led to a rapid increase in the soil water content during the same rainfall event (Table 1). A crisscrossed root network increases the soil’s porosity and gives priority to the formation of infiltration channels [90], rapidly reducing the soil’s shear strength and increasing soil-sliding dynamics [56,91], promoting the occurrence of landslide disasters. This viewpoint [34,92] is consistent with the shallow landslide density being higher in vegetated areas.

Moreover, the mechanism between eco-hydrological hazards and vegetation is a complex process [41]. Tree vegetation is also prone to death, and the soil reinforcement capacity of dead tree roots decreases as their strength and stiffness decrease [93]. When trees die, the favorable effects on the soil water potential regime (root water absorption) are immediately lost [94]. There are many voids in the internal root system of neighboring trees due to the gradual decomposition of the roots [95], which decreases the tensile strength of the roots and the total shear capacity provided by the roots to the soil. Over time, the risk of landslide occurrence will progressively increase [96,97,98]. At this time, the decaying roots increase the soil permeability and the chance of rainwater infiltration, leading to positive pore water pressure and widening of the scar area of the landslide [99]. In addition, the primary root extension depth of tall trees is significantly deeper than that of shrubs and herbaceous vegetation, and the depth of rainfall infiltration deepens as the duration of rainstorms increases. The soil weight associated with tree vegetation roots increases the weight acting on the failure surface, thus reducing the slope stability [34,100].

In summary, these detrimental influences of vegetation on the soil may balance out the weak support provided by the roots. Continuous rainfall causes the root–soil composite layer to progressively become saturated, increasing the weight of the loess and decreasing its effective strength until a landslide eventually occurs [63,78]. This inevitably implies that vegetation may promote shallow landslides, that is, the soil in woodland structures is more sensitive to the expected disturbance of heavy rainfall.

6. Conclusions

In this study, the influences of the types of vegetation cover on loess landslides and soil properties were investigated using the TRIGRS model, combined with field investigations, indoor experiments, and simulation prediction. It was found that rainfall was the main trigger for shallow landslides, and vegetation promoted the occurrence of landslides under prolonged rainfall conditions.

In terms of the mechanical parameters of the soil and the partial simulation results, the contribution of the vegetation roots to the soil cohesion was undeniable and it controlled the landslide occurrence. However, vegetation accelerated the occurrence of landslides under extremely heavy rainfall patterns. The results revealed that under the same rainfall intensity and duration, the probabilities of landslides in woodland, grassland, and cultivated land were 6.7%, 6.0%, and 1.3%, respectively, with increments of 0.35 km², 0.24 km², and 0.05 km². This fully demonstrates that the effect of vegetation on landslide control is limited.

These results are in good agreement with the satellite images and field observations. In the study area, the woodland-covered soil slopes contained the most landslide scars, while the grass-covered slopes contained few landslide scars. The degree of coincidence between the actual landslides and predicted landslides was 79.88%, and the results were reliable. Specifically, under extreme rainfall, future landslides may be more sensitive to the disturbance of a woodland structure.