Landslide Inventory in the Downstream of the Niulanjiang River with ALOS PALSAR and Sentinel-1 Datasets

Abstract

Landslide inventory and deformation monitoring is an essential task for human life and property security during the exploitation process of hydroelectric power resources. Synthetic Aperture Radar Interferometry (InSAR) is recognized as an effective tool for ground displacement monitoring with the advantages of wide coverage and high accuracy. In this study, we mapped the unstable slopes in the downstream of the Niulanjiang River with 22 ALOS PALSAR SAR images acquired from 2007 to 2011, and 90 Sentinel-1 SAR images from 2015 to 2019. A total of 94 active slopes are identified using a displacement map from the two datasets based on Small BAseline Subset (SBAS) InSAR analysis. By comparing the results from ALOS PALSAR and Sentinel-1 data stacks, we find that the number of active slopes increased dramatically. Several impact factors, e.g., earthquake, concentrated rainfall, and construction of hydropower stations, are discussed through time series analysis of typical landslides. Furthermore, nonlinear displacement of natural unstable slopes are found to be correlated with rainfall. A climate-driven model is used to qualify the relationship between rainfall and landslide displacement. Our results can provide valuable information for landslide detection and prevention.

1. Introduction

The cheap and renewable hydroelectric power resource is widely used, with huge economic benefits produced in China. However, reservoirs of hydroelectric power stations are usually built in valleys, where landslides are more likely to occur. Construction of reservoirs, which is another inducing factor of landslides, may also induce instability in its surrounding geological environments [1]. “Landslide” refers to the downward movement of rock or soil masses under the joint influence of internal and external factors, e.g., extreme rainfall, earthquakes, mining, and other human activities [2,3]. Landslides cause tremendous life and property loss worldwide each year. Hence, it is essential to conduct landslide inventory and surface deformation monitoring around hydroelectricity facilities and other possible regions for life and property protection. Additionally, landslide inventory maps can assist the reasonable use of renewable resources in mountainous regions [4,5,6].

China is identified as having one of the world’s highest landslides incidence rates, occupying 83% of the total landslides in East Asia. During January 2004 and December 2016, 4862 non-seismic landslides took the lives of 55,957 people [7]. Post-disaster investigations indicated that 70% of the fatal landslides were not included in previous inventory maps with huge threats [8]. The Guanling landslide in Guizhou Province, which occurred on June 28 in 2010, was triggered by heavy rainfall and resulted in the deaths of 99 people [9]. The Xinmo landslide in Mao County, Sichuan Province, induced by rainfall on 24 June 2017, destroyed the Xinmo village, causing 83 deaths [10,11,12]. Therefore, landslide inventory is an urgent task for landslide prevention and preparation.

Traditional methods, e.g., leveling, crack meters, and the global positioning system (GPS) are generally considered to be suitable methods for monitoring landslides, however, the identification of unstable slopes in inaccessible high mountainous areas is usually challenging with these methods. Synthetic aperture radar interferometry (InSAR) technology is able to accurately measure ground displacement at the level of millimeters or centimeters with the advantage of large-scale coverage and all-weather working capability [12]. The first study in China on landslide monitoring using differential synthetic aperture radar interferometry (D-InSAR) was conducted on typical landslides in the Three Gorges area, e.g., Xintan, Shuping landslides and Lianziya dangerous rock [13]. It was noticed that decorrelation problems induced by dense vegetation could seriously affect the application of InSAR. Time series analysis algorithms, e.g., permanent scatterer InSAR (PSI) [14], small baselines subset (SBAS) InSAR [15], and SqueeSAR [16], are able to overcome decorrelation and obtain high-precision time series ground surface displacement by making use of stable and slowly decorrelated pixels in SAR data stacks. With the massive accumulation of synthetic aperture radar (SAR) data, time series InSAR analysis methods are successfully employed to large-scale elaborate landslides inventory in mountainous areas, e.g., the Three Gorges area [17,18], Bailong River Basin [19,20], Jinsha River Basin [21], and the Dadu River Basin [11,22].

Massive collapses and landslides have occurred in the downstream of the Niulanjiang River due to intense tectonic movements, concentrated rainfall, and increasing anthropogenic activities. Large-scale landslide inventory is seldom carried out in this area. In this paper, 22 ALOS PALSAR images from January 2007 to March 2011 and 90 Sentinel-1 images from November 2015 to June 2019 are collected to map the unstable slopes in the downstream of the Niulanjiang River using the SBAS InSAR method. Multiple impact factors, e.g., rainfall, earthquakes, and impoundment of reservoirs are analyzed on selected landslides. A climate-driven model is also estimated to reconstruct the relationship between rainfall and slope displacement.

2. Study Area and Datasets

2.1. Study Area

Our study area is located in the southeast margin of the Qinghai-Tibet Plateau, which is the transition zone between the Yunnan-Guizhou Plateau and Sichuan Basin, as you can see in Figure 1. The elevation is high in the west and low in the east. The landform of this area is complex and diverse, with vertical and horizontal gullies and drastic elevation changes [23]. Due to the huge topographic elevation difference of the gully, the loose clastic materials at high places have huge potential energy. The steep hillside provides favorable conditions for the release of energy within loose deposits on the slope and the conversion of potential energy into kinetic energy, making it easier to form disasters such as landslide and debris flow.

A basic geological map of our study area is given in Figure 1b. Stratums from the Paleozoic and Mesozoic eras cover most of our study area. The Permian, Jurassic, Triassic and Cretaceous stratums are mainly composed of mudstone and siltstone [24]. The Ordovician, Silurian, Devonian, Carboniferous, and Permian stratums are dominated by limestone, dolomite, and shale [24]. The easily weathered limestone and dolomite are sources for large-scale landslides, e.g., the Hongshiyan and Ganjiazhai landslides in our study.

Multiple faults are developed with active tectonic activities in this study area. According to statistics, eighteen earthquakes with magnitude $\ge$ 5 have occurred in the Zhaotong–Ludian fault zone and surrounding areas between 1885 and 2014 [25,26,27]. Historical earthquakes larger than 2.5 are marked in Figure 1 as white circles. An earthquake with Ms. 6.5 occurred at 3 August 2014 in Ludian County, Yunnan Province. The epicenter is marked with a pink pentagram in Figure 1. This catastrophic earthquake caused 112 people to go missing, 617 deaths, and 3143 injured, with a direct economic loss of 23.6 billion Yuan [26].

The water system in the study area consists of the Niulanjiang River in the southeast, and the Sayu River in the central north, all flows west or north into the Jinsha River. The Niulanjiang River, as a tributary of the Jinsha River, is originated from Kunming and flows into the Jinsha River at Zhaotong in Yunnan Province. The main stream of the Niulanjiang River has a total length of ~423 km, and the altitude drop of its river basin is approximately 1725 m [28]. The significant elevation difference makes the Niulanjiang River an ideal choice for hydropower stations. As a third-stage hydropower station in the Niulanjiang River, the Xiangbiling hydropower station marked in Figure 1a started the first impoundment in August 2017, the nearby Hejiapingzi landslide was identified sliding rapidly after the impoundment [29].

Our study area has a plateau climate with a distinct vertical distribution of temperature, with dry winters and wet summers. Generally, the rainy season endures from June to October, which occupies approximately 80% of the annual rainfall [22,24].

2.2. Datasets

One stack of ALOS PALSAR and one stack of Sentinel-1 images are collected, respectively, in our study area. Basic parameters are given in

Table 1. Basic information of ALOS PALSAR and Sentinel-1 data stacks.

Sensor	ALOS PALSAR	Sentinel-1
Orientation	Ascending	Ascending
Image number	22	90
Heading angle	−10.45°	−12.51°
Incidence angle	34.51°	35.50°
Band	L	C
Acquisition mode	FBD/FBS	IW

. The images acquired from fine beam dual polarization (FBD) mode are oversampled to the pixel spacing of fine beam single polarization (FBS) mode for cross-mode interferometry. An image obtained on 9 January 2009 is selected as a reference and all the other images are co-registered with respect to it using the DEM assisted co-registration method. Images with temporal baselines ($\le$600 days) and spatial baselines ($\le$2600 m) are combined to generate 68 interferograms for SBAS analysis as shown in Figure 2a.

The Sentinel-1 datasets contain 90 images acquired in interferometric wide-swath (IW) mode spanning from December 2015 to June 2019. An image acquired on 21 December 2017 is selected as reference. Image pairs with temporal baselines ($\le$50 days) and perpendicular baselines ($\le$20 m) are combined and 312 interferograms are generated, as shown in Figure 2b. Different with the ALOS PALSAR images, significant variations of Doppler centroid frequency within each burst requires co-registration of 0.001 pixels to avoid phase jumps between bursts [30]. An enhanced spectral diversity (ESD) method is carried out to co-register all the images with respect to the reference with one arc second resolution SRTM DEM and Sentinel-1 precise orbit products [31,32]. Image pairs with temporal baselines ($\le$50 days) and perpendicular baselines ($\le$0 m) are combined and 312 interferograms are generated, as shown in Figure 2b. The co-registration of both datasets is carried out using GAMMA software [33].

One arc second resolution (approximately 30 m) SRTM data are used for differential interferogram generation and geocoding. Daily rainfall (resolution 0.25°) and temperature (resolution 0.5°) data are also collected from the China Metrological Data Service Center to model the driving factors of landslides.

3. Methods

3.1. SBAS InSAR Method

The single look complex (SLC) SAR images contain amplitude and phase information. Amplitude information represents the intensity of backscattered signals of ground target while phase information represents its distance to the SAR sensor. SAR interferometric techniques can monitor the displacement of a single SAR pixel in the line of sight (LOS) direction on the basis of its back scattering information by joint analysis of neighboring stable pixels.

Pixels in SAR images can be divided into point-like scatterer (PS) pixels and distributed scatterer (DS) pixels [34]. Generally, backscattered signals from point-like scatterers remain stable over a very long time period, which is ideal for ground displacement monitoring [14]. However, point-like scatterers, e.g., buildings and bare rocks, are sparsely distributed over mountainous regions. In contrast, the DS pixels are widely distributed with low phase stability.

The SBAS InSAR method [15] makes use of image pairs with short temporal and perpendicular baselines that can preserve the phase of slowly decorrelated DS pixels, e.g., bare soils and low grass areas. To further enhance the signal noise ratio of DS pixels, statistically homogeneous pixels with similar distributions can be selected and averaged [34]. The optimal phase of DS pixels are retrieved using phase linking from the statistically homogeneous pixels [35].

We initially selected pixel candidates using a loose amplitude dispersion value (0.6 in this study). We estimated the spatially correlated phase through a band-pass filter in the time domain and spatially uncorrelated phase though linear regression between the wrapped phase and perpendicular baselines. After removing the estimated phase terms, phase stability analysis was carried out with phase residuals. Pixels with temporal coherence larger than 0.6 were kept for the following analysis:

$$\gamma =\frac{1}{N}\left|{{\displaystyle \sum }}_{i=1}^N{e}^{\sqrt{-1}{\phi }_r}\right|,$$

(1)

where $\gamma$ is the temporal coherence, N represents the number of images, ${\phi }_r$ represents the residual phase, and i represents the image index.

Three-dimensional phase unwrapping on the selected pixels is carried out on the final pixels [36]. The unwrapped phase of a given pixel in each interferogram can be expressed as:

$${\phi }_{uw}={\phi }_{disp}+{\phi }_{topo}+{\phi }_{orb}+{\phi }_{atmos}+{\phi }_{noise},$$

(2)

where ${\phi }_{ uw}$ represents the unwrapped phase of pixel x, ${\phi }_{x, disp}$, ${\phi }_{x, topo}$, ${\phi }_{x, orb}$, and ${\phi }_{x, atmos}$ represent the component induced by ground displacement, topographic residuals, orbital errors, and atmospheric delay. ${\phi }_{noise}$ is the noise term. The orbital phase ramp in each interferogram can be estimated and removed with a bilinear model. The topographic residual errors are estimated by linear regression using unwrapped phases and perpendicular baselines. The undulating terrains in mountainous regions make the APS complex. The APS can be divided into stratified APS and turbulent APS. The empirical elevation dependent linear regression model is used to remove the stratified APS. Any turbulent APS was removed using temporal high-pass and spatially low-pass filters. The displacement of the ground target can be expressed as following equation [37]:

$${\phi }_{disp}={\phi }_{ nl}+\Delta t·\Delta v,$$

(3)

where ${\phi }_{disp}$ is the phase component induced by ground displacement, ${\phi }_{ nl}$ is the long-term linear displacement rate, and $\Delta t$ is the time interval. $\Delta v$ is the nonlinear displacement which might be related with rainfall or temperature changes. The linear displacement rate and nonlinear displacement can be retrieved with the least square method.

3.2. Climate-Driven Displacement Model

Rainfall is identified as an important triggering factor of landslide displacement [7]. Seasonal rainfall generally induces seasonal acceleration proportional to climate factors, e.g., rainfall and temperature. A climate-driven displacement model [38] is employed in this study to model the rainfall induced nonlinear displacement. The nonlinear displacement of the slope at time t is proportional to the nonlinear displacement at previous time and residence time of rainfall in the slope, which can be expressed as:

$$d\left(t\right)=d(t-1)·e^{-\frac{1}{\tau \left(t\right)}}+k_1·R\left(t\right),$$

(4)

where $t$ is the daily scale time vector, $d\left(t\right)$ is the rainfall-related nonlinear displacement, $R\left(t\right)$ is the daily rainfall vector, and $k_1$ is the scale factor.$\tau \left(t\right)$ is the rainfall residence time, which is a indicator of rainfall loss rate consist of run off, infiltration, and evaporation.

Generally, rainfall residence time in summer is significantly different from that in winter at the same location. A temperature related equation is used to describe the seasonal variation of rainfall residence time:

$$\tau \left(t\right)=k_2+k_3·T_z\left(t\right),$$

(5)

where $k_2$ and $k_3$ are positive model parameters which need to be solved. $T_z\left(t\right)$ are standardized values of daily temperature change ranging between 0 and 1. We assume residence time is only sensitive to temperature above zero. Temperatures below 0 are count as 0 by the following equation:

$$T\left(t\right)=\{\begin{array}{c}0, T\left(t\right)<0 \\ T\left(t\right), T\left(t\right)\ge 0\end{array},$$

(6)

Then the standardization of daily temperature changes are carried out with Equation (8) to eliminate the impact of extreme weather.

$$T_z=\frac{1}{2}\left(1-tanh\left(\frac{mean\left(T_0\right)-T\left(t\right) }{SD\left(T\right)}\right)\right),$$

(7)

where $SD\left(T\right)$ represent the standard deviation of temperature changes. The initial value d(0) can be determined by the analytical solution for the equilibrium state of Equation (4) given a mean rainfall.

$$d\left(0\right)=\frac{mean\left(R\left(t\right)\right) }{1-mean\left(e^{-\frac{1}{\tau \left(t\right)}}\right)},$$

(8)

The observed time series nonlinear displacements $d_{nl}$ of the landslide together with daily scale temperature and rainfall data were used to fit the mode.

$$d_{nl}\left(t\right)=d\left(t\right)+\epsilon ,$$

(9)

where $\epsilon$ is the error of InSAR measurements. The Markov chain Monte Carlo [39] method was used to solve the model parameters $k_1$, $k_2$, and $k_3$.

This model can be used to predict landslide displacement in the future. Landslides with abnormal displacement (e.g., significantly larger than modelled values) should be carefully monitored in case of failures. However, the coarse spatial resolutions of daily temperature (0.5°) and daily rainfall (0.25°) makes the metrological data on landslides not accurate enough which will cause the molded accuracy decrease.

4. Results and Discussion

4.1. Mean Displacement Rate and Landslide Inventory

The mean displacement rate map obtained from ALOS PALSAR and Sentinel-1 data stacks is given in Figure 3. Positive value means moving away from the sensor, while negative value means moving towards the sensor. Measurement points are detected from both data stacks. The point density for the ALOS PALSAR and Sentinel-1 data stacks are 1880 pixels/km² and 712 pixels/km², respectively. Denser measurement points are selected from the ALOS PALSAR image stack, the reason we assume, is due to the longer wavelength of L-band SAR data.

Landslide inventory is carried out by combining displacement rates and geomorphological analysis. We first extract pixels with absolute LOS displacement rates > 10 mm/yr. Clustered pixels (>10) within a diameter of 300 m are selected as hotspots. We then manually select landslides from the hotspots through texture information from optical images and geomorphological analysis of the digital elevation model.

A total of 94 active slopes with an area of 39.83 km² are detected and 34 of them are located at the overlapping area of these two datasets in Figure 4.

The maximum and minimum areas of the detected unstable slopes are 2.26 km² and 0.0054 km², respectively. There are 8 active slopes detected by ALOS PALSAR during 2007 and 2011, while the number of active slopes increased to 29 during 2015 and 2019. The increasing of unstable slopes might be related to the 2014 Ludian earthquake and intense anthropogenic activities. We also note five of the slopes detected by the ALOS PALSAR marked by the blue polygons turned stable during 2015–2019. More investigation should be carried out to confirm the results, e.g., combining descending InSAR measurements, high-resolution optical images, and filed investigations.

4.2. Time Series Displacement of Typical Landslides and Impact Factors Analysis

4.2.1. Longdongshui-Yanjiao Area

The Longdongshui and Yanjiao area (104.02°E, 27.49°N) is located at the north side of the Heng River in Yiliang County, Yunnan Province. The terrain incised from northeast to southwest with a height difference of 300 m. Optical images obtained in April 2008 and May 2016 shown in Figure 5a,b indicate a significantly reduced amount of vegetation in this area. The reclaimed land is mainly used to grow crops. The irrigation activities might increase the ground water level and decrease slope stabilities [40].

Figure 5c,d gives the mean displacement rate in the LOS direction from ALOS PALSAR and Sentinel-1 datasets.

The displacement rates are generally less than 10 mm/yr during 2007–2011 and increase to 20–50 mm/yr during 2015–2019. The displacement rate decreased from the east to the west, which is correlated with terrains. Time series displacement of point P₁ and P₂ marked in Figure 5 are given in Figure 6. P₁ is located on steep terrain with a mean slope angle of 45° while P₂ is located on gentle terrain with angles of 25°. The time series displacement of P₁ and P₂ from ALOS PALSAR in Figure 6a confirms the stable state during 2006–2011. The cumulative displacement of P₁ and P₂ reached 139 mm and 90 mm, respectively, during 2015–2019.

4.2.2. Hejiapingzi Landslide

The Hejiapingzi landslide is located at the east bank of the Yulong River in Weining County, Guizhou Province. The area of the landslide is 0.28 km², with a maximum thickness of 80 m [28]. Its center location is 103.71°E and 27.06°N, approximately 4.7 km northwest of the Xiangbiling hydropower station. The slope is steep, ranging from 40° to 60°. The impoundment of the Xiangbiling Reservoir started on 26 April 2017. Since then, the Yulong River, a tributary of the Niulanjiang River, became a part of the flooded reservoir. As we can infer from the optical images in Figure 7a,b, the river channels are significantly widened. The toe of the landslide is submerged into the Yulong River. Several cracks on the ground and houses were identified on 20 August 2017 due to the slow movement of the Hejiapingzi landslide.

The mean displacement rate map of the Hejiapingzi landslide measured from ALOS PALSAR and Sentinel-1 is given in Figure 7c,d, respectively.

The displacement rate during 2007 to 2011 are mostly less than 10 mm/yr, indicating the stable state of the landslide. However, the displacement rate in the central part of the landslide increases with a maximum number of 80 mm/yr detected from the Sentinel-1 datasets during the period of 2015–2019. It can be inferred that the landslide is a northeast-facing slope, thus the detected LOS displacement rate is positively moving toward the ascending satellite. The time series displacements of P₃, located on the north side of the Hejiapingzi landslide, are given in Figure 8. As we can infer from Figure 8a, the displacement rate of P₃ is approximately 22.6 mm/yr and the cumulative displacement rate reached 80 mm during 2015–2019. An obvious acceleration during the water level decline period is observed in Figure 8b. Rapid water level decline will make the water pressure inside the slope larger than outside, which will decrease the slope stability [41].

4.2.3. Hongshiyan and Ganjiazhai Landslides

The Hongshiyan landslide and the Ganjiazhai landslide are two of the largest landslides triggered by the Ludian earthquake in 2014 [42]. Displacement rate maps and profiles of these two landslides are given in Figure 9 and Figure 10.

The Hongshiyan landslide (103.40° E and 27.04° N) is located at the left bank of the Niulanjiang River, as shown in Figure 9a,b. It is a typical rock landslide that is mainly composed of limestone [43]. The elevation of the original landslide ranges from 1320 to 1840 m. The volume of failed deposit is approximately 12.24 × 10⁶ m³ [44]. The landslide rushes into the Niulanjiang River and forms a landslide dam as indicated in Figure 9b. The dammed lake is converted into the Hognshiyan reservoir after reinforcement (Figure 9b). As we can infer from Figure 9a and Figure 10, the displacement rate of the head area is less than 10 mm/yr, indicating a stable state during 2015–2019.

However, a displacement rate of about 15 mm/yr is identified on the landslide dam, which might be correlated with the consolidation of loose landslide deposits. We also noticed that there is an ancient landslide located in the left bank of the Niulanjiang River shown in Figure 9a,b. The area of this ancient landslide is approximately 0.831 km² and the estimated volume is 60 ± 5 × 10⁶ m³ [44]. The landslide is active and a maximum displacement rate of ≥30 mm/yr is identified at the head area of the landslide. The active state of the ancient landslide should be closely monitored in case of failures. The time series displacement of P₄ and P₅ in Figure 11 indicate both points were experiencing decelerating processes. The mean displacement rates in the LOS direction are 26.86 and 28.06 mm/yr for P₄ and P₅, respectively, and the cumulative displacement reached 95 mm and 100 mm from November 2015 to June 2019.

The Ganjiazhai landslide (103.38°E and 27.07°N) is a fan-shaped slope (Figure 9c,d) located at the west side of Shaba River, a tributary of the Niulanjiang River. The area and volume are 0.18 km² and 17 × 10⁶ m³, respectively. This catastrophic failure took over 100 people’s lives [42]. Through comparison of optical images acquired before and after the failure in Figure 9c,d, the Ganjiazhai landslide can be divided into the upper and lower parts.

The elevation of the upper part and lower part are about 1800 and 1500 m (Figure 10b). The displacement rate of the upper part is less than 5 mm/yr (Figure 9d and Figure 10b). The head area of the lower part is active with a displacement rate of about 15 mm/yr. The time series displacement of P₆ in Figure 11 indicates that the landslide was in a deceleration state. However, filed investigations found that there are large amounts of accumulation soils and limestone blocks, which is likely to fail under extreme rainfall [45].

4.3. Climate-Driven Displacement Modelling Results

The 12-day revisit cycle of Sentinel-1 can provide us with dense observations and enable us to explore the relationship between nonlinear displacement and rainfall. Locations of the three points P₇, P₈, and P₉ are selected for this purpose and given in Figure 12. The corresponding cumulative LOS displacement and the nonlinear displacement of P₇, P₈, and P₉ are given in Figure 13a,c,e. We clearly find accelerations in rainy seasons on these three points. Rainfalls that infiltrate into the slopes will increase the ground water level and decrease the shear strength. The response time might be correlated with landslide materials, thickness, and rainfall intensity [38].

Our modelled displacements in Figure 13b,d,f can describe nonlinear displacement using the climate-driven displacement model in Section 3.2. The correlation coefficients (R²) between the measured nonlinear displacement and the model results for P₇, P₈, and P₉ are 0.76, 0.68, and 0.63, respectively. The resolved parameters k₁, k₂, and k₃ are 0.0003, 26.48, and 1.18 for P₇; 0.0007, 35.27, and 0.64 for P₈; and 0.0007, 35.47, and 0.58 for P₉. We notice that parameters of P₇, P₈, and P₉ are very similar, which might be related with the close distance (about 10.6 km) and similar meteorological data of P₆ and P₇. The difference of resolved parameters might be related to the difference of meteorological data and materials of landslides. Our results show the potential of the climate-driven model in modeling the rainfall-induced seasonal accelerations.

4.4. Limitations and Future Works

The side-looking geometry makes InSAR suffer from geometric distortions. Limited information can be extracted from layover or foreshortening areas. Furthermore, the criterion that only clustered pixels (>10) are selected might omit some landslides. Combination of descending and ascending orbits datasets can decrease these impacts to some extent. The identified pixel density might vary with land covers, which will affect the detection of landslides. More rigorous landslide detection criterion should be made to improve the detection ability. We should note that InSAR can detect movement of N-S oriented slopes although it is insensitive to deformations in the N-S direction. Since movements of slopes generally occur at slope-parallel direction which means there are also vertical displacement which InSAR can measure. At the meantime, in-situ measurements, e.g., precipitation and landslide composition, are very important for the interpretation of our results. The detected landslide can also be used to produce landslides susceptibility maps for risk evaluation.

5. Conclusions

China is rich in renewable hydropower resources. In order to promote its development in human life and property safety, as well as the possible reuse of middle-risk areas with photovoltaic deployment, we mapped the distribution of landslides in the downstream of the Niulanjiang River using ALOS PALSAR acquired from 2006–2011 and Sentinel-1 from 2015–2019 through the SBAS InSAR method. A total of 94 active slopes with areas ranging from 0.0054 km² to 2.26 km² were utilized. Comparing the results from the ALOS PALSAR and Sentinel-1 datasets, a number of active slopes in the overlapping area were found to be increasing from 8 to 27 due to the joint effect of earthquakes, rainfall, and anthropogenic activities. The expansion of constructed areas induced new unstable areas, e.g., the case of the Longdongshui and Yanjiao villages. According to the ALOS PALSAR results, the Heijiapingzi landslide was stable during 2006 and 2007 and the impoundment of the Xiangbiling hydropower station contributed to the movement. The movements of the Hejiapingzi landslide should be closely monitored, especially during water level rise and decline periods. As two typical landslides were triggered by the 2014 Ludian earthquake, the Hongshiyan and Ganjiazai landslides are now also in an active state with displacement rates of ~15 mm/yr. We also evaluated the potential of a climate-driven model to reconstruct the relationship between rainfall and nonlinear slope displacement. Our model can describe the rainfall-induced seasonal displacement well, and it is promising for correlate rainfall and displacement rates. We will further explore the climate-driven displacement model with more reasonable impact factors and in-situ data.

There are a lot of inaccessible areas in mountainous southwestern China, which would be very challenging for traditional methods. InSAR provides us an alternative way to map and monitor the landslide with high accuracy. With the operation of Sentinel-1, NISAR, and Lutan-1 satellites, InSAR technology will play more and more of an important role in geohazard detection and monitoring.