Landslide Characteristics and Evolution: What We Can Learn from Three Adjacent Landslides

: Landslide processes are a consequence of the interactions between their triggers and the surrounding environment. Understanding the differences in landslide movement processes and characteristics can provide new insights for landslide prevention and mitigation. Three adjacent landslides characterized by different movement processes were triggered from August to September in 2018 in Hualong County, China. A combination of surface and subsurface characteristics illustrated that Xiongwa (XW) landslides 1 and 2 have deformed several times and exhibit signiﬁcant heterogeneity, whereas the Xiashitang (XST) landslide is a typical retrogressive landslide, and its material has moved downslope along a shear surface. Time-series Interferometric Synthetic Aperture Radar (InSAR) and Differential InSAR (DInSAR) techniques were used to detect the displacement processes of these three landslides. The pre-failure displacement signals of a slow-moving landslide (the XST landslide) can be clearly revealed by using time-series InSAR. However, these sudden landslides, which are a typical catastrophic natural hazard across the globe, are easily ignored by time-series InSAR. We conﬁrmed that effective antecedent precipitation played an important role in the three landslides’ occurrence. The deformation of an existing landslide itself can also trigger new adjacent landslides in this study. These ﬁndings indicate that landslide early warnings are still a challenge since landslide processes and mechanisms are complicated. We need to learn to live with natural disasters, and more relevant detection and ﬁeld investigations should be conducted for landslide risk mitigation.


Introduction
Landslide processes and evolution are influenced by the interactions between triggering events and local natural conditions, including hillslope topography and geological background [1][2][3][4]. These processes can reflect the differences in landslides' deformation patterns and the spatial redistribution characteristics of their materials [5][6][7][8][9][10], which is helpful in understanding the mechanisms of slope instability. Landslides processes vary even for adjacent hillslopes with similar geological and topographical conditions due to the complexity of the relationship between their internal structure and triggers [11][12][13][14]. Thus, it is important to provide new insights for hazard mitigation based on landslides characterized by different movement processes.
The deformation signal of landslide is a key parameter for early landslide warning systems. It reflects the deformation degree of the hillslope and determines how the local sustained uplift and regional tectonic movement ( Figure 1) [38,39]. The annual precipitation is 451 mm (China Meteorological Data Service Center, http://data.cma.cn/ (accessed on 2 September 2021)). A number of landslides have been triggered in this region, and they always cause casualties and severe damage to infrastructure. From 2 July 2018 to 4 October 2018, three landslides occurred successively in Jinyuan Tibetan township, Hualong County, Qinghai Province, China (Figure 1a). Xiongwa (XW) landslides 1 and 2 ( Figure 1c) occurred on 2 September 2018. These two landslides tore apart about 3.8 km of road S306. It should be noted that XW landslide 2 occurred on the hillslope where a landslide had previously occurred on 8 September 2007. The other landslide was the Xiashitang (XST) landslide (Figure 1c), which occurred from 6 August 2018 to 10 September 2018, according to Sentinel-2 images. The houses and farmland located in the toe area of the XST landslide continue to be at risk since the landslide body is still unstable.
Remote Sens. 2021, 13, x FOR PEER REVIEW 3 of 20 Hualong County is located in Qinghai Province, where the Tibetan Plateau and the Loess Plateau connect. This region is characterized by high relief and steep slopes due to sustained uplift and regional tectonic movement ( Figure 1) [38,39]. The annual precipitation is 451 mm (China Meteorological Data Service Center, http://data.cma.cn/ (accessed on 2 September 2021)). A number of landslides have been triggered in this region, and they always cause casualties and severe damage to infrastructure. From 2 July 2018 to 4 October 2018, three landslides occurred successively in Jinyuan Tibetan township, Hualong County, Qinghai Province, China (Figure 1a). Xiongwa (XW) landslides 1 and 2 ( Figure 1c) occurred on 2 September 2018. These two landslides tore apart about 3.8 km of road S306. It should be noted that XW landslide 2 occurred on the hillslope where a landslide had previously occurred on 8 September 2007. The other landslide was the Xiashitang (XST) landslide (Figure 1c), which occurred from 6 August 2018 to 10 September 2018, according to Sentinel-2 images. The houses and farmland located in the toe area of the XST landslide continue to be at risk since the landslide body is still unstable. The lithosomes of the three landslides are mainly covered by orange glutenite (interbedding hard and soft clastic rocks), which are categorized as terrigenous clastic sediments ( Figure 2a). There is a weathered soil layer covered on top of the substratum with a thickness around 1-2 m, which is the weathering product of orange glutenite. Field investigation revealed that the landslide bodies mainly consist of unweathered substratum The lithosomes of the three landslides are mainly covered by orange glutenite (interbedding hard and soft clastic rocks), which are categorized as terrigenous clastic sediments ( Figure 2a). There is a weathered soil layer covered on top of the substratum with a thickness around 1-2 m, which is the weathering product of orange glutenite. Field investigation revealed that the landslide bodies mainly consist of unweathered substratum (glutenite) and weathered soil (Figure 2b,c). This orange glutenite is characterized by poor physical and mechanical properties and poor weathering resistance. Moreover, the well-developed fissures result in rocks having low strength; thus, the rocks are prone to sliding due to the formation of weak geological structures [40,41]. Thus, the hillslope is prone to sliding under the influence of rainfall. (glutenite) and weathered soil (Figure 2b,c). This orange glutenite is characterized by poor physical and mechanical properties and poor weathering resistance. Moreover, the welldeveloped fissures result in rocks having low strength; thus, the rocks are prone to sliding due to the formation of weak geological structures [40,41]. Thus, the hillslope is prone to sliding under the influence of rainfall.

Measurement of Landslide Surface Movement Distance and Direction
The aerial images of pre-failure and post-failure moments allow us to track the movement of the surface features. Based on the aerial images obtained on 2 July 2018 (SPOT7) and 4 October 2018 (SPOT6), with a resolution of 1.5 m, the locations of the same surface features (e.g., roads) were first identified in different aerial images. Then, the same surface features were connected with lines. The lengths of the lines were measured to quantify movement distance, and the angles between the lines and north were confirmed as the movement direction of the surface features using ArcGIS 10.2 (ArcGIS by Esri). In this manner, a vector displacement field was mapped.

Time-Series InSAR and DInSAR Analyses
A total of 31 Sentinel-1 ascending orbit interferometric images acquired from 4 September 2017 to 30 August 2018, with a spatial resolution of 20.0 × 5.0 m (Azimuth × Range, A × R), were obtained from the European Space Agency (ESA). The image obtained on 10 October 2017 was chosen as the super master ( Figure 3). A Shuttle Radar Topography Mission (SRTM) Digital Elevation Model (DEM) (30 m) with a vertical accuracy of ±10 m was used as the topographic reference (https://earthexplorer.usgs.gov/ (accessed on 11 June 2021)). The small-baseline subset (SBAS) technique was used to obtain cumulative displacements of the three landslides in this study. Forty-five days was taken as the temporal baseline, and 200 m was taken as the spatial baseline. The orbit datasets generated 87 interferometric pairs. We also conducted DInSAR analysis based on two pre-failure interferometric images (acquired on 18 August and 30 August 2018) and one post-failure image (acquired on 11 September 2018) in order to detect sudden movement of the landslides ( Figure 4).

Measurement of Landslide Surface Movement Distance and Direction
The aerial images of pre-failure and post-failure moments allow us to track the movement of the surface features. Based on the aerial images obtained on 2 July 2018 (SPOT7) and 4 October 2018 (SPOT6), with a resolution of 1.5 m, the locations of the same surface features (e.g., roads) were first identified in different aerial images. Then, the same surface features were connected with lines. The lengths of the lines were measured to quantify movement distance, and the angles between the lines and north were confirmed as the movement direction of the surface features using ArcGIS 10.2 (ArcGIS by Esri). In this manner, a vector displacement field was mapped.

Time-Series InSAR and DInSAR Analyses
A total of 31 Sentinel-1 ascending orbit interferometric images acquired from 4 September 2017 to 30 August 2018, with a spatial resolution of 20.0 × 5.0 m (Azimuth × Range, A × R), were obtained from the European Space Agency (ESA). The image obtained on 10 October 2017 was chosen as the super master ( Figure 3). A Shuttle Radar Topography Mission (SRTM) Digital Elevation Model (DEM) (30 m) with a vertical accuracy of ±10 m was used as the topographic reference (https://earthexplorer.usgs.gov/ (accessed on 11 June 2021)). The small-baseline subset (SBAS) technique was used to obtain cumulative displacements of the three landslides in this study. Forty-five days was taken as the temporal baseline, and 200 m was taken as the spatial baseline. The orbit datasets generated 87 interferometric pairs. We also conducted DInSAR analysis based on two pre-failure interferometric images (acquired on 18 August and 30 August 2018) and one post-failure image (acquired on 11 September 2018) in order to detect sudden movement of the landslides (Figure 4).

UAV Surveys
We carried out UAV photogrammetry on 21 December 2020 and 28 June 2021 ( Table  1). The UAV used was a DJI PHANTOM 4 RTK. We designated the flight heights as 300 m on the first survey and 200 m on the second survey. The flight speed was 7.9 m/s for both surveys. We set 6 and 21 ground control points (GCPs) to ensure the accuracy of the orthophotos and the DEMs in the UAV surveys, respectively. The coordinates of each GCP were measured by using a Global Navigation Satellite System Real-Time Kinematic (GNSS RTK) (CHCNAV, i50, China). The orthophoto and DEM were generated by using the Pix4Dmapper software (Pix4Dmapper, 2017). The coordinate system used was the WGS_3_Degree_GK_CM_102E. In this manner, two high-resolution DEMs and orthophotos were generated.

UAV Surveys
We carried out UAV photogrammetry on 21 December 2020 and 28 June 2021 ( Table  1). The UAV used was a DJI PHANTOM 4 RTK. We designated the flight heights as 300 m on the first survey and 200 m on the second survey. The flight speed was 7.9 m/s for both surveys. We set 6 and 21 ground control points (GCPs) to ensure the accuracy of the orthophotos and the DEMs in the UAV surveys, respectively. The coordinates of each GCP were measured by using a Global Navigation Satellite System Real-Time Kinematic (GNSS RTK) (CHCNAV, i50, China). The orthophoto and DEM were generated by using the Pix4Dmapper software (Pix4Dmapper, 2017). The coordinate system used was the WGS_3_Degree_GK_CM_102E. In this manner, two high-resolution DEMs and orthophotos were generated.

UAV Surveys
We carried out UAV photogrammetry on 21 December 2020 and 28 June 2021 ( Table 1). The UAV used was a DJI PHANTOM 4 RTK. We designated the flight heights as 300 m on the first survey and 200 m on the second survey. The flight speed was 7.9 m/s for both surveys. We set 6 and 21 ground control points (GCPs) to ensure the accuracy of the orthophotos and the DEMs in the UAV surveys, respectively. The coordinates of each GCP were measured by using a Global Navigation Satellite System Real-Time Kinematic (GNSS RTK) (CHCNAV, i50, China). The orthophoto and DEM were generated by using the Pix4Dmapper software (Pix4Dmapper, 2017). The coordinate system used was the WGS_3_Degree_GK_CM_102E. In this manner, two high-resolution DEMs and orthophotos were generated.

Electrical Resistivity Tomography Detection of the Internal Structures of the Landslides
Most landslides have large heterogeneities due to variations in materials and local slope topography. Electrical resistivity tomography can reveal water content distributions at different depths along a profile, which constitutes key evidence for investigating the internal structure of landslide deposits [36]. In this study, we used electrical resistivity tomography to detect the internal structures of the landslides along three profiles in XW landslides 1 and 2 and three profiles in the XST landslide. The ERT instrument used was a GeoERT IP 2401 (Beijing, China). Electrode spacing was set at 5 m, and the effective measuring depth reached about 75 m along profiles AA', BB', and CC' and about 75 m, 55 m, and 55 m along profiles DD', EE', and FF', respectively (Figure 1c). The coordinates of each electrode were measured by using GNSS RTK.

Effective Antecedent Precipitation
Precipitation is regarded as a key trigger for landslides in most mountainous and hilly areas. It should be noted that hillslope stability is not only influenced by the precipitation on the day of the event but also by the precipitation within a certain period preceding a landslide [42]. This is referred as to the effective antecedent precipitation, and its influence on the soil water content will gradually decay due to evaporation. Thus, we followed the method of Ma et al. [42] and Guo et al. [43] and used the effective antecedent precipitation (EAP) to evaluate the contribution of precipitation to hillslope instability: where EAP is the effective antecedent precipitation, P i is the daily precipitation on the i th day before the landslide, n is the number of days before the landslide (1 ≤ i ≤ n), and K is the decay coefficient due to evaporation [42]. Here, following Ma et al. [42], K was set to 0.84.

Evolution of XW Landslides 1 and 2 and Its Influence
As is shown in Figure 5, road S306 was mainly destroyed by the landslides on 8 September 2007 and 2 September 2018. Historical aerial images revealed that the earlier landslide, which occurred on 8 September 2007, was located in the middle of the hillslope, and it destroyed about 2.7 km of road S306 ( Figure 5b). As shown in Figure 5a, the main sliding direction of XW landslides 1 and 2 was east. Due to the fact that road S306 is the key road from Bayan Town to Guanting Town, the government had to repair the damaged road in order to ensure that traffic could pass through, but the risk of future landslides to inhabitants and their houses and farmland remained. On 2 September 2018, about 3.8 km of road S306 was destroyed again by XW landslides 1 and 2 (Figure 5d), which disrupted traffic for 10 days. Then, road S306 was repaired again after the second landslide. It is likely that this portion of road S306 will be destroyed again as the landslide region remains unstable.
Road construction is difficult and expensive, especially in underdeveloped mountainous areas. Thus, even though the road was damaged by landslides several times, the government had to repair the road multiple times over a short period. Short-term evacuation, but not relocation, is feasible for most villages because their property, such as their houses and farmland, cannot be moved. As a result, we have to learn how to live with the risk of landslides. Fortunately, with the development of monitoring techniques, more early warning systems are being used to provide key information for hazard warning announcements.  Figure 6a shows that only a few points can be treated as reliable ground features and were used to identify the movement direction and distance of XW landslides 1 and 2. The main sliding direction of XW landslide 1 was east, with a mean slip distance of 93.5 m. Moreover, the local topographic features helped us to estimate the slip directions in regions 1 and 2 (black dashed circles 1 and 2 in Figure 6a). Part of the deposits moved downward into the gully, with a S-SE direction in region 1 and a N-NE direction in region 2. The ground features in the middle of the hillslope were completely destroyed, so the displacement in this region could not be identified. The results of the vector fields in XW landslide 2 indicate that the displacement in the right part (with a mean slip distance of 26.0 m) was less than that in the left part (with a mean slip distance of 81.5 m).

Landslide Surface Movement Vector Field
The vector displacement field shown in Figure 6b was formed based on 18 pairs of ground features, which were identified in pre-landslide and post-landslide aerial images. The vector fields of the XST landslide were completely different from those of XW landslides 1 and 2. With the exception of region 1 (black dashed circle 1 in Figure 6b), which lacked the reliable ground features needed to confirm the direction and distance of the movement, clear vector fields were obtained for the other parts of the XST landslide. Their displacements ranged from 16.0 to 46.1 m, with a mean distance of 30.9 m. Displacement gradually decreased from the NW to the SE. All of the directions of the vectors showed that the main slip direction of the XST landslide was east. The vector displacement fields demonstrated that the XST landslide was a typical retrogressive landslide; that is, the movement distance in its upper part was greater than that in its lower part. Most of the topography and geomorphology in this region differ from those of a natural hillslope due to the construction of terraced fields. These changes in topography may be the key factors that caused the old landslide and the XST landslide. The entire landslide body, except for the upper part, moved downslope, but the relative spatial positions of the surface features shows that only a few points can be treated as reliable ground features and were used to identify the movement direction and distance of XW landslides 1 and 2. The main sliding direction of XW landslide 1 was east, with a mean slip distance of 93.5 m. Moreover, the local topographic features helped us to estimate the slip directions in regions 1 and 2 (black dashed circles 1 and 2 in Figure 6a). Part of the deposits moved downward into the gully, with a S-SE direction in region 1 and a N-NE direction in region 2. The ground features in the middle of the hillslope were completely destroyed, so the displacement in this region could not be identified. The results of the vector fields in XW landslide 2 indicate that the displacement in the right part (with a mean slip distance of 26.0 m) was less than that in the left part (with a mean slip distance of 81.5 m).
The vector displacement field shown in Figure 6b was formed based on 18 pairs of ground features, which were identified in pre-landslide and post-landslide aerial images. The vector fields of the XST landslide were completely different from those of XW landslides 1 and 2. With the exception of region 1 (black dashed circle 1 in Figure 6b), which lacked the reliable ground features needed to confirm the direction and distance of the movement, clear vector fields were obtained for the other parts of the XST landslide. Their displacements ranged from 16.0 to 46.1 m, with a mean distance of 30.9 m. Displacement gradually decreased from the NW to the SE. All of the directions of the vectors showed that the main slip direction of the XST landslide was east. The vector displacement fields demonstrated that the XST landslide was a typical retrogressive landslide; that is, the movement distance in its upper part was greater than that in its lower part. Most of the topography and geomorphology in this region differ from those of a natural hillslope due to the construction of terraced fields. These changes in topography may be the key factors that caused the old landslide and the XST landslide. The entire landslide body, except for the upper part, moved downslope, but the relative spatial positions of the surface features only changed slightly, which is key evidence confirming the movement process of the XST landslide.  Figure 7 shows the cumulative displacements of the three landslides. We found that only the area along road S306 exhibited a small displacement and that most of the regions did not exhibit significant displacement for XW landslide 1. In particular, the main scarp region of XW landslide 1 was almost stable. Thus, the XW landslide was a sudden landslide, and its short deformation period was less than the Sentinel-1 satellite revisit period (minimum of 6 days). The cumulative displacement of XW landslide 2, except for point L6, was less than 11 mm, illustrating that XW landslide 2 was also still stable, except for the toe region (point P6) (Figure 7b). However, the cumulative deformation of the XST landslide gradually increased from 4 September 2017 to 30 August 2018.  (Figure 7j,k). The timeseries InSAR analysis revealed that the displacement at the top of the XST landslide was negligible since this part did not move downward, but it was cracked by the movement of the main body. This is also key evidence for illustrating that the XST landslide was a retrogressive landslide. Figure 8c shows that the displacement at point R2 in the XST landslide was larger than those at points R3 and R4, which were larger than that at point R1. Moreover, the upper part of the XST landslide failed first (Point R2 in Figure 8a). This is consistent with the results of the surface movement vector field based on pre-failure and post-failure aerial images. As cumulative displacement increased, the lower part of the landslide moved downward due to the extrusion of material in the upper part. Thus, the XST landslide was a typical retrogressive landslide.  Figure 7 shows the cumulative displacements of the three landslides. We found that only the area along road S306 exhibited a small displacement and that most of the regions did not exhibit significant displacement for XW landslide 1. In particular, the main scarp region of XW landslide 1 was almost stable. Thus, the XW landslide was a sudden landslide, and its short deformation period was less than the Sentinel-1 satellite revisit period (minimum of 6 days). The cumulative displacement of XW landslide 2, except for point L6, was less than 11 mm, illustrating that XW landslide 2 was also still stable, except for the toe region (point P6) (Figure 7b). However, the cumulative deformation of the XST landslide gradually increased from 4 September 2017 to 30 August 2018. Figure 7c-e illustrates that the instability occurred in the upper part of the XST landslide first and then the deformation area gradually increased (Figure 7f-i). From 20 June 2018 to 30 August 2018, except for the upper part of the XST landslide, the main landslide body exhibited clear displacement (Figure 7j,k). The timeseries InSAR analysis revealed that the displacement at the top of the XST landslide was negligible since this part did not move downward, but it was cracked by the movement of the main body. This is also key evidence for illustrating that the XST landslide was a retrogressive landslide.    Figure 8c shows that the displacement at point R2 in the XST landslide was larger than those at points R3 and R4, which were larger than that at point R1. Moreover, the upper part of the XST landslide failed first (Point R2 in Figure 8a). This is consistent with the results of the surface movement vector field based on pre-failure and post-failure aerial images. As cumulative displacement increased, the lower part of the landslide moved downward due to the extrusion of material in the upper part. Thus, the XST landslide was a typical retrogressive landslide. Remote Sens. 2021, 13, x FOR PEER REVIEW 10 of 20

Large Displacement Detection Revealed via DInSAR
An obvious phase jump was identified based on the interferograms of two adjacent periods. It corresponds to severe low coherence, which means that a large amount of displacement occurred in a short period of time. It general, this can be treated as the signal of the landslide's occurrence and can be used to identify where the larger deformation occurred [19]. Figures 9a and b show that most of the parts of XW landslides 1 and 2 did not exhibit significant interferometric fringes and were characterized by a higher correlation. Thus, XW landslides 1 and 2 remained stable until 30 August 2018. However, the significant interferometric fringes (Figure 9c) and low correlation (Figure 9d) indicate that a larger displacement occurred in XW landslides 1 and 2 and part of the XST landslide from 30 August to 11 September 2018. This DInSAR analysis provides key evidence of the occurrences of larger displacements, which means that XW landslides 1 and 2 occurred without significant early deformation until 30 August 2018.

Large Displacement Detection Revealed via DInSAR
An obvious phase jump was identified based on the interferograms of two adjacent periods. It corresponds to severe low coherence, which means that a large amount of displacement occurred in a short period of time. It general, this can be treated as the signal of the landslide's occurrence and can be used to identify where the larger deformation occurred [19]. Figure 9a,b show that most of the parts of XW landslides 1 and 2 did not exhibit significant interferometric fringes and were characterized by a higher correlation. Thus, XW landslides 1 and 2 remained stable until 30 August 2018. However, the significant interferometric fringes (Figure 9c) and low correlation (Figure 9d) indicate that a larger displacement occurred in XW landslides 1 and 2 and part of the XST landslide from 30 August to 11 September 2018. This DInSAR analysis provides key evidence of the occurrences of larger displacements, which means that XW landslides 1 and 2 occurred without significant early deformation until 30 August 2018.
Most landslides around the world are preceded by deformations [15,24]. Deformation detection of landslides that remain stable until several days or even one day before their failure is undoubtedly difficult when using time-series InSAR analysis because Sentinel-1 images have a relatively long repeating period (minimum of 6 days). For example, for Shuicheng landslide-which occurred on 23 July 2019, killed 42 people, and destroyed 21 houses in Guizhou, China-even in the latest image, which was acquired on 20 July 2019 (three days before the landslide), the surface deformation process was not detected via multitemporal InSAR [22]. Thus, the InSAR technique can track the movement processes of some potential landslides, but it ignores some unstable slopes that may transform into catastrophic landslides. We have to learn how to live with this risk because natural hazards are unavoidable and numerous landslides still cannot be detected [44]. The early warning of landslides is still a challenge even with the development of the Earth observation system, InSAR, UAV surveys, and in situ monitoring sensors. Determining where and when landslides will occur still requires further study [15]. Remote Sens. 2021, 13, x FOR PEER REVIEW 11 of 20 Most landslides around the world are preceded by deformations [15,24]. Deformation detection of landslides that remain stable until several days or even one day before their failure is undoubtedly difficult when using time-series InSAR analysis because Sentinel-1 images have a relatively long repeating period (minimum of 6 days). For example, for Shuicheng landslide-which occurred on 23 July 2019, killed 42 people, and destroyed 21 houses in Guizhou, China-even in the latest image, which was acquired on 20 July 2019 (three days before the landslide), the surface deformation process was not detected via multitemporal InSAR [22]. Thus, the InSAR technique can track the movement processes of some potential landslides, but it ignores some unstable slopes that may transform into catastrophic landslides. We have to learn how to live with this risk because natural hazards are unavoidable and numerous landslides still cannot be detected [44]. The early warning of landslides is still a challenge even with the development of the Earth observation system, InSAR, UAV surveys, and in situ monitoring sensors. Determining where and when landslides will occur still requires further study [15].

Surface Characteristics Based on UAV and Field Investigations
We conducted field investigations and UAV surveys on 21 December 2020 and 28 June 2021. Figure 10a-d shows that XW landslides 1 and 2 tore apart the hillslope surface. Most of the damage to the road and hillslope surface was X-shaped. A witness reported that XW landslide 1 began as collapses in the main scarp area (Figure 10a) on the morning of 2 September 2018, and the entire hillslope moved slowly downward until the afternoon

Surface Characteristics Based on UAV and Field Investigations
We conducted field investigations and UAV surveys on 21 December 2020 and 28 June 2021. Figure 10a-d shows that XW landslides 1 and 2 tore apart the hillslope surface. Most of the damage to the road and hillslope surface was X-shaped. A witness reported that XW landslide 1 began as collapses in the main scarp area (Figure 10a) on the morning of 2 September 2018, and the entire hillslope moved slowly downward until the afternoon [45]. The deformation and movement processes of XW landslides 1 and 2 were different from those of the landslides characterized by long-runout distances and great mobility. The unstable material of latter landslides is usually semi-fluid, and it quickly slides along the surface. However, the main body of XW landslides 1 and 2 did not exhibit obvious liquefaction and, thus, moved slowly, which caused most of the parts of the landslide body to deform several times due to interactions between the material and local topography. The entire unstable hillslope length (including XW landslides 1 and 2) was more than 1200 m. Figure 10b,c shows that the damaged road exhibited a different shape, which indicates that the stress conditions varied in different parts of XW landslides 1 and 2.
the surface. However, the main body of XW landslides 1 and 2 did not exhibit obvious liquefaction and, thus, moved slowly, which caused most of the parts of the landslide body to deform several times due to interactions between the material and local topography. The entire unstable hillslope length (including XW landslides 1 and 2) was more than 1200 m. Figure 10b,c shows that the damaged road exhibited a different shape, which indicates that the stress conditions varied in different parts of XW landslides 1 and 2. The XST landslide was a reactivated landslide, which was located in an old landslide body (Figure 1c). Figure 10e shows that the head of the XST landslide was cracked and formed several vertical offsets. In contrast, the foot of the XST landslide was not significantly deformed since part of the deposition was transported into the river along the gully Two dammed lakes formed in the toe the XST landslide, which may become sources of debris flow (Figure 10i). Thus, the houses and farmland located in the toe are still threatened.
High resolution UAV-based DEMs and orthophotos have become powerful methods in landslide surveying and mapping [46,47]. Here, by using a UAV-based DEM obtained on 28 June 2021, which was resampled to a 2.5 m resolution in ArcGIS 10.2, in addition to an Advanced Land Observing Satellite World 3-D Digital Surface Model (AW3D DSM) [48] obtained from 2006 to 2011 with a resolution of 2.5 m, we examined elevation differences during pre-failure and post-failure periods by using the Raster Calculator tool in ArcGIS 10.2. Figures 11a and 11e show that the upper and middle parts of the hillslope were actually destroyed by two landslides, i.e., XW landslides 1 and 2, respectively. Moreover, the slope ( Figure 11c) and hillshade (Figure 11d) clearly show the main scarp of XW landslide 2, which was difficult to identify in the orthophotos (Figure 11b). Although part The XST landslide was a reactivated landslide, which was located in an old landslide body (Figure 1c). Figure 10e shows that the head of the XST landslide was cracked and formed several vertical offsets. In contrast, the foot of the XST landslide was not significantly deformed since part of the deposition was transported into the river along the gully. Two dammed lakes formed in the toe the XST landslide, which may become sources of debris flow (Figure 10i). Thus, the houses and farmland located in the toe are still threatened.
High resolution UAV-based DEMs and orthophotos have become powerful methods in landslide surveying and mapping [46,47]. Here, by using a UAV-based DEM obtained on 28 June 2021, which was resampled to a 2.5 m resolution in ArcGIS 10.2, in addition to an Advanced Land Observing Satellite World 3-D Digital Surface Model (AW3D DSM) [48] obtained from 2006 to 2011 with a resolution of 2.5 m, we examined elevation differences during pre-failure and post-failure periods by using the Raster Calculator tool in ArcGIS 10.2. Figure 11a,e show that the upper and middle parts of the hillslope were actually destroyed by two landslides, i.e., XW landslides 1 and 2, respectively. Moreover, the slope ( Figure 11c) and hillshade (Figure 11d) clearly show the main scarp of XW landslide 2, which was difficult to identify in the orthophotos (Figure 11b). Although part of the landslide body has been artificial disturbed, the pre-failure and post-failure characteristics of profile A-D still show the erosion and deposition regions of each landslide.

Electrical Resistivity Tomography Detection of the Internal Structures of Landslides
We used electrical resistivity tomography to investigate the influence of landslides with different movement processes on internal structures along six profiles (Figure 1c). Figure 10 shows the distribution of the electrical resistivity in the XST landslide along profile DD' after four iteration calculations. The significant electrical resistivity changes are interpreted as a possible shear surface. The depth of the shear surface varies from 20 to 40 m along the profile. Most of the region above the shear surface is characterized by lower electrical resistivities (<10 Ωm). The region below the shear surface has a higher electrical resistivity than compared to the surrounding area (>60 Ωm) and could be bedrock or compacted soil. In the upper part of the XST landslide (upper 200 m) (red irregular region 1 in Figure 12a), surface cracks promote water infiltration and result in this region having high-water content.

Electrical Resistivity Tomography Detection of the Internal Structures of Landslides
We used electrical resistivity tomography to investigate the influence of landslides with different movement processes on internal structures along six profiles (Figure 1c). Figure 10 shows the distribution of the electrical resistivity in the XST landslide along profile DD' after four iteration calculations. The significant electrical resistivity changes are interpreted as a possible shear surface. The depth of the shear surface varies from 20 to 40 m along the profile. Most of the region above the shear surface is characterized by lower electrical resistivities (<10 Ωm). The region below the shear surface has a higher electrical resistivity than compared to the surrounding area (>60 Ωm) and could be bedrock or compacted soil. In the upper part of the XST landslide (upper 200 m) (red irregular region 1 in Figure 12a), surface cracks promote water infiltration and result in this region having high-water content.  Figure 13 shows the distributions of electrical resistivity in the XST landslide along profiles EE' and FF' after four and five iteration calculations, respectively. Profile EE' is nearly perpendicular to the sliding direction and is longer than the landslide's boundary (Figure 13a). We found that both landslide flank regions (black and blue dashed circles in Figure 13b) are characterized by lower electrical resistivity values than the surrounding areas, which indicates that where soil structure has been disturbed, the water concentration is higher. According to the differences in the electrical resistivity, we speculate that the shear surface along profile EE' varied from 20 to 25 m in depth. The landslide body has lower electrical resistivities than the bedrock region. The same pattern was found along profile FF' (Figure 13d), but the shear surface along profile FF' varied from 15 to 45 m in depth. This indicates that the depth of the XST landslide varied in different parts of the landslide, which may have been caused by local topography and soil properties.  Figure 13 shows the distributions of electrical resistivity in the XST landslide along profiles EE' and FF' after four and five iteration calculations, respectively. Profile EE' is nearly perpendicular to the sliding direction and is longer than the landslide's boundary (Figure 13a). We found that both landslide flank regions (black and blue dashed circles in Figure 13b) are characterized by lower electrical resistivity values than the surrounding areas, which indicates that where soil structure has been disturbed, the water concentration is higher. According to the differences in the electrical resistivity, we speculate that the shear surface along profile EE' varied from 20 to 25 m in depth. The landslide body has lower electrical resistivities than the bedrock region. The same pattern was found along profile FF' (Figure 13d), but the shear surface along profile FF' varied from 15 to 45 m in depth. This indicates that the depth of the XST landslide varied in different parts of the landslide, which may have been caused by local topography and soil properties.  Figure 13 shows the distributions of electrical resistivity in the XST landslide along profiles EE' and FF' after four and five iteration calculations, respectively. Profile EE' is nearly perpendicular to the sliding direction and is longer than the landslide's boundary (Figure 13a). We found that both landslide flank regions (black and blue dashed circles in Figure 13b) are characterized by lower electrical resistivity values than the surrounding areas, which indicates that where soil structure has been disturbed, the water concentration is higher. According to the differences in the electrical resistivity, we speculate that the shear surface along profile EE' varied from 20 to 25 m in depth. The landslide body has lower electrical resistivities than the bedrock region. The same pattern was found along profile FF' (Figure 13d), but the shear surface along profile FF' varied from 15 to 45 m in depth. This indicates that the depth of the XST landslide varied in different parts of the landslide, which may have been caused by local topography and soil properties.  Figure 14 shows the distribution of electrical resistivities in XW landslides 1 and 2, which are the result of five iteration calculations, along profiles AA', BB', and CC' to a maximum depth of about 70 m. Shear surface recognition of XW landslides 1 and 2 from electrical resistivity tomography was difficult because the landslides are characterized by significant heterogeneities along the profiles and with depth, which makes them difficult to identify in 2D images. The post-failure road repair also changed the surface structure (to a depth of about 5 m) and affected the electrical resistivity of the soil. However, the locations of the road and damaged road are characterized by a larger resistivity (Figure 14).  Figure 14 shows the distribution of electrical resistivities in XW landslides 1 and 2, which are the result of five iteration calculations, along profiles AA', BB', and CC' to a maximum depth of about 70 m. Shear surface recognition of XW landslides 1 and 2 from electrical resistivity tomography was difficult because the landslides are characterized by significant heterogeneities along the profiles and with depth, which makes them difficult to identify in 2D images. The post-failure road repair also changed the surface structure (to a depth of about 5 m) and affected the electrical resistivity of the soil. However, the locations of the road and damaged road are characterized by a larger resistivity ( Figure  14). Electrical resistivity tomography has been widely used to detect the internal structures of landslides based on 2D or 3D images of the spatial distribution of electrical resistivities [36]. ERT provides us with an opportunity to understand the landslide's internal structure without using boreholes. However, the internal structure was not speculated based on the exact value of the electrical resistivity but on its spatial distribution characteristics and even on the stratigraphic data from boreholes [34,36]. In our study, the electrical resistivity tomography profiles illustrate that XW landslides 1 and 2 were characterized by cracked surfaces and significant internal heterogeneities. This is the complex result of the influence of multiple deformations between deposition periods and local topography, as well as human activities. Compared with the XW landslides, the XST landslide may have a clear shear surface since it is a retrogressive landslide. Both the timeseries InSAR analysis and surface movement vector fields illustrate that the main body flowed down as a whole. Thus, the ERT technique may be more suitable for detecting the internal structure of landslides similar to XST.

Effective Antecedent Precipitation
Precipitation intensity is a common and easily accessible landslide warning signal. It is crucial for hazard warnings in most mountainous and hilly areas, especially in the under-developed regions where there is a lack of ground-based sensors obtaining Electrical resistivity tomography has been widely used to detect the internal structures of landslides based on 2D or 3D images of the spatial distribution of electrical resistivities [36]. ERT provides us with an opportunity to understand the landslide's internal structure without using boreholes. However, the internal structure was not speculated based on the exact value of the electrical resistivity but on its spatial distribution characteristics and even on the stratigraphic data from boreholes [34,36]. In our study, the electrical resistivity tomography profiles illustrate that XW landslides 1 and 2 were characterized by cracked surfaces and significant internal heterogeneities. This is the complex result of the influence of multiple deformations between deposition periods and local topography, as well as human activities. Compared with the XW landslides, the XST landslide may have a clear shear surface since it is a retrogressive landslide. Both the time-series InSAR analysis and surface movement vector fields illustrate that the main body flowed down as a whole. Thus, the ERT technique may be more suitable for detecting the internal structure of landslides similar to XST.

Effective Antecedent Precipitation
Precipitation intensity is a common and easily accessible landslide warning signal. It is crucial for hazard warnings in most mountainous and hilly areas, especially in the underdeveloped regions where there is a lack of ground-based sensors obtaining deformation information of unstable hillslopes [46][47][48][49][50][51][52]. However, daily precipitation on 2 September 2018 (timing of XW landslides 1 and 2) was only 0.4 mm. We suggest that cumulative effective antecedent precipitation (EAP) may have played a key role in the occurrences of these landslides. According to the China Meteorological Administration [53], the number of heavy rainfall days (25 mm < precipitation < 50 mm within 24 h) was seven during the study period ( Figure 15). The EAP on 2 September 2018 reached 34.8 mm (n = 7) and 39.0 mm (n = 10). Thus, although the daily precipitation on 2 September 2018 was only 0.4 mm, the landslides occurred due to the effect of the antecedent precipitation. Moreover, this study area is mainly covered by layered clastic rocks (Figure 2). The poor physical and mechanical properties of clastic rocks render them more susceptible to sliding than compared to other hard rocks, such as intrusive rocks [40,41]. The strength of clastic rocks is also significantly influenced by water content. Thus, effective antecedent precipitation played a key role in landslide failure.
during the study period ( Figure 15). The EAP on 2 September 2018 reached 34.8 mm (n = 7) and 39.0 mm (n = 10). Thus, although the daily precipitation on 2 September 2018 was only 0.4 mm, the landslides occurred due to the effect of the antecedent precipitation. Moreover, this study area is mainly covered by layered clastic rocks (Figure 2). The poor physical and mechanical properties of clastic rocks render them more susceptible to sliding than compared to other hard rocks, such as intrusive rocks [40,41]. The strength of clastic rocks is also significantly influenced by water content. Thus, effective antecedent precipitation played a key role in landslide failure.
In fact, EAP is a key parameter in most landslide early warning systems, which are based on precipitation thresholds, e.g., the daily rainfall in 24 h and the cumulative rainfall in the previous 15 days are used for landslide advisory in Hong Kong [54]. In the Seattle area, USA, the empirical rainfall thresholds for landslides were confirmed to be a result of antecedent precipitation 3-15 days before landslide activity [49,55]. Landslides triggered by EAP are always potentially catastrophic because villagers have lowered their alertness. Thus, longer rainfall durations, even when each daily amount of rainfall is small, always results in a larger EAP for landslide occurrences. More attention should be paid to EAP, especially in under-developed mountainous regions. Figure 15. Daily precipitation and effective antecedent precipitation from 1 July 2018 to 30 September 2018 at the Tuwacang meteorological station. K is the decay coefficient due to evaporation, and n is the days before the landslide. Here, K was identified as 0.84.

Influence of Adjacent Landslides
Rainfall, earthquakes, and anthropogenic activities such as road construction, irrigation, and coal mining are responsible for the occurrences of most landslides [56][57][58][59][60][61][62]. In this study, we found that the deformation of an existing landslide itself can also trigger new adjacent landslides. In fact, many studies have reported that early landslides are likely to cause subsequent landslides in adjacent areas due to local topography and morphology changes caused by earlier landslides [63,64]. The degree of overlap between earlier and later landslides is a key assessment factor for previous landslides triggering later landslides [3,64]. In this study, we calculated the degree of overlap between the old landslide, which occurred on 8 September 2007, and the XW landslides 1 and 2, which occurred on 2 September 2018, following Samia et al. [64]. The result showed that 56.8% of XW landslide 2 spatially overlapped the older landslide (Figures 1b,c and Figure 5b). Thus, the XW landslide 2 may be reactivated with a larger scale due to the extrusion of the XW landslide 1. Hillslope remained in a critical state after the previous landslide. This means that the topographic instabilities remaining after an earlier landslide need to be removed Figure 15. Daily precipitation and effective antecedent precipitation from 1 July 2018 to 30 September 2018 at the Tuwacang meteorological station. K is the decay coefficient due to evaporation, and n is the days before the landslide. Here, K was identified as 0.84.
In fact, EAP is a key parameter in most landslide early warning systems, which are based on precipitation thresholds, e.g., the daily rainfall in 24 h and the cumulative rainfall in the previous 15 days are used for landslide advisory in Hong Kong [54]. In the Seattle area, USA, the empirical rainfall thresholds for landslides were confirmed to be a result of antecedent precipitation 3-15 days before landslide activity [49,55]. Landslides triggered by EAP are always potentially catastrophic because villagers have lowered their alertness. Thus, longer rainfall durations, even when each daily amount of rainfall is small, always results in a larger EAP for landslide occurrences. More attention should be paid to EAP, especially in under-developed mountainous regions.

Influence of Adjacent Landslides
Rainfall, earthquakes, and anthropogenic activities such as road construction, irrigation, and coal mining are responsible for the occurrences of most landslides [56][57][58][59][60][61][62]. In this study, we found that the deformation of an existing landslide itself can also trigger new adjacent landslides. In fact, many studies have reported that early landslides are likely to cause subsequent landslides in adjacent areas due to local topography and morphology changes caused by earlier landslides [63,64]. The degree of overlap between earlier and later landslides is a key assessment factor for previous landslides triggering later landslides [3,64]. In this study, we calculated the degree of overlap between the old landslide, which occurred on 8 September 2007, and the XW landslides 1 and 2, which occurred on 2 September 2018, following Samia et al. [64]. The result showed that 56.8% of XW landslide 2 spatially overlapped the older landslide (Figure 1 b,c and Figure 5b). Thus, the XW landslide 2 may be reactivated with a larger scale due to the extrusion of the XW landslide 1. Hillslope remained in a critical state after the previous landslide. This means that the topographic instabilities remaining after an earlier landslide need to be removed by further mass movement. In this manner, the failure of XW landslide 1 became a key trigger for the reactivation of the old landslide on a larger scale.

Conclusions
In this study, we observed three landslides that occurred on adjacent hillslopes in the same period and were characterized by different movement processes using UAV surveys, time-series InSAR and DInSAR analyses, and electrical resistivity tomography. This combination of methods provided us with the opportunity to obtain a better understanding of the landslide processes in the surface and subsurface regions of these landslides. Road S306 was torn apart by XW landslides 1 and 2 on 2 September 2018. The UAV and field investigations revealed that the hillslope was deformed due to interaction between landslide materials and local topography. Only a few of the surface features could be used to build a surface movement vector field. Based on time-series InSAR and DInSAR analyses, we confirmed that the XW landslides 1 and 2 occurred suddenly without significant early deformations under the effect of effective antecedent precipitation. Moreover, the poor physical and mechanical properties of clastic rocks were also key factors in the formation of multiple landslides. We found that the failure of XW landslide 1 was partly in response to XW landslide 2. This means that a landslide can also trigger another landslide under special topographic and local natural conditions. However, the XST landslide, which occurred on an adjacent hillslope, is a typical retrogressive landslide that occurred through slow creep deformation over about one year and eventually slipped. Our results suggest that InSAR detection of regional potential hazards may have difficulty identifying some catastrophic landslides that lack significant early deformation. We need to learn to live with the risk of landslides because natural hazards are unavoidable.

Data Availability Statement:
The data presented in this study are available upon request from the corresponding author.