2.1.

Study Site

The study area lies in the vicinity of Wudongde hydropower reservoir at the junction of Luquan and Huidong counties in Yunnan and Sichuan provinces, China, respectively, as shown in Fig. 1, and the study area is about $400{\mathrm{km}}^2$. The hydropower station is located in the downstream reach of Jinsha River, which cuts to a depth of about 1000 to 2000 m in the Wudongde hydropower reservoir region. Elevations range from 2000 to 3500 m above sea level on both sides of the valley, which has an average slope of $>30\mathrm{deg}$ to 45 deg. The exposed Cenozoic, Mesozoic, Paleozoic, and Mesoproterozoic strata in the reservoir area mainly include Quaternary, Cretaceous–Jurassic, Triassic, Permian, Cambrian, Sinian, Gabbro, and Granite, as shown in Fig. 2. Therefore, the rising and falling water level in the reservoir can lead to adverse changes in the geologic and anthropomorphic environment that have resulted in the occurrence of landslides at all scales on the surrounding slopes. In particular, the Jinpingzi landslide, which occurred about 900 m from the dam site, directly affects the safety and normal operation of the reservoir. A background map of the study field is shown in Fig. 1, and the suspected landslides are the areas whose features were found to be very similar to the landslide in the field survey. The investigated area is framed in a purple square in this figure and represented by a red circle in the vicinity map; detailed information (zonation and ETS point arrangement) of the Jinpingzi landslide is inserted in the lower right corner.

Fig. 1

Topographically shaded aerial photo showing names and outlines of supposed landslide areas in the study area. A vicinity map in the upper left shows the location of the study area within China. An inset map in the lower right shows the zones and ETS monitoring sites within the Jinpingzi landslide area.

Fig. 2

Geologic map of the study region.

2.2.

Available Data

Nineteen scenes of phased array-type L-band synthetic aperture radar (PALSAR-1) ascending acquisitions of the ALOS-1 satellite and two scenes of PALSAR-2 descending acquisitions of ALOS-2, with a wavelength of 230 mm (L-band), were applied. The Shuttle Radar Topography Mission digital elevation models (DEMs) (3 arc sec) were interpolated and corrected to provide auxiliary 10-m DEM data. The parameters of PALSAR data are shown in Table 1.The spatial baseline of the two scenes of PALSAR-2 descending acquisitions was 97 m. The time-position plot of the 19 PALSAR-1 ascending acquisitions is shown in Fig. 3. The minimum and the maximum spatial baselines were 205 and 4392 m, which are represented by green and red lines in Fig. 3, respectively.

Table 1

Parameters of the available PALSAR data.

Fig. 3

Time-position plot of the 19 ascending pass data scenes (the yellow point is the master image).

2.3.

Small Baseline Subset Interferometric Synthetic Aperture Radar Technology

The conventional SBAS-InSAR¹⁴ is a time series-based technique and is considered advanced compared to conventional InSAR. The proposed SBAS-InSAR is an improvement on the conventional SBAS-InSAR method. The proposed SBAS-InSAR introduced phase coherence of lower threshold into the inverse model, instead of discarding any noisy data present in a fraction of the interferograms. All the pixels were retained in the processing chain, and the weight of the observed phase data was based on the coherence of each pixel in each differential interferogram using a weighting matrix. The specific data processes are shown in Fig. 4.

Fig. 4

Flowchart of SBAS-InSAR technology. Note: SLC, single-look complex; DEM, digital elevation model; LP, low pass; SVD, singular value decomposition.

2.4.

Ground Local Incidence Angle and its Calculation Model

The GLIA is defined as the intersection angle between the propagation vector of the radar wave and the surface normal of a ground point. It is determined by the incidence and azimuth angles of the satellite as well as the slope and aspect of each ground point.^47–49

A geometric sketch of GLIA is shown in Fig. 5, where $\theta$ is the incident angle, which is the angle between the radar wave propagation vector and the vertical direction, $\partial$ is the local slope angle of the ground point, and ${\theta }_{\mathrm{loc}}$ is the GLIA. The $\delta$ and $\omega$ are the local aspect angle of the ground point and the horizontal angles of line of sight (LOS) direction, respectively, where they are clockwise rotational horizontal angles and their zero direction is the north. Finally, ${\mathbf{V}}_{\mathbf{s}}$, ${\mathbf{V}}_{\mathbf{n}}$, and ${\mathbf{V}}_{\mathbf{a}}$ are the vector of LOS direction, the vector of the ground point normal direction, and the vector composition of ${\mathbf{V}}_{\mathbf{s}}$ and ${\mathbf{V}}_{\mathbf{n}}$, respectively.

Fig. 5

Geometric sketch of the GLIA, where the three dimensions are shown as $X,Y$, and $Z$. Note: $\theta$, incidence angle; ${\theta }_{\mathrm{loc}}$, GLIA; $\delta$, local aspect angle of the ground point; $\partial$, local slope angle of the ground point; $\omega$, horizontal angles of LOS direction.

Thus, under the right sight condition, $\omega$ can be expressed as $\omega =$ azimuth angle $+90\mathrm{deg}$; under the left sight condition, $\omega =$ azimuth angle $-90\mathrm{deg}$. The right sight condition is shown in Fig. 5, as an example. The three components of ${\mathbf{V}}_{\mathbf{s}}$ can be written using Eqs. (1)–(3):

2.5.

Analysis of the Reliable Interference Region

2.5.1.

Calculation of GLIA

Both ascending and descending data from ALOS PALSAR were used in this study. The $\theta$ and $\omega$ data for every ground point could be calculated using the ALOS PALSAR data; $\partial$ and $\delta$ were obtained from the DEM (10-m spatial resolution) raster data spatial analysis.

According to Eq. (11), the GLIA of every point can be obtained. The range of GLIA values for the ascending and descending data were 26.58 deg to 101.97 deg and 32.51 deg to 116.50 deg, respectively. Negative values of GLIA represent layover areas (an overlapping of radar responses from different positions), whereas the values of GLIA higher than 90 deg represent shadowed areas (e.g., areas behind steep mountains, which are not illuminated by the radar beam). The values of GLIA at each point show differences over area. For the sake of simplicity in the following analysis, the values of GLIA were reclassified into groups with the interval of 10 deg (considering DEM data of 10-m resolution and radar data of ALOS-1 are used); values of GLIA representing layover and shadowed areas were placed into two other groups, separately. Therefore, the 11 angle groups were scattered in the study area as shown in Figs. 6 and 7.

Fig. 6

Distribution and reclassification map of the GLIA for the ascending pass data.

Fig. 7

Distribution and reclassification map of the GLIA for the descending pass data.

2.5.2.

Correlation analysis of the GLIA and interference characteristics

The interference characteristics form a description of the interference capability indices of the InSAR data set. This study mainly refers to the backscattering coefficient (BSC) and coherence coefficient (CC) of the InSAR data set. The BSC is the ratio of the grayscale value of each pixel to the maximum grayscale value (255) in power images.

To acquire the relationship of GLIA and backward scattering elements, the average values of BSCs were computed for the ascending and descending SAR images. The average values of BSC as a regional statistic were obtained according to the reclassified interval of GLIA. The specific variation trends of GLIA and BSC are shown in Fig. 8, in which BSC is the average value of all pixels in the mean power image rather than the average value of all interference point targets for the SBAS-InSAR.

Fig. 8

Variation trend diagram of the BSC and the GLIA.

The BSC decreased with an increase in GLIA (Fig. 8). However, the values of BSC were higher and more stable in ascending data with less than a 10-deg interval and in descending data with less than a 30-deg interval. Then, the gradient variation exhibited rapid slowing down after the 50-deg to 60-deg interval for both of the ascending and descending data. The BSC values were almost highest in the layover areas, but they were lowest in the shadowed areas.

To investigate the relationship between GLIA and coherence, the average values of the CC were computed according to the reclassified GLIA for the ascending and descending SAR images. The specific variation trends are shown in Fig. 9, in which CC is the average value of all pixels in the mean coherence image rather than the average value of all interference point targets for the SBAS-InSAR.

Fig. 9

Variation trend diagram of the CC and the GLIA.

The average values of CC increased with an increase of the GLIA at first and peaked at 10 deg to 20 deg (Fig. 9). Then, the CC values declined with an increase of the GLIA. However, the CC value of the ascending data had a slight increasing trend after 50 deg to 60 deg, whereas the descending data continued to decline. This showed that the variation tendency was consistent and stable before 50 deg to 60 deg and that it had a different variation tendency after the 50-deg to 60-deg interval.

2.5.3.

Locating of the reliable interference region

Several conclusions can be drawn based on Figs. 6 and 7 and the correlation of GLIA with both BSC (Fig. 8) and CC (Fig. 9). One can conclude that a smaller GLIA will produce a more intense backscatter in mountainous and valley areas. In addition, a smaller GLIA will also produce better interferometric properties in general for InSAR when applied in mountainous and valley areas. However, the detailed interference properties of SAR data were influenced by the ground target properties. Nevertheless, the interferometric properties of SAR data are relatively reliable with a 1-deg to 90-deg range of GLIA. The layover ($<0\mathrm{deg}$) and shadowed ($>90\mathrm{deg}$) areas are noneffective for InSAR when applied in mountainous and valley areas.

In this work, the distribution of effective interferometric regions for the 19 PALSAR-1 ascending acquisition scenes is shown in Fig. 10. The red layover regions and the blue shadowed regions are both noneffective regions. The other gray regions for the 1-deg to 90-deg range of the GLIA are reliable interferometric regions.

Fig. 10

Distribution map of effective interferometric regions.

3.1.

Locating Results of Regional Landslides

The displacement map (Fig. 11), which contains a detailed map showing the LOS moving average velocity, was obtained between July 12, 2007, and March 7, 2011, using the 19 PALSAR-1 ascending SAR image scenes with the SBAS-InSAR technique, and the gray areas are the noneffective interference areas marked “noneffective areas” in Fig. 11. The noneffective interference areas are not the absence of landslide disasters. They indicate that landslide disasters cannot be monitored using the 19 PALSAR-1 ascending SAR image scenes with the SBAS-InSAR technique in the noneffective interference areas. Other data or methods are needed for monitoring. Therefore, only the monitoring results of effective interference areas are analyzed in this study.

Fig. 11

Velocity map of LOS between July 12, 2007, and March 7, 2011, period.

The annual deformation rate of the LOS was also obtained in Fig. 12. Red and brown polygons are used in Fig. 11 to mark landslides and debris flows, respectively, which have occurred or are in motion based on field survey.

Fig. 12

Annual deformation rate of the LOS.

In Figs. 11 and 12, the minus signs indicate areas of subsidence and the greater than zero indicate areas of uplift. The area of subsidence was investigated because landslide movement is always accompanied by subsidence. The velocity of the various points could be clearly identified, and the range of moving velocity was 0 to 89.44 mm/year (Fig. 11). The annual deformation rate of the LOS was slightly different (Fig. 12), but the variation trend was basically the same in different regions.

Theoretically, the maximum detection capability of SAR system along the LOS is half of the wavelength by two SAR acquisitions,^10,50 and the wavelength of the L-band employed by PALSAR data was 230 mm. Therefore, considering the continuous change zone in the annual deformation rate in Fig. 12 and that the time period of the velocity was about 3.6 years in Fig. 11, the moving areas with velocity larger than 30 mm/year along the LOS direction were considered to be potentially moving landslide areas and landslide hazard areas.

As previously mentioned, the areas of brilliant color in Fig. 11, where the average velocity was larger than 30 mm/year, were mainly concentrated in the Jinpingzi landslide, at the mouth of the Anlonggou debris flow, in the catchments of the Yindigou and Matoushang debris flows, and in the Dacun-Fujiapingzi landslide group.

3.2.

Monitoring Results of the Jinpingzi Landslide

To verify the accuracy of the improved SBAS-InSAR technique for the location and monitoring of the monomerous Jinpingzi landslide, the deformation evolution of this active landslide was investigated. The displacement maps of the LOS direction were obtained from July 12, 2007, to March 7, 2011 (Fig. 13). At the same time, the annual deformation rate in the LOS direction was also obtained (Fig. 14). The displacement value of every monitoring point was revealed, and the moving displacement of the different regions can be clearly identified in the displacement map. The Jinpingzi landslide can be divided into four zones, namely I, II, III, and IV (Fig. 13), based on the local stratigraphic composition of the slide material. The results (Figs. 14 and 15) show that the displacement of zone II and a partial region of zone I were the largest; these must be regarded as the most important potential landslide disaster areas needing further investigation. The displacement of the other zones was relatively small. However, note that zone III is experiencing relative uplift that may be caused by the effects of human activities that are already in progress. Therefore, the results obtained via SBAS-InSAR were consistent with those obtained via ground monitoring results, as shown in Fig. 13 and Table 2.

Fig. 13

Geological map of Jinpingzi landslide. An inset picture in the lower right shows the physical environment of the Jinpingzi landslide.

Fig. 14

Velocity map of the Jinpingzi landslide between July 12, 2007, and March 7, 2011, period.

Fig. 15

Annual deformation rate of the Jinpingzi landslide.

Table 2

Comparison of SBAS-InSAR and ETS measurement results (units: mm).

4.1.

Locating Accuracy of Regional Landslides

The SBAS-InSAR monitoring results compared with the field survey results, which were measured by Three Gorges Geotechnical Consultants Co., Ltd., almost all measured landslides and debris flows (marked by red and brown polygons, respectively) were located in subsidence areas. In particular, landslides with a large amount of subsidence where the velocity was larger than 30 mm/year (the Jinpingzi landslide and the Dacun-Fujiapingzi landslide group) have become important study objects in ground measurement. The comparison shows that the locating results of SBAS-InSAR were exactly accurate and consistent with the field survey results.

4.2.

Single-Point Monitoring Accuracy

To test the accuracy of single-point monitoring when using the SBAS-InSAR technique, the LOS displacement values of the leveling points (green dots with black serial numbers on the Jinpingzi landslide in Fig. 16) obtained via SBAS-InSAR were compared with the ground-based ETS survey results. Therefore, the displacement vector $d_N$, $d_E$, and $d_U$ in north, east, and up direction, respectively, which were monitored by ETS, will be projected to one slant-range component $d_{\text{LOS}}$ in the radar LOS. Equation (12)⁵⁰ was used for computing the $d_{\text{LOS}}$ value of ETS survey results.

Fig. 16

LOS displacement map of the Jinpingzi landslide (92 days and 368 days).

When compared with the 92-day (January 12, 2008, to April 13, 2008) SBAS-InSAR results, the ETS survey period was from January 11, 2008, to April 19, 2008, for zone II and from January 12, 2008, to April 18, 2008, for zones I, III, and IV. When compared with the 368-day (July 12, 2007, to July 14, 2008) SBAS-InSAR results, the ETS investigation period was from June 23, 2007, to July 15, 2008, for zone II and from June 28, 2007, to July 14, 2008, for zones I, III, and IV (Table 2). The root-mean-square error was computed, in which the results of the ETS were regarded as the ground-truth values. In addition, a comparison of trend charts in Fig. 17 indicates that the variation trends of SBAS-InSAR results were in perfect agreement with the high-precision ETS data.

Fig. 17

LOS displacement trends for SBAS-InSAR and ETS (unit: mm). The compared trends of (a) 92 days LOS displacement and (b) 386 days LOS displacement.