A Regional Landslide Stability Analysis Method under the Combined Impact ofRainfall andVegetationRoots inSouthChina

)e aim of this study was to develop a regional landslide stability analysis method considering the combined impact of rainfall and the roots of vegetation in densely vegetated areas. A typical mountainous watershed in the Nanling National Nature Reserve of South China was chosen as the study area. First, the unmanned aerial vehicle (UAV) method was used to obtain surface element information including topography, vegetation, and landslides. Five main plant species were identified. )e RipRoot model was then used to calculate the additional cohesion of these five plant species, and the relationship between the root systems of the different plant species and the soil shear strength was subsequently revealed. Finally, the root cohesion was introduced into the stability index mappingmodel (SINMAP), and the receiver operating characteristic curve (ROC)method was used to calculate the accuracy of slope stability when considering only soil cohesion as well as the composite cohesion of both soil and roots.)e results showed significant differences in the root cohesion of different plants in the study area and a significant increase in the calculation accuracy (from 90% to 95.6%) when root cohesion was considered in the landslide stability calculation. )ese study results not only enrich theoretical studies on the impact of vegetation roots on landslide stability but also provide a scientific support for preventing disasters in mountainous landslide-prone areas.


Introduction
Landslides occur due to the deformation of rock and soil on slopes under the influence of rainfall, human activities, slope topography, geology, vegetation, and other factors [1]. Rainfall (especially heavy rainfall) is the main cause of landslide disasters generally causing a regional mass effect that can result in catastrophic consequences within a short time frame, such as huge economic losses, heavy casualties, and significant damage to the ecological environment [2][3][4]. According to a survey of 290 counties and cities in China, 90% of the total number of landslides were induced by heavy rainfall [5]. Such landslides are thus representative of serious geological disasters that occur in China and globally.
It is widely known that the main aim of landslide mitigation is to determine the probability and location of the potential occurrence of shallow landslides. Landslide stability analyses are conducted in this respect, and statistical analyses and physical model methods are the main method employed for rainfall-induced landslides. Statistical methods are based on the use of long-term rainfall and disaster monitoring data [6][7][8][9][10]. However, shallow landslides usually occur as short sudden outbreaks with few signs of early deformation, and historical records and data on the landslide occurrence time and corresponding cumulative rainfall are often lacking. erefore, it is difficult to utilize statistical samples to form a statistically significant data system that can be employed to conduct stability analyses, and the lack of such data may result in large errors. Quantitative research can be conducted using physical methods based on a hydromechanical model, and the migration of precipitation on the surface and its infiltration into the ground can be clearly described. Such methods can be used to analyze and observe changes in pore water pressure of soils that occurs due to precipitation. In addition, specific changes in slope and the corresponding safety factor value can be calculated using the slope stability model. e main physical models used to evaluate landslides associated with rainfall are Shallow Slope Stability model (SHALSTAB) [11,12], the Stability Index MAPping model (SINMAP) [13,14], and the Transient Rainfall Infiltration and Grid-based Regional Slope Stability model (TRIGRS) [15][16][17][18][19]. e most recognized and widely used models are currently the SINMAP [13,[20][21][22][23][24][25]. e theoretical basis of using the SINMAP model arises from the fact that it is a large-scale slope stability model utilizing the terrain moisture index obtained from the steady-state hydrological model based on grid units that can be quickly and conveniently calculated.
During landslide formation, each slope occurs in relation to the incline at which it is located, and each slope unit has a relatively clear topographical boundary. Landslides are controlled by the geological elements associated with the slope unit. erefore, slope is the most ideal unit for studying landslides. However, the SINMAP is still grid unit model and is not extended to slope unit. is study innovatively extended the SINMAP model to slope unit to calculate the stability coefficient of each slope under different rainfall conditions.
Vegetation plays an important role in slope stability, and roots have a mechanical effect in increasing slope stability. To date, studies on the effects of the root fixing of soil have focused mainly on single tensile properties of different plant types [26,27]. In addition, studies have focused on the shearresistance characteristics of root-soil composites [28,29], mechanical reinforcement effect of the root fixing of soil [30,31], and mechanism of root slope protection using a soil model [32][33][34]. However, most previous research on root systems has been conducted on small or single-plant scales, whereas for regional slopes with a high vegetation coverage, few studies have investigated the roots of vegetation and their effects on slope stability. Furthermore, most research has focused mainly on conducting laboratory tests on the tensile strength of the root systems of different tree species, and related studies evaluating the application of vegetation roots to slope stability zoning on a slope scale are relatively rare. erefore, this study innovatively develops a regional landslide stability assessment method that considers the combined impact of different rainfall types and the roots of vegetation on landslide stability in densely vegetated areas. e specific objectives were to provide a scientific and reliable theoretical basis for the mitigation of landslide disasters by (i) proposing a method based on the use of UAV remote sensing data of the surface elements of small watersheds (such as the terrain, vegetation, and landslides); (ii) proposing an improvement of the physical SINMAP model by introducing root cohesion into the landslide stability analysis model (SINMAP) and employing the slope unit as the basic unit; and (iii) producing a landslide susceptibility zoning map associated with different rainfall intensities. To achieve these objectives, a watershed in the core area of the Nanling National Nature Reserve in South China was selected as the study area. e area has a subtropical monsoon climate with abundant rainfall and a high vegetation coverage. e surfaces of slopes are mostly covered by residual strong weathered rocks with loose structures and fissures, which renders the area extremely prone to landslide disasters.

Characterization of Study Area.
e Nanling Mountains within the Nanling National Nature Reserve of South China are the largest mountain range in southern China; they act as an important natural geographical boundary between tropical and subtropical regions and they separate four provinces (study location is shown in Figure 1). e mountains also form a watershed dividing between the Pearl River and the Yangtze River, and they play an important ecological role in South China. e study area has a typical subtropical climate with average total annual rainfall of 1705 mm (multiyear average precipitation ranges from 1452.1 to 2154.5 mm) ( Figure 2). e horizontal zonal soil in the study area is red soil. Owing to the large difference in the relative elevation of the mountains and the vertical distribution of moisture and temperature conditions, the distribution of soil types differs considerably with altitude. Yellow soil is distributed below 1500 m, latent yellow soil is distributed within 1500-1800 m, and meadow soil exists at elevations higher than 1800 m. Because of their unique climatic conditions, the Nanling Mountains are an oasis that contain the largest remaining area of unspoiled land at the same latitude in the world. e various types of forest vegetation within the mountain range include subtropical evergreen broad-leaved forests, coniferous and broad-leaved mixed forests, coniferous forests, and hilltop lowland forests. In the study area, Yushania basihirsuta is mainly distributed on peaks and ridges above 1500 m where the soil is comprised of granite or sandy shale weathered meadow soil and is thin and rocky with a soft, thick, humus and moss layer on the surface. Trees are extremely low in height and most are small trees or shrubs. Coniferous and broad-leaved mixed forests include the species Cunninghamia lanceolata, Pinus kwangtungensis, and Fagus longipetiolata and are distributed mainly on ridges and on steep slopes at 1000-1500 m. Constructive species include Pinus massoniana and Cunninghamia lanceolata. e soil at this altitude is mountainous yellow soil covered by a thick humus layer. Plant roots are concentrated mainly on the 0-40 cm surface. e mountainous part of the study area has steep unstable slopes, strongly cut river valleys, thick weathered soil, strong erosion, and complex climatic conditions. It is prone to mountain disasters, such as collapses, landslides, and mudslides. e area percentage of different slope class is shown in Figure 3. Mountain disasters, such as collapses and landslides, occur mainly in areas where the slopes are steep, where differences in terrain exist, and where the slope of the terrain ranges from 30°to 40°, and particularly in regions where extensive weathering and rock fragmentation has occurred. Such regions include slopes, the steep portions of river valleys, and slopes along reservoirs and other constructed infrastructure. However, the scale of collapses and landslides is predominantly small, and mass movements tend to be shallow.

2.2.
e Flowchart of the Method. e flowchart of this method is shown in Figure 4. e process can be divided into five parts: the topography and vegetation extraction, the vegetation root investigation, the slope unit division based on GIS, the root cohesion calculation, and the slope stability analysis with SINMAP model under the combined impact of rainfall and vegetation roots.

e Topography and Vegetation Extraction.
is study used high-resolution UAV remote sensing images obtained in August 2017 as the data source to define the topography and vegetation. In this respect, the UAV was used to acquire stereo image pairs with a high overlap in the study area.
ree-dimensional (3D) relative point cloud data were then used to extract 3D point cloud, which were subsequently used to generate a digital elevation model (   Advances in Civil Engineering spatial resolution of 0.5 m. e species of flora were then visually interpreted according to the results of vegetation classification research obtained in the study area [35], which were subsequently verified by intensive field investigations. e main plant species in the study area were finally determined as Yushania basihirsuta, Cunninghamia lanceolata, Pinus massoniana, Pinus kwangtungensis, and Fagus longipetiolata.

2.2.2.
e Vegetation Root Investigation. Vegetation root data were obtained via field sampling, and sampling sites were established at different elevations. Plant diameter and height and the numbers of roots with different root diameters were measured. Landslide identification was also conducted using positive UAV remote sensing images. e precipitation data employed were rainfall monitoring data spanning 1950 to 2016 obtained from two rainfall monitoring stations in the study area. e maximum daily rainfall at the peak station was 105.3 mm and that at the foot of the mountain was 207.6 mm.

e Slope Unit Division Based on GIS.
e slope unit is often used to analyze slope stability with respect to landslides. We extended the cells in the original SINMAP model into slope cells, and the stability of each slope under the vegetation coverage and rainfall exposure was then calculated. e method used to divide slope units was based on a DEM generated from the UAV remote sensing results. In this respect, based on the source cutting method and ridgelines and valley lines, slope units were divided using the GIS platform.

2.2.4.
e Root Cohesion Calculation. Soil reinforcement research is based on the mechanical properties of plant roots and root-permeated soil is a composite material that has enhanced strength [36]. Vertical deep roots are anchored to stable rock layers through a shallow weathering layer, and shallow roots convert the soil on a slope into a root-soil composite material that improves soil strength. is composite material improves the cohesion of slope soil and increases the safety and stability of the slope. e slip surface 18  of shallow landslides in forested areas mostly occurs between the layer with the largest root density and the lower soil body, and such a phenomenon proves that root systems have a controlling effect on shallow landslides. Wu [37] and Waldron [38] made an important contribution to the theory of soil reinforcement and determined that soil strength (c r ) is a function of where c r is the cohesion due to roots (kPa), T r is the mean tensile strength of roots (kPa), ϕ is the friction angle of the soil (degrees), θ is the angle of shear distortion (degrees), A r is the area of roots in the plane of the shear surface, A s is the area of the shear surface, and RAR is the root area ratio. e root area ratio and tensile strength are the most important mechanical indicators of the strength of a root system. However, Pollen and Simon [39] found that models based on equation (1) tended to overestimate root reinforcement.
e RipRoot model considers different root system tolerances and the process of gradual root system destruction, and it therefore provides a more reasonable root system and associated enhancement of soil strength. e RipRoot model was therefore selected in this study to estimate the additional cohesion due to roots.
e Bank Stability and Toe Erosion (BSTEM) model proposed by the Agricultural Research Service of the United States is a comprehensive model used to predict the slope foot erosion rate and slope stability. erefore, the effects of vegetation on slope stability were simulated using the root reinforcement model and the RipRoot model within the BSTEM model. e RipRoot model within the BSTEM model was employed to calculate the root cohesion of vegetation in the study area. When calculating root cohesion, it is usually necessary to determine the species and ages of trees, the relationship between the root diameter and tensile strength, and the distribution of different sized root systems. Tree species of trees were identified by remote sensing, and their ages and root diameters (Table 1) were obtained through a field investigation. According to Burylo et al. [40], the tensile strength of roots, T r , decreases with increasing root diameter and follows a simple power-law equation, where T r is the tensile strength of roots (kPa), α and β are empirical values depending on the species, and D is the root diameter. e larger the root diameter, the greater the tensile strength.

e Slope Stability Calculation with SINMAP Model under the Combined Impact of Rainfall and Vegetation Roots.
Deterministic models based on landslide physics processes are widely used to predict the spatiotemporal stability of shallow landslides [41]. To evaluate the spatial distribution of the stability of shallow landslides caused mainly by rainfall, Borga et al. [42] constructed a quantitative evaluation model of shallow landslide stability (Stability Index MaPping, SINMAP) based on the following: a DEM, a coupled steady-state hydrological model, and an infinite slope stability model [13,[20][21][22][23][24][25]. SINMAP is a large-scale slope stability model that uses topographic humidity index data obtained from the steady-state hydrological model, slope information obtained from the gridded DEM, maps of the effective catchment area and other GIS maps, and field investigation data. e model can be used to classify and evaluate surface stability [43]. e basic unit of the SINMAP model is a grid cell, which is quick and easy to calculate. However, the square grid in this respect poorly reflects the mechanism of disaster formation. Each slope has a well-defined terrain boundary, and each slope has significantly different characteristics from its adjacent slopes.
As the topographic features within slopes are consistent, "the slope" is an idea unit for use in studying landslides. erefore, in this study, we extended the cell of the original SINMAP model into a slope cell.
We used the SINMAP model to calculate the factor of safety (F s ), which relates to the quotient of the upward and downward sliding forces on a weak structural plane parallel to the ground surface. e formula for calculating F s is as follows [44]: F s � c r + c s + cos 2 θρ s D − D w + ρ s g − ρ w g D w tan φ Dρ s g sin θ cos θ + D w ρ w g sin θ cos θ , where c r is the cohesion of plant roots (N/m 2 ), c s is the cohesion of soil (N/m 2 ), θis the terrain gradient, ρ s is the moist soil density (kg/m 3 ), g is the gravitational acceleration (9.81 N/kg), D is the soil thickness (m), ρ w is the water density (kg/m 3 ), φ is the soil friction angle, and D w is the depth of the aquifer from the surface (m).

Its dimensionless expression follows
where is the terrain gradient, φ is the soil friction angle, r is the water density divided by the soil density, C is the dimensionless cohesion, and ω is the topographic wetness index, which was calculated from the TOPMODEL model using the following formula: where q is the effective rainfall, a is the ratio of the area of rain collection (the catchment area divided by the drainage width), and T is the transmissibility (T � KZ cos θ), where Advances in Civil Engineering 5 K is the permeability coefficient and Z is the soil thickness. Surface runoff occurs when the topographic wetness index (ω) is greater than 1, and the maximum value of the topographic wetness index ω is 1.
when F s > 1, the slope is stable; when F s � 1, the slope is in a state of limit equilibrium; and when F s < 1, the slope is unstable. a and θ were extracted from the DEM, r depends on the density of the soil in the study area, C is the value of cohesion used in this study, and φ and T were calculated from the results obtained from laboratory tests and literature surveys. e range of the study area was small, and the internal friction angle, permeability coefficient, and the soil density values were typically identical within it. e values of all parameters are listed in Table 1.
To study the effect of root cohesion on slope stability, we defined the root coefficient (R) as the influence of root cohesion on stability according to the SINMAP model: where R is the root coefficient, c r is the cohesion of plant roots (kPa), θ is the terrain gradient, ρ s is the moist soil density (kg/m 3 ), g is gravitational acceleration (9.81 N/kg), and D is soil thickness (m).

e Field Instigation Result of Root Diameters.
We measured the plant diameter and height and the numbers of roots with different root diameters of five dominant plant species in the field. e root depths of the main plant species in the study area were found to range from 30 cm to 40 cm, and the various vegetative characteristics of the roots are listed in Table 2.

e Result of Slope Unit Division.
e method used to divide slope units was based on a DEM generated from the UAV remote sensing results. In this respect, based on the source cutting method and ridgelines and valley lines, slope units were divided using the GIS platform. e average terrain gradient (θ) and the ratio of the area of rain collection (a) of each slope unit are shown in Figures 5(a) and 5(b), respectively.

e Result of Root Cohesion.
According to existing research, the relationships between root tensile strength and the root diameters of five dominant plant species in the study area are shown in Table 3.
By considering variations in the mean root tensile strength (T r ) with root diameter, root cohesion for the five vegetation types was calculated using equations (1) and (8) via the RipRoot model, and the results are given in Table 4. e root cohesive forces produced by the five plant species were significantly different and varied from 6.36 kPa (Yushania basihirsuta) to 17.6 kPa (Fagus longipetiolata). Such variations relate to differences between the root area ratios and tensile strengths.

Slope Stability under Different Rainfall Conditions and When considering and Not considering Root Cohesion.
Based on the characteristics of the study area and daily rainfall data for the past 50 years, it was assumed that the effective rainfall amounts were 50 mm, 105.3 mm, and 207.6 mm. e SINMAP model was used to calculate the stability factor for each slope under these three rainfall amounts. Steady-state classification was then conducted according to the stability coefficient. e slope stability coefficient obtained using equation (9) is an absolute value, and it directly reflects the degree of slope stability. It is of note that the classification of safety level standards for slope stability factors is currently not uniform. Rainfall erosion in the study area is extremely strong and it results in large slope disturbances. According to the geological environment and the degree of human activity in the study area, the safety standard for slope stability was integrated.
According to the calculated F s value, slope stability was divided into five grades: unstable (F s < 1.00), potentially unstable (F s is 1.00-1.25), basically stable (F s is > 1.25-1.50), increasingly stable (F s is > 1.50-1.75), and stable (F s > 1.75), and the results are shown in Figures 6-8 with respect to different amounts of precipitation. With an increase in the amount of effective rainfall, the proportion of unstable regions that are F s < 1.00 gradually increases.
When root cohesion was not considered, the proportion of such slope units increased from 1.08% under 50 mm of precipitation to 11.7% at 207 mm, and when root cohesion was considered, the proportion of slope units increased from 11.5% under 50 mm of rainfall to 34.36% under 207 mm. According to the distribution of past disaster sites, there has been a significant increase in the proportion of landslide disaster points distributed in unstable regions in relation to the effective rainfall amounts. It is thus evident that landslides in this area have been significantly affected by rainfall.
In this study, the receiver operating characteristic (ROC) analysis method was used to analyze and evaluate the model results under the three rainfall conditions, both in consideration of roots and without the consideration of roots. e sensitivity index (where a landslide area is identified as an unstable slope) was used as the ordinate, and a specificity of 1− (determining an area as stable, unstable, and misjudgments) was used as the abscissa. e size of the area under the curve (AUC) was used to measure the accuracy of the model's calculations. In this respect, a larger area indicates a stronger ability to predict and evaluate, and a curve closer to the upper-left corner indicates a higher model accuracy. When considering and not considering the root system effect, F s < 1 was determined as unstable and assigned a value of 1, and F s > 1 was determined as stable and assigned a value of 0. Additionally, a slope for which a landslide had occurred was assigned a value of 1, and a unit in which a landslide had not occurred was assigned a value of 0. 6 Advances in Civil Engineering e results of disaster conditions and the safety factors associated with the three rainfall conditions were imported into the SPSS software to conduct the ROC analysis, and the results are shown in Figure 9. As shown in the figure, introducing root cohesion into the model improved the prediction accuracy of the model, which was highest when the rainfall was 103.5. e AUC surface area when considering the contribution of roots was 0.962, and that without considering the contribution of roots was 0.915.

Effect of Root Cohesion on the Safety Factor.
e influence of root cohesion on the factor of safety (F S ) of an infinite slope considering the three precipitation levels is shown in Figure 10. It is evident that root cohesion increases slope stability. However, the safety factor does not change linearly with the change in root cohesion, but it rather shows a gradual changing trend. It can be inferred from formula (10) that, under the same surface soil and hydrological environment, the coefficient of the action of the root system is mainly affected by the slope of the terrain. In addition, when the cohesive force of the root system increases to a certain threshold, the rate of increase in the safety factor begins to slow. In the interval of 0-45°, the coefficient of action of the root system is positively correlated with the slope; that is, the larger the slope with the same vegetation type, the more effective the effect on slope stability. e root system had the largest coefficient of action with a slope of 45°, but when the slope was in the range of 45-90°, the root coefficient of action had a negative correlation with slope; that is, the greater the slope with the same kind of vegetation, the weaker the stability of the slope.
e root system coefficient (R) was found to be affected by the slope and the soil thickness. Under the same soil type and soil thickness, the influence of vegetation on slope stability was controlled mainly by the degree of slope. e safety factor (Fs) did not change linearly with changes in root cohesion. When the cohesive force of the root system increased to a certain threshold, the rate of increase in the safety factor began to slow.

Discussion.
e forest coverage rate in South China exceeds 90%, and plant roots play a very important role in the stability of landslides. At present, most studies on the stability of high-coverage vegetation landslides in South China do not consider the role of roots on the stability of landslides, nor do      they introduce the root-soil composite cohesion into the analysis of landslide stability, which seriously affects the calculation accuracy of landslide stability in areas with high vegetation coverage in South China. is study explored the influence of the additional tensile stress and cohesion of the root system on the strength of the soil, so as to analyze the influence of the root system on the factor of landslide safety (Fs). We modified the SINMAP model, extended its grid unit to slope unit, calculated the slope stability coefficient of each slope unit in the study area with and without roots, and used the landslide inventory data and the ROC curve to validate the study results. e results show that, considering the root-soil composite cohesion, the calculation accuracy of landslide stability can be improved to a large extent. e mechanical properties of the root-soil complex mainly include shear strength, elastic modulus, ultimate tensile strength of roots, etc. ere are many influencing factors, which can be roughly divided into two categories: (1) soil factors, such as soil water content, cohesion, internal friction angle; and (2) plant factors, such as tree age, root diameter, root length density, total root length, root area ratio, and root diameter class. Different scholars chose different plants, so the generalization of their research results to other plants has yet to be found in commonality. In addition, the study area is different, and the geological conditions and the nature of the soil are also different, which is also a problem that needs to be paid attention to in future research.  e model in this paper is mainly aimed at the main factors affecting the stability of the landslide in South China, mainly considering the root diameter of the vegetation, the strength of the soil, the thickness of the soil, the permeability coefficient of the soil, the effective rainfall, and other factors. e parameters were simplified in the model operation process, and slight differences in the study area were neglected to some extent, which would have had some influence on the accuracy of the evaluation results. In a future study, the water and soil parameters will be assigned through experimental methods, and the intrinsic relationship between vegetation and slope stability under different rainfall conditions will be further elucidated.

Conclusions
e objective of this study was to propose an improved physically based SINMAP model for predicting rainfallinduced landslides in densely vegetated areas, with the aim of providing a support for disaster mitigation in mountainous areas. e unit of the SIMMAP model was improved using a slope unit that can reflect landslide disasters, facilitate the rapid calculation of results, and provide calculation results that are comparatively more realistic. Several conclusions can be made, and these are presented as follows.
(1) e root cohesion force was found to differ considerably between the main plant species in the study area. Root cohesion is mainly related to root diameter; as the root diameter increases, the root crosssectional area also rapidly increases, with the proportion of lignification of the root system; therefore, the maximum tensile strength of the roots rapidly increases. However, the same plant species has a different effect on slope stability under different geological conditions. (2) Under three different rainfall conditions, the stability characteristics of each slope unit were calculated using the improved SINMAP model while considering vegetation root cohesion; the results spatially reflected the slope stability under different conditions. e calculation results also showed that the accuracy of the improved SINMAP model improved when root cohesion was considered. In addition, as the amount of rainfall increased, there was an increase in the area where potentially dangerous landslides could occur. (3) e Nanling Mountains are characterized by high vegetation coverage, and the roots of plants form a continuous root net in the shallow soil beneath the forest, forming a special composite material. e soil has a relatively low-strength soil (matrix), but the root system (fiber) has a high tensile strength and is embedded within the soil matrix; this is important for controlling shallow landslides. e root cohesion derived from the root-soil interaction model in this study is only an approximate value of the actual amount, but it reveals the potential of using vegetation to protect slopes. is study thus provides a theoretical basis for planting vegetation to prevent disasters by protecting slopes.

Data Availability
e data used to support the findings of this study are available from the corresponding author upon request.

Conflicts of Interest
e authors declare that there are no conflicts of interest regarding the publication of this paper.