4.1. Effect of Layover and Shadow
Ascending and descending satellite imagery are acquired from different positions in space and, therefore, illuminate different areas on the ground so that the areas affected by layover and shadow are different. In other words, an area in shadow in the ascending pass might be illuminated in the descending pass and vice versa. Within the Lake Sarez study area, from the geometry of tracks N° 5 and N° 100, 23% and 46% of the area on the ground, respectively, is occluded by either layover or shadow. Much of this difference between the two tracks is accounted for by the lack of shadow areas in the descending track, a consequence of a smaller incidence angle and the specific viewing geometry. The distribution of layover and shadow areas in each case is shown in Figure 3
For Lake Sarez, it is clear that the descending geometry would be the best choice in terms of coverage if SAR data from only one track was available, with 77% of the ground able to be surveyed compared to 54% for the ascending track. However, there are still significant areas of slope on the lakeshore that would be excluded from the analysis if only the descending track was available. A potential solution to the lack of coverage from any single geometry is to combine the two geometries, in some way, to create a synergistic effect. This is certainly possible and, in this case, the coverage is increased to 88% of the land surface by combining the individual geometries, in doing so covering most of the slopes missing using the descending track only (Figure 4
Another possible analysis over such sites is to take areas that are mutually covered by the ascending and descending geometries and perform a stereo InSAR analysis, to separate out the Up-Down and East-West components of motion [50
]. In this case, the potential area available for a stereo survey is only 32% of the total land surface (Figure 4
b). Although the benefits of a stereo survey are manifold, it may be that, in this case, critically hazardous areas could be excluded from such an analysis.
For ground motion hazard monitoring in the Lake Sarez area, it is possible to conclude that it would be more advantageous to perform separate interpretations of the ascending and descending pass surveys where, in combination, 88% of the area can be assessed, instead of a stereo analysis where only 32% coverage is possible. A broader conclusion would be that, in high mountain terrain, data from both ascending and descending tracks is needed to guarantee the fullest possible coverage over steep slopes. However, opposite pass surveys are not always possible due to a lack of data in some parts of the world. For instance, the Sentinel-1 mission prioritises stereo (ascending and descending) surveys over Europe and tectonically active areas only, in order to support scientific and policy objectives, and so routine access to both ascending and descending acquisitions cannot be relied upon. Other SAR satellites are available, but these are usually at a premium, meaning that obtaining a pair of InSAR stacks over an area using such data may not be as cost-effective.
The significant difference between the coverage of the ascending and descending geometries illustrates the importance of the incident geometry and the specific slope and aspect of the terrain. If there are multiple options for surveying an area, via a choice of different satellites and geometries, a simple apriori
estimation of the layover and shadow may, therefore, help to determine which of those are optimal for a given location. Although a full simulation, as performed in this study, is necessary to completely delineate layover and shadow, estimating it using an approximate geometry is sufficient to identify the potential problems in achieving adequate coverage. This can be readily performed in a Geographic Information System (GIS) environment using only an estimate of the radar incident geometry and a DEM [42
4.2. Ground Motion
For each InSAR method and track, the following ground motion products were generated at 20-m resolution: (i) the average line-of-sight (LOS) velocity for each pixel over the 2017–2019 period of observation; (ii) a time-series of ground motion throughout the same period of observation. The former assumes a linear model of deformation, whereas the latter determines the relative change in LOS displacement between each SAR image acquisition assuming a non-linear model of deformation.
Ground motion is measured in the satellite LOS, which varies 29–46° from the surface normal across the swath from the near-to-far range. It should be noted that results from opposing geometries can appear to contradict each other. For example, the motion of an area of ground towards the satellite in the ascending data and might appear as motion away from the satellite in the descending data. However, this is simply a consequence of the side-looking geometry of the satellite, incidence angles, and steep topography (see Figure 5
The difference in coverage between the conventional SBAS method and the ISBAS method for the western part of the study area is illustrated in Figure 6
. Owing to only sparse vegetation, both techniques provide reasonable coverage of measurements over the site and identify areas that appear to be in significant motion. However, there are some significant gaps in the coverage of the SBAS results when compared to the ISBAS survey. The coverage and measurement densities for the different approaches, excluding layover and shadow areas, are shown in Table 2
. Providing more than double the spatial coverage, it is clear that the ISBAS survey is superior to the SBAS method in being able to provide ubiquitous coverage of all visible slopes in the Lake Sarez area. This concurs with the observations of studies in other regions (e.g., [44
]) and can be attributed to the intermittent nature of coherence associated with land cover types, such as vegetation and snow cover, that vary over time. With this in mind, only the ISBAS results will be considered from this point onwards.
shows the ISBAS average velocities from both geometries, and it can be seen that the wider surroundings of Lake Sarez are very dynamic. In this figure, the blue colour indicates motion towards the sensor and red away. Additionally, in this case, the ascending geometry (Figure 7
a) is viewing from the left and the descending (Figure 7
b) from the right. With a few exceptions, the observed motion can be predominantly attributed to mass movements down the steep slopes.
A notable observation is that there is very little motion towards the sensor (i.e., blue) in either ground motion map. This is in part due to the presence of layover, which may obscure foreslopes, but may also be due to the fact that the sensor is less sensitive to motion on foreslopes than backslopes. This is because it takes a larger amount of motion on a foreslope to elicit the same LOS change seen by a small shift on the backslope. This is clear from Figure 5
, where the same motion down a west-facing slope elicits a larger signal in the descending pass than the ascending pass.
Considering the known landsliding areas (adapted from [12
] and shown in Figure 7
), motion is observed within most of those locations, but not in both ascending and descending geometries simultaneously. On the whole, this is due to occlusion by layover and shadow in one result or the other.
Other significant landsliding areas are also observed outside of the known locations. One of these is a large area on the north shore that appears in both surveys, between the Bazaytash and Batasayf landslide areas (indicated as location P in Figure 6
and Figure 7
). This motion occurs on a steep south-facing slope that is not subject to significant layover or shadow. The observation that this area is subject to motion away from the sensor in both surveys implies that there is no strong lateral motion and the downward motion of the landslide is far greater than its motion towards the sensor in either of the viewing geometries. In fact, the presumption that this is a landslide can only be made based on the fact that it occurs on a steep slope, not from any notion of lateral motion. This illustrates a major drawback with the use of satellite InSAR from a single platform in a polar orbit; ascending and descending geometries may not detect the lateral movement of landslides on the north- or south-facing slopes as there will be little motion in the LOS direction. Possible solutions to this may be the integration of observations from different satellite platforms [8
] or using the multi-aperture interferometry (MAI) technique [52
Time-series of ISBAS-derived measurements corresponding to points A and B (locations indicated in Figure 6
and Figure 7
) are shown in Figure 8
and Figure 9
. Point A is located on the Right Bank landslide (also known as the Pravoberezhniy landslide), which has an area of around 3 km2
and lies approximately 5 km east of the Usoy Dam. This is an unstable slope with numerous cracks that has attracted the attention of researchers since the beginning of Lake Sarez investigations [12
]. Cracks on the slope have been monitored for many years and the main landslide area was recorded as experiencing movement of up to 100 mm/year between 1985 and 1990 [53
]. Significant downslope movement (~47 mm/year on average) is clearly detectable using InSAR (Figure 7
and Figure 8
), with motion on parts of the mass reaching a maximum velocity of 105 mm/year. This is in accordance with localised field measurements of 10–20 mm/year [22
] and ENVISAT-derived InSAR measurements up to 120 mm/year [35
], both made during the period 1998–2006. Moreover, this supports the notion that the rate of movement on the Right Bank has been constant over the past 15–20 years. It is also noted that the motion, although clear from both geometries, has a smaller range and is noisier in the ascending geometry, which again illustrates the above discussion regarding the relative lack of sensitivity over foreslopes.
Point B is situated on the Usoy Dam. The Usoy Dam is generally regarded as being stable, with reports of contemporary surface displacements not exceeding 5–10 mm/year [17
]. This is corroborated by the time-series shown in Figure 9
, which reveals a general trend of stability overprinted with a clear seasonal effect, with displacements of up to ~8 mm between spring and autumn. This motion is likely attributable to either the varying hydrostatic load of the lake acting on the dam or changes in the internal pore water pressure [53
] which will vary with the seasonal rise and fall of the water level. This may be as much as 12 m for Lake Sarez [54
], coinciding with the behavior of lake levels across the Himalayas where peak levels usually occur in the summer months when there is a higher amount of precipitation [55
]. The rise and fall of point B are in opposite sense in the ascending and descending data, which strongly indicates that this motion is much more lateral than vertical. This concurs with the suggestion of a sub-horizontal (25°) surface beneath the dam by Hanisch and Söder [53
], and implies that the observed motion is most likely attributable to the higher horizontal water load forcing the dam away from the centre of the lake in the summer and its subsequent rebound in the winter when the lake level is at its lowest.