Spatial and Temporal Evolution of the Inﬁltration Characteristics of a Loess Landslide

: Inﬁltration plays an important role in inﬂuencing slope stability. However, the inﬂuences of slope failure on inﬁltration and the evolution of inﬁltration over time and space remain unclear. We studied and compared the inﬁltration rates in undisturbed loess and disturbed loess in di ﬀ erent years and at di ﬀ erent sites on loess landslide bodies. The results showed that the average initial inﬁltration rate in a new landslide body (triggered on 11 October 2017) were dramatically higher than those in a previous landslide body (triggered on 17 September 2011) and that the inﬁltration rates of both landslide types were higher than the rate of undisturbed loess. The initial inﬁltration rate in the new landslide body sharply decreased over the 4–5 months following the landslide because of the appearance of physical crusts. Our observations indicated that the inﬁltration rate of the disturbed soil in a landslide evolved over time and that the inﬁltration rate gradually approached that of undisturbed loess. Furthermore, in the undisturbed loess, both the initial and quasi-steady inﬁltration rates were slightly higher in the loess than in the paleosol, and in the previous landslide body, the inﬁltration rate was highest in the upper part, intermediate in the middle part, and lowest in the lower part. This study can help us to better understand the evolution process of inﬁltration in undisturbed loess, previous landslides, and new landslides.


Introduction
The infiltration and the strong collapsibility of loess are the main exogenic and endogenic factors, respectively, that contribute to loess landslides and the secondary failure of landslide bodies [1][2][3][4][5][6]. In recent years, loess-covered areas have been prone to rainfall-induced landslides due to variable climatic conditions and anthropogenic activities [7][8][9]. Moreover, large-scale landslide bodies tend to exhibit secondary failure due to an increased infiltration rate [10][11][12]. This phenomenon worsens the situation in landslide-prone areas.
The infiltration rate can be used to evaluate how quickly water enters the soil, and a high infiltration rate tends to greatly decrease the shear strength of the soil due to increases in the water content and pore water pressure [13][14][15][16][17]. Loess is characterized by well-developed vertical jointing and strong collapsibility, which significantly influence the infiltration process [3,18]. The collapsibility of loess tends to worsen due to the destruction of the soil structure during the landslide process, triggering

Study Area
The Bailu Loess Tableland (263 km 2 ) is located 5 km east of Xi'an city, Shaanxi Province, China (Figure 1a,b). The length and width are 28 km and 6-10 km, respectively. The stratigraphic unit comprises deposits of Malan loess, which was formed in the Late Pleistocene and is very sensitive to water because of its collapsibility [30], and Lishi loess, which was formed in the Middle Pleistocene and consists of interlayered loess and paleosol layers [3]. The Bailu Loess Tableland area has a semi-humid continental monsoon climate and is in the warm temperate zone, with a mean annual precipitation of 560 mm and a mean temperature of 13.1 • C [31]. The rainfall is mostly concentrated in the form of heavy precipitation in the summer and autumn, and the intensities vary from 1 mm/h to 20 mm/h. The Jiwachang Brickyard Landslide (34 • 16 00" N, 109 • 05 49" E) is located in the northwest corner of the Bailu Loess Tableland (Figure 1b). This landslide is characterized by a high speed and a long runout. The Jiwachang Brickyard Landslide was triggered on 17 September 2011 by heavy precipitation. In addition, the excavation of the brickyard contributed to this landslide to some extent, and this landslide destroyed the brickyard and killed 32 people [7]. The QR code ( Figure 1g) shows a panorama of the Jiwachang Brickyard Landslide generated by an Unmanned Aerial Vehicle (UAV).

Field Investigation
Based on the field investigation, the typical landslide N3# was selected (hereinafter referred to as NLB) for the infiltration experiments. The characteristic parameters of the PLB, recent landslides and flowslides, were recorded during the investigation (Table 1, Figures 1c-h and 2). The sedimentary thicknesses of L1 (the first loess layer in the Malan loess), S1 (the first paleosol layer in the Lishi loess), and L2 (the first loess layer in the Lishi loess) are 8 m, 2 m, and 6 m, respectively. In total, approximately 90% of the PLB was covered by grass (Figure 2a), and biological crusts ( Figure  2c), moss crusts ( Figure 2d) and physical crusts were found to be abundant on the soil surface. However, grass and soil crusts were not observed on the NLB on 24 October 2017 (Figure 2b,e). The diameter of the fragmented loess blocks was 5-120 cm, but the diameter of most of the clods was less than 10 cm, which means that the structure of the soil in the NLB was severely disturbed.

Field Investigation
Based on the field investigation, the typical landslide N3# was selected (hereinafter referred to as NLB) for the infiltration experiments. The characteristic parameters of the PLB, recent landslides and flowslides, were recorded during the investigation (Table 1, Figures 1c-h and 2). The sedimentary thicknesses of L1 (the first loess layer in the Malan loess), S1 (the first paleosol layer in the Lishi loess), and L2 (the first loess layer in the Lishi loess) are 8 m, 2 m, and 6 m, respectively. In total, approximately 90% of the PLB was covered by grass (Figure 2a), and biological crusts (Figure 2c), moss crusts ( Figure 2d) and physical crusts were found to be abundant on the soil surface. However, grass and soil crusts were not observed on the NLB on 24 October 2017 (Figure 2b,e). The diameter of the fragmented loess blocks was 5-120 cm, but the diameter of most of the clods was less than 10 cm, which means that the structure of the soil in the NLB was severely disturbed.

In Situ Infiltration Rate and the Physical Index Measurement
A single-cylinder permeameter (S-CP) (diameter of 4.8 cm, length of 10 cm) was used to measure the quasi-steady infiltration rate (QIR) and the initial infiltration rate (IIR) of the soil [13,32,33]. During the in-situ infiltration experiments, the vegetation was trimmed to the soil surface before inserting the S-CP, and animal burrows were avoided when choosing the experiment site. Several measurements may be needed to determine the QIR. The entire process required approximately 1-3 h for each experiment site. When the infiltration experiment was finished, 5 soil samples were

In Situ Infiltration Rate and the Physical Index Measurement
A single-cylinder permeameter (S-CP) (diameter of 4.8 cm, length of 10 cm) was used to measure the quasi-steady infiltration rate (QIR) and the initial infiltration rate (IIR) of the soil [13,32,33]. During the in-situ infiltration experiments, the vegetation was trimmed to the soil surface before inserting the S-CP, and animal burrows were avoided when choosing the experiment site. Several measurements may be needed to determine the QIR. The entire process required approximately 1-3 h for each experiment site. When the infiltration experiment was finished, 5 soil samples were collected by slowly pushing a core ring (dimension of 5 cm, volume of 100 cm 3 ) into the soil near the experiment site. The samples were then dried at 105 • C for 12 h in the laboratory and were used to measure the antecedent moisture content, bulk density, and porosity of the soil. The IIR and QIR were calculated using the following equation [34]: where i is the IIR or QIR, VW is the volume of infiltrated water, t is the time required for the water to infiltrate, and A is the area of the cylinder (A = πr 2 , r = 2.4 cm).
The bulk density (g/cm 3 ) was calculated using the following equation [34]: where Bd is the bulk density, ρ is the density of wet soil, m 0 is the weight of wet soil, v is the volume of the core ring (100 cm 3 ), and w is the moisture content (%). The porosity (%) was calculated using the following equation [34]: where P is the porosity, Bd is the bulk density, ρ d is the specific gravity, as measured in the laboratory.

Experimental Design
To compare the differences in the infiltration rate among the UDL, PLB, and NLB, a series of experiments were conducted on 18 to 26 October 2017, in the L1, S1, and L2 soil layers (Figure 3a), the upper, middle, and lower parts of the PLB (Figure 3b), and the upper and lower parts of the NLB (Figure 3c). Flat surfaces were selected for the infiltration experiment, and the distances between the experimental points on the UDL, PLB, and NLB were 5 m, 50 m, and 5 m, respectively. It should be noted that the QIR of the NLB was difficult to measure during the experiment due to the strong collapsibility of the fragmented loess. Therefore, only the IIR of the soil was measured on the NLB. Successive infiltration experiments were conducted to identify changes in the infiltration rate in the NLB on 14 March 2018, and 17 July 2018.

Statistical Analysis
The differences in the IIR, bulk density and porosity in the UDL, PLB, and NLB were evaluated with one-way ANOVA (p < 0.05) after testing for homogeneity of variance. The results in this study are expressed as the mean ± standard error of the mean (SEM). An independent t-test (p < 0.05) was used to evaluate the differences in the results of the infiltration experiments conducted on the NLB on 14 March 2018, and 17 July 2018. All statistical analyses were performed with SPSS 17.0 software

Statistical Analysis
The differences in the IIR, bulk density and porosity in the UDL, PLB, and NLB were evaluated with one-way ANOVA (p < 0.05) after testing for homogeneity of variance. The results in this study are expressed as the mean ± standard error of the mean (SEM). An independent t-test (p < 0.05) was used to evaluate the differences in the results of the infiltration experiments conducted on the NLB on 14 March 2018, and 17 July 2018. All statistical analyses were performed with SPSS 17.0 software (SPSS, Chicago, IL, USA).

The Spatial Differences in Infiltration Characteristics across Sites
The different parts of the UDL, PLB, and NLB are characterized by large differences in the infiltration rates ( Figure 5). In the UDL, the IIR was highest in soil layer L1 (mean ± SEM = 5.8 ± 0.9 mm/min), intermediate in soil layer L2 (4.9 ± 1.8 mm/min), and lowest in soil layer S1 (2.6 ± 0.7 mm/min) ( Figure 5), but the differences were very small. In the PLB, the IIR was highest in the upper part (20.9 ± 4.0 mm/min), intermediate in the middle part (16.4 ± 5.2 mm/min), and lowest in the lower part (11.3 ± 3.8 mm/min) ( Figure 5). In the NLB, the IIR was higher in the lower part (968.3 ± 68.1 mm/min) and lower in the upper part (742.6 ± 89.4 mm/min) ( Figure 5). Furthermore, the QIR was highest in L1, intermediate in L2, and lowest in S1 ( Figure 5) in the UDL. In the PLB, the QIR was highest in the upper part, intermediate in the middle part, and lowest in the lower part ( Figure 5).

The Spatial Differences in Infiltration Characteristics across Sites
The different parts of the UDL, PLB, and NLB are characterized by large differences in the infiltration rates ( Figure 5). In the UDL, the IIR was highest in soil layer L1 (mean ± SEM = 5.8 ± 0.9 mm/min), intermediate in soil layer L2 (4.9 ± 1.8 mm/min), and lowest in soil layer S1 (2.6 ± 0.7 mm/min) ( Figure 5), but the differences were very small. In the PLB, the IIR was highest in the upper part (20.9 ± 4.0 mm/min), intermediate in the middle part (16.4 ± 5.2 mm/min), and lowest in the lower part (11.3 ± 3.8 mm/min) ( Figure 5). In the NLB, the IIR was higher in the lower part (968.3 ± 68.1 mm/min) and lower in the upper part (742.6 ± 89.4 mm/min) ( Figure 5). Furthermore, the QIR was highest in L1, intermediate in L2, and lowest in S1 ( Figure 5) in the UDL. In the PLB, the QIR was highest in the upper part, intermediate in the middle part, and lowest in the lower part ( Figure 5). part (20.9 ± 4.0 mm/min), intermediate in the middle part (16.4 ± 5.2 mm/min), and lowest in the lower part (11.3 ± 3.8 mm/min) ( Figure 5). In the NLB, the IIR was higher in the lower part (968.3 ± 68.1 mm/min) and lower in the upper part (742.6 ± 89.4 mm/min) ( Figure 5). Furthermore, the QIR was highest in L1, intermediate in L2, and lowest in S1 ( Figure 5) in the UDL. In the PLB, the QIR was highest in the upper part, intermediate in the middle part, and lowest in the lower part ( Figure 5).  Notably, the average IIR in the NLB was almost 200 times and 60 times higher than those in the UDL and PLB, respectively. Unsurprisingly, the rate at which water penetrated into the cracked soil was much higher than that into UDL and the seven-year-old landslide body. A comparison of the PLB and the UDL reveals a small but significant increase in the IIR, and the average IIR in the PLB was 4.4 times higher than that in the UDL (p < 0.05).

Physical Parameter Differences across Sites
Bulk density and porosity are the basic physical properties of soil. These properties have a direct influence on the soil infiltration properties and solute migration [35][36][37][38]. Based on the results generated for the UDL, PLB, and NLB ( Figure 6), the UDL had the highest soil bulk density and the lowest porosity, while the NLB had the lowest soil bulk density and the highest porosity. The disturbed soils (PLB and NLB) exhibited a large and significant (p < 0.05) increase in porosity compared to the UDL (loess-paleosol sequences). Among the disturbed soils, there was a small and insignificant (p = 0.074) increase in the PLB compared with the NLB.
influence on the soil infiltration properties and solute migration [35][36][37][38]. Based on the results generated for the UDL, PLB, and NLB (Figure 6), the UDL had the highest soil bulk density and the lowest porosity, while the NLB had the lowest soil bulk density and the highest porosity. The disturbed soils (PLB and NLB) exhibited a large and significant (p < 0.05) increase in porosity compared to the UDL (loess-paleosol sequences). Among the disturbed soils, there was a small and insignificant (p = 0.074) increase in the PLB compared with the NLB.

Temporal Evolution of the Infiltration Rate in the Recent Landslide
To study the temporal evolution and the influence of physical crusts on the infiltration rate in the recent landslide, successive infiltration experiments were conducted in the upper part and lower part of the NLB on 14 March 2018 and 17 July 2018 ( Table 2). The distance between the crusted and scalped experiment sites was 0.4-0.5 m. The results showed that the average IIR was dramatically greater on 26 October 2017 (no crust) than on 14 March 2018 (crusted) and 17 July 2018 (crusted) ( Table 2), while the difference in the average infiltration rate between the crusted sites was small, exhibiting an insignificant decrease from 14 March 2018 to 17 July 2018 (p = 0.794) ( Table 2). However, there was a significant decrease from 14 March 2018 to 17 July 2018 in the scalped sites (p < 0.05) ( Table 2), and the results from both dates were greater in the scalped plots than in the crusted sites (p < 0.05) ( Table 2).

Temporal Evolution of the Infiltration Rate in the Recent Landslide
To study the temporal evolution and the influence of physical crusts on the infiltration rate in the recent landslide, successive infiltration experiments were conducted in the upper part and lower part of the NLB on 14 March 2018 and 17 July 2018 ( Table 2). The distance between the crusted and scalped experiment sites was 0.4-0.5 m. The results showed that the average IIR was dramatically greater on 26 October 2017 (no crust) than on 14 March 2018 (crusted) and 17 July 2018 (crusted) ( Table 2), while the difference in the average infiltration rate between the crusted sites was small, exhibiting an insignificant decrease from 14 March 2018 to 17 July 2018 (p = 0.794) ( Table 2). However, there was a significant decrease from 14 March 2018 to 17 July 2018 in the scalped sites (p < 0.05) ( Table 2), and the results from both dates were greater in the scalped plots than in the crusted sites (p < 0.05) ( Table 2). Note. Different superscript letters within a row or a column indicate significant differences exist between the infiltration rate of crusted and scalped (p < 0.05) (e.g., a and b indicate significant differences between the infiltration rate of crusted and scalped). Same superscript letter within a column indicate no significant differences between the infiltration rate on different infiltration experiments (p > 0.05).

The Reason for Spatial Differences in the Infiltration Rate
In this study, the infiltration experiment results showed that the average infiltration rate was dramatically highest in the NLB, intermediate in the PLB, and lowest in the UDL. As expected, the highest IIR occurred in the NLB, which can be attributed to the high degree of collapsibility of fragmented loess and the lack of obstruction by crusts on the soil surface [4,29,39]. The IIR value in the PLB was 4.4 times higher than that in the UDL, and we attributed this result to the loose structure of the disturbed loess and the weak cementation between soil particles, which is not easily restored to the level of UDL within a short time [37]. Furthermore, the recovery state and surface soil crusts can account for the spatial differences in the PLB. In the UDL, the infiltration rate was higher in the loess layers (L1 and L2) than in the paleosol layer (S1), which can be attributed to the loess having a lower bulk density and a higher porosity than the paleosol.

Role and Function of Physical Crusts in the Infiltration Rate in a Recent Landslide
The average IIR decreased sharply from 855.4 mm/min on 26 October 2017 (without crust), to 11.6 mm/min on 14 March 2018 (crusted) ( Table 2). We attributed this decline to the appearance of physical crusts. On the other hand, the infiltration rate was significantly lower on 17 July 2018 than on 14 March 2018 on the scalped plots (Table 2), which can be attributed to the compaction effect related to the quality of the soil. The ecosystem processes of infiltration, surface runoff and erosion are influenced by the surface conditions of soil [40,41]. The infiltration rate is likely to be higher in areas without the obstruction of surface soil crusts [13,42,43], especially for disturbed soils. Given the way in which crusts affect the distribution of surface water, the effects of the crusts on the infiltration rate and variability on a recent landslide are unknown.
First, to better study the temporal evolution of infiltration on recent landslides, it is necessary to understand the formation mechanism and duration of physical crusts. Previous studies have assumed that raindrop impacts are the main driver of physical crust formation [44,45] and that the rainfall intensity influences the physical crust thickness. Simulated rainfall experiments have shown that the crust thickness slowly increases with increasing rainfall intensity [45,46]. However, researchers have different opinions on the duration of physical crust formation. Eldridge believed that this process requires one year in situations of suitable rainfall [29], while Bresson believed that a physical crust is likely to form after one or two rainfall and drying events [22]. In Han's study, the crusts formed rapidly under laboratory conditions (within one hour) [44]. Regardless, these researchers all agree that a physical crust can dramatically decrease the infiltration rate. In this study, data from successive investigations conducted on the NLB (Figure 7) suggested that physical crust formation requires four to five months and that the crust thickness increased slowly (Table 3). Data from the Baqiao meteorological station showed that one third of the days in January 2018 were snowy or rainy, with a mean daily temperature of −1 • C. Then, the temperature gradually increased from 12 February 2018. Therefore, because of the effects of heavy snowfall, rainfall and temperature increases, distinct physical crusts were found on 14 March 2018 (Figure 7b), and the crust thickness increased slowly from 0.5-1.5 mm (14 March 2018) to 1-2 mm (17 July 2018) ( Table 3). Based on the effect of physical crusts on the infiltration rate, the results showed that the physical crusts were responsible for the initial rapid decrease in the infiltration rate; however, over longer time scales, the infiltration rate slowly decreased. February 2018. Therefore, because of the effects of heavy snowfall, rainfall and temperature increases, distinct physical crusts were found on 14 March 2018 (Figure 7b), and the crust thickness increased slowly from 0.5-1.5 mm (14 March 2018) to 1-2 mm (17 July 2018) ( Table 3). Based on the effect of physical crusts on the infiltration rate, the results showed that the physical crusts were responsible for the initial rapid decrease in the infiltration rate; however, over longer time scales, the infiltration rate slowly decreased.

Role and Function of Biological Crusts and Moss Crusts on a Landslide Body
Physical crusts, biological crusts, and moss crusts are the principal types of soil crusts. However, the duration of biological crust formation is much longer than that of physical crusts. In fact, the recovery of biological crusts requires 5 to 20 years [47], and other studies have suggested that biological crusts can form within 6 years on bare soil after shrub death [48]. Higher rainfall amounts will promote moss and lichen growth after the soil is stabilized by cyanobacteria [47]. Therefore, this slow growth process is the reason that no biological or moss crusts were found on the NLB in the recent investigation (17 July 2018) (Figure 7d). In contrast, physical crusts, biological crusts, and moss crusts were observed on the PLB, and the thicknesses of the crusts were 0.5-2 cm, 0.5-1 cm, and 2-3 cm, respectively. It is speculated that biological crusts can develop on disturbed soil within 5 to 7 years following landslide formation. Furthermore, there has long been a debate regarding whether biological crust formation decreases or increases the infiltration rate. The research of Warren [49] suggested that most biological crusts have a positive effect on infiltration according to 13 studies. This result may account for the higher infiltration rate in the PLB compared to that in the UDL.

Role and Function of Biological Crusts and Moss Crusts on a Landslide Body
Physical crusts, biological crusts, and moss crusts are the principal types of soil crusts. However, the duration of biological crust formation is much longer than that of physical crusts. In fact, the recovery of biological crusts requires 5 to 20 years [47], and other studies have suggested that biological crusts can form within 6 years on bare soil after shrub death [48]. Higher rainfall amounts will promote moss and lichen growth after the soil is stabilized by cyanobacteria [47]. Therefore, this slow growth process is the reason that no biological or moss crusts were found on the NLB in the recent investigation (17 July 2018) (Figure 7d). In contrast, physical crusts, biological crusts, and moss crusts were observed on the PLB, and the thicknesses of the crusts were 0.5-2 cm, 0.5-1 cm, and 2-3 cm, respectively. It is speculated that biological crusts can develop on disturbed soil within 5 to 7 years following landslide formation. Furthermore, there has long been a debate regarding whether biological crust formation decreases or increases the infiltration rate. The research of Warren [49] suggested that most biological crusts have a positive effect on infiltration according to 13 studies. This result may account for the higher infiltration rate in the PLB compared to that in the UDL.

The Temporal Evolution of the Infiltration Rate in a Recent Landslide
Loess failure can change the microstructure of the soil, further influencing its macroscopic geotechnical properties. Therefore, the interparticle cementation and shear strength of disturbed soil are lower than those of undisturbed soil [37,50]. In fact, these variations have a remarkable impact on infiltration. Therefore, loess landslides can cause the infiltration rate to significantly increase during the one to three months after landslide formation because of the changes in the loess structure [51,52]. Due to the influences of the recovery state, rainfall, temperature, and microbial activities, moss crusts and biological crusts begin to form and gradually thicken within 5 to 20 years [28,29,53], causing the infiltration rate in recent landslides to decrease slowly [54]. According to the results of this study, more than 10 years may be required for the disturbed soil caused by a landslide to recover the same infiltration rate as UDL.

Conclusions
Our experiments revealed that the loess landslide process resulted in greatly increased infiltration, especially in a recent landslide. The average IIR in the NLB was approximately 200 times and 60 times higher than that in the UDL and PLB, respectively. Moreover, the infiltration capacity in the loess layer was slightly higher than that in the paleosol layer. Spatially, the infiltration rate in the PLB was highest in the upper part, intermediate in the middle part, and lowest in the lower part. Successive investigation and experiments conducted on the NLB revealed that the formation of raindrop-induced physical crusts caused the infiltration rate to rapidly decrease for four to five months, after which it slowly decreased. The results may decrease the risk of damage and offer suggestions when a landslide body needs to be used for industrial applications or as farmland.