Giant, very slow rock slope deformations known as deep-seated gravitational slope deformations (DSGSD) are widespread in high mountain environments worldwide [1
]. The evolution of these phenomena is strongly constrained by inherited geological structure and long-term climatic forcing and occurs through progressive damage accumulation. This can lead to progressive strain localization and nucleation of secondary landslides, nested within the main DSGSD mass, that become faster and more sensitive to hydrological processes as damage accumulates [6
DSGSDs creep slowly over thousands of years, affecting entire valley slopes and displacing rock volumes up to billions of cubic meters [4
]. Therefore, even if displacement rates are low (usually smaller than 5 cm/year [11
]) and apparently steady, DSGSDs result in large cumulative deformations that are mirrored by different morpho-structural field evidence. These include extensional features as double-crested ridges, trenches, scarps, counterscarps, and half-grabens, usually dominating in the upper slope sectors, and compressional features like bulging, thrusting and folding, and nested large landslides in the middle-lower sectors. The kinematic interpretation of individual features and their associations provide critical information on the overall slope deformation mechanism and the patterns of both distributed and localized strain, reflecting deep-seated deformation mechanisms [14
Slow rock slope deformations cause severe degradation of rock mass strength, damage to infrastructures [15
] and possible evolution of slope sectors towards faster rates [6
]. A reliable characterization of DSGSD mechanisms and activity is thus crucial to identify possible failure scenarios and for risk reduction. Nevertheless, DSGSDs are usually characterized by: (a) strongly heterogeneous displacement patterns, associated with complex mechanisms (strain partitioning, damage localization and/or secondary landslides nested at different depths); (b) variable and often unknown trends of activity and sensitivity to external forcing [4
]. In order to investigate these points, displacements must be characterized in a spatially-distributed fashion and with a sufficiently high rate of temporal sampling.
Among the available in situ and remote landslide mapping and monitoring techniques, spaceborne differential radar interferometry (DInSAR) allows characterizing ground deformation rates from a few millimeters to centimeters per year [18
], maximizing the spatial and temporal coverage at a relatively low cost [12
]. Persistent-scatterer interferometry (PSI) techniques allow extracting coherent information on line-of-sight (LOS) displacement rates at point targets, even when spatial baselines and revisit times of satellite platforms were not favorable [20
]. Thus, PSI became a standard tool for landslide investigation, also applied to a regional appraisal of DSGSDs [12
]. However, the application of point-like data extracted by PSI processing chains to the detailed analysis of slow-moving DSGSDs in high mountain environments remains challenging. This is mainly due to low signal to noise ratio, severe atmospheric disturbances and snow cover effects that limit the occurrence of scatterers with sufficiently high coherence. This results in a sparse point cloud that is not dense enough to fully analyze the spatial displacement pattern.
Such specific problems can be overcome only by targeting DInSAR on the specific features and typical spatial-temporal scales of very slow rock slope deformations, integrating: (a) DInSAR processing over multiple temporal baselines; (b) detailed field morpho-structural constraints; (c) ground-based measurements.
In this perspective, we investigate the Mt. Mater DSGSD in Valle Spluga (Italian Central Alps), which affects a 1300 m-high glacial valley flank over an area of 3 km2
. The slope impends over Madesimo village and ski resort and, according to published data [12
], it is one of the fastest DSGSD in the Lombardia region. In order to unravel the kinematics, segmentation, and style of activity of Mt. Mater DSGSD, we integrate available SqueeSARTM
products, site-specific DInSAR processing of Sentinel-1 radar images, GPS data, and detailed morpho-structural information from aerial photointerpretation and field surveys. Our results outline a strong spatial segmentation and suggest a complex seasonal pattern of slope deformations at Mt. Mater, providing important guidance to further modeling activities aimed at risk reduction.
4. Data and Results
We characterized the global kinematics (i.e., sliding mechanism) of the Mt. Mater DSGSD by exploiting the available SqueeSARTM
dataset based on Sentinel-1 radar images (Figure 3
a,b), from which we derived the products of 2D displacement rate vector decomposition (2DInSAR). The spatial distribution of the dipT
variable along the slope allows identifying sectors undergoing lowering, slope-parallel translation or bulging/uplift, thus providing a consistent figure of global slope kinematics [31
For Mt. Mater, our results indicate an average dipT
value of about 30° over the slope, while higher values (around 80°) can be detected only close to the steep DSGSD upper scarps and to the headscarps of nested large landslides. These values and their spatial distribution provide robust insights into the global kinematics of the DSGSD. This is characterized by dominant translational sliding, with rotational sliding components associated with the head sectors of the nested large landslides mapped in the field (Figure 1
and Figure 3
values (Figure 3
d) are generally higher in the upper DSGSD sectors, reaching 80 mm/year above the curved scarp at around 2600 m a.s.l. (Figure 1
a) and testifying the ongoing active deformation of the entire slope. At and above the head sectors of the two nested large landslides, VT
values range between 30 and 45 mm/year (head of northern landslide; Figure 1
a,b) and 25–30 mm/year (above the head of southern landslide; Figure 1
a,d). Displacement rates fade to 1–5 mm/year moving towards the slope toe, where landslide materials become more crushed and crumpled and/or the true displacement vector may significantly deviate from the radar line-of-sight.
However, the sparse nature of these data hampers a sound interpretation of the spatial pattern of measured displacements with respect to the scale (i.e., shallow vs deep-seated) and heterogeneity of slope failure mechanisms characteristic of different slope sectors. Thus, to better investigate displacement patterns along the slope, we analyzed the multi-temporal interferograms generated through targeted DInSAR processing across the period June 2016-October 2019 (Figure 4
In all the processed interferograms, a low coherence area (<0.4) due to patchy vegetation and thermal stratification of the valley air, hampers the possibility to detect deformations close to the slope toe. This effect becomes more severe when increasing the temporal baseline from 24 days to 1-year, in which, the low coherence sector extends up to the middle slope sector because of increasing temporal decorrelation.
Interferograms with a baseline shorter than 24 days (i.e., 6, 12 days) do not show significant displacements, except for localized superficial debris covers or periglacial forms. From 24 days on, a triangular-shaped moving sector emerges from an almost homogeneous ground displacement signal (Figure 4
a). It extends between the N-S curvilinear scarp trending at 2500 m a.s.l. and the DSGSD headscarp at 2900 m a.s.l. (Figure 1
) and is made of damaged rock with extensive debris cover and periglacial features. This is the fastest slope sector, as it always decorrelates in longer baseline interferograms and is characterized by average displacement rates of 10 mm/month (Figure 4
These values are consistent with data provided by GPS benchmarks, all located within the upper slope sector between 2500 and 2650 m a.s.l. (Figure 1
and Figure 5
a). As few GPS measurements were carried out during each year, for this comparison we referred to displacement rates averaged over the entire time span covered by GPS (2015 to 2019), that were projected along the descending Sentinel-1 satellite LOS.
GPS data show evident seasonal trends, characterized by increased displacement rates in the early summer-autumn period (June to October) as an effect of combined snowmelt and rainfall (Figure 5
b). In fact, due to the high elevation (2800 m a.s.l.), snowmelt in this sector usually starts around the end of May, later than usually observed at the valley floor. In the snowmelt period, major portions of the slopes are still covered by snow, hampering DInSAR processing [12
] and accurate quantification of snowmelt contributions to displacements.
Seasonal interferograms (Figure 4
c) with longer temporal baselines clearly highlight continuous displacement fields, corresponding with: (a) localized deformation along the major DSGSD scarps at 2600 m and 2700 m a.s.l. (Figure 1
and Figure 4
d), occurring at rates of several millimeters per seasonal cycle. This signal is very different from that of shallow debris movement detected in 24-day interferograms (Figure 4
b) and reflects deep-seated slow deformations; (b) displacements of the head and internal sectors of the northern nested landslide below 2400 m a.s.l. (Figure 1
and Figure 4
d), at rates up to 5 mm/month depending on the period (Figure 6
); (c) movements up to 3–4 mm/month (Figure 6
) of the southern nested landside head at 2100 m a.s.l. and the strongly segmented above sector, up to a neat curved scarp at 2500 m a.s.l. (Figure 1
and Figure 4
d). DInSAR-derived displacement maps report for these sectors velocities ranging from a mean value of 1.5 to 4 mm/month, depending on the considered seasonal period (Figure 6
). This suggests a differential seasonal response also for the middle slope sector, with variable trends conditioned by the snowmelt timing and rainfall input in the across winter and springtime.
To better constrain this observation, we selected three slope sectors (Figure 4
f) corresponding to the nested landslides heads and above scarps (sectors 1 and 2) and the area in between (sector 3). We compared displacement rate distributions between the three sectors during consecutive seasonal periods, from 2016 to 2019. Boxplots in Figure 6
show that displacement rates change seasonally in response to the magnitude and timing of different hydrological inputs, with variable trends through the considered years. In 2016–2017, when snowmelt started early in April (Figure 5
b), displacement rates recorded across the snow cover periods (Oct-Jun; about 4 mm/month) were higher than in the summer period (Jun-Oct; about 1.5 mm/month).
An opposite behavior is observed for 2017–2018. In this case, snowmelt occurred later (Figure 5
b) and its contribution is recorded by the summer interferogram, with larger displacement rates between June and October (Figure 6
). In the 2018–2019 period, despite late snowmelt (May; Figure 5
a), abundant rainfall during autumn and spring triggered an anticipated acceleration, with mean LOS displacement rates of about 3.5 mm/month. While consistently reflecting seasonal changes, displacement rates in the three considered sectors confirm a sharp internal segmentation of the DSGSD. The northern nested landslide (sector 1) is always faster than the southern one (sector 2) and both are faster than sector 3, which is interpreted as representative of the background DSGSD activity.
Because of the low signal to noise ratio and poor coherence, we could not derive reliable unwrapped displacement maps from 1-year interferograms. However, wrapped phase maps (Figure 4
e) still provide valuable information on the persistent, pluri-annual displacement signal of the DSGSD. In fact, two fringe cycles are evident in the nested landslide sectors and a coherent signal turns out between them in the central part, corresponding to the previously identified sector 3. Such signal unlikely results from shallow movements in the debris cover, that would decorrelate with such a long temporal baseline.
A detailed slope-scale characterization of very slow rock slope deformations like deep-seated gravitational slope deformations [4
] remains challenging. Successful slope-scale DInSAR applications have been presented in the literature for slow-moving landslides characterized by displacement rates far exceeding 5 cm/year and relatively homogeneous displacement fields [39
]. Slower movements typical of DSGSD [13
] are close to the limits of detection of DInSAR. In fact, DSGSD is characterized by complex mechanisms, reflected by heterogeneous displacement fields and creep at very slow rates. Moreover, these phenomena occur in high mountain environments, where atmospheric and snow-cover disturbances strongly affect the quality of remote sensing products. Although persistent-scatterers interferometry (PSI) techniques proved their ability to measure displacement rates of a few mm/year for coherent targets [12
] due to their point-like nature, they often fail at capturing a complete picture of the spatial variability associated to slope failure mechanisms.
Our study shows that integrating field morpho-structural observations, remote sensing data from different InSAR techniques and ground-based monitoring allows a successful detail-scale investigation of heterogeneous, very slow-moving rock slope deformations. In particular, we characterized the activity and kinematics of the Mt. Mater DSGSD (Figure 7
), wisely using remote sensing techniques to cope with practical difficulties due to: (a) displacement rates close to the limits of remote sensing techniques; (b) strong segmentation of displacement fields, due to secondary large landslides nested at different depths and characterized by different degree of both localized and distributed material damage; (c) lack of subsurface borehole or geophysical investigation data providing constraints to the interpretation of the deep structure of DSGSD.
In the first stage, we tested the potential of available SqueeSARTM
PSI data, that are now used as a standard in the scientific community, for the characterization of very slow rock slope deformations. Performing 2DInSAR analysis to combine information from Sentinel 1A/B ascending and descending orbits, we fully exploited PSI data to retrieve a sound interpretation of the global kinematics of the DSGSD (Figure 3
). However, these data have a point-like, sparse nature, making them unsuitable to capture the spatial heterogeneity of the phenomenon. Moreover, SqueeSARTM
data processed over large areas are usually unable to account for the temporal patterns of activity of individual very slow-moving DSGSD and the site-specific atmospheric and snow cover disturbances, which greatly affect the local signal to noise ratio. Thus, we adopted a targeted DInSAR processing approach integrating multi-temporal baseline interferograms (2016–2019) constrained by detailed field morpho-structural observations and GPS data to unravel the deformation patterns and trends of activity in a spatially distributed fashion (Figure 4
DInSAR processing, specifically designed on the spatial and temporal scales of slope processes associated with DSGSD, provides data consistent with ground-based GPS measurements. As the ability of GPS to measure very small displacements is limited by several error sources such as satellite coverage, temperature changes, and ground dilation, for each GPS benchmark we considered displacement rates averaged over the entire 4-year measurement period. Despite smoothing non-linearities of the GPS time series, this reduces the errors associated with individual measurements and provides a consistent figure of persistent displacement rates. GPS and DInSAR displacement rates consistently settle within about ±2 mm/24 days, according to the specific 24-day period covered by interferograms. Despite keeping the same spatial trend of DInSAR products obtained with targeted temporal baselines, SqueeSARTM
data processed over large areas [37
] always underestimate displacement rates (Figure 5
a). This might be caused by reduced spatial coherence (quality) of the interferograms included in the SqueeSARTM
analysis (mainly associated with long temporal baselines considered and/or the occurrence of relatively large spatial/ temporal phase gradients), which might result in underestimations of local surface velocities [42
Selecting image pairs covering different temporal windows (i.e., 24-days, seasonal, annual) allows us to maximize the sensor detection capability for complex phenomena, extracting slope displacement rates, that in the studied case range between more than −15 mm/month and less than −3.5 mm/month. These very different values are typical of associated slope instability processes acting on different scales (e.g., shallow vs deep-seated sliding, nested landslides). Failing to discriminate among these different processes can lead to a misleading interpretation of the overall slope instability mechanisms with major impacts on the assessment of risk components (e.g., scenario volumes, intensity), that can be strongly over- or underestimated. Moreover, 24-day and seasonal interferograms effectively highlighted active sectors characterized by different spatial patterns and rates of displacements.
Combining mapping and morpho-structural information with remote sensing constraints on kinematic and spatial segmentation of the slope, we were able to provide an interpretative geological model of Mt. Mater slope (cross-section in Figure 7
), even without the support of investigations. Different morpho-structures and their associations are witnesses of different deformation mechanisms and provide a first insight into the kinematic behavior of the landslide. We mapped several orders of persistent scarps that dissect the slope from the edge to the toe and only a few minor counterscarps that cut the southern border of the DSGSD towards the Groppera valley. This morpho-structural assemblage suggests a translational sliding mechanism characterized by synthetic structures developing at different depth levels and bounding discrete sectors. Field surveys and aerial photo-interpretation revealed that the slope deformation is sliced by secondary large landslides, nested at different depths within the main DSGSD. These are characterized by rotational movements in the head sectors, where the highest vertical displacements are recorded, and by a variable amount of internal damage. The presence of nested sectors reflects a deep complexity of slope deformation mechanisms, controlled by multiple shear zones almost parallel to the slope profile (Figure 7
While 1-year interferograms provide a picture of long-term background DSGSD displacement signals, the combined analysis of seasonal interferograms and GPS data outline a sensitivity of the different slope sectors to hydrological forcing. Displacement rates in the middle-upper slope are dominantly sensitive to snowmelt, modulated by spring rainfall depending on the relative timing and magnitude of the two contributions. This induces differential responses across the snow cover period and the summer season. The large landslides nested in the main DSGSD mass exhibit similar styles of activity, but deform at displacement rates higher than the background signal of the DSGSD.
This may be due to the development of different shear zones at shallower depth from the toe to the top and their different degree of evolution that condition the response of the relative slope sectors to hydrological inputs.
Our approach, that can be applied also to other slow rock deformations in different geological and geomorphological settings, proved to supply key information (i.e., internal segmentation, style of activity, forcing) required to define reference scenarios for risk analysis and mitigation of a widespread, yet challenging class of slope instabilities.
Very slow rock slope deformations (e.g., DSGSD) in high mountain environments are challenging phenomena for InSAR applications, due to slow movements (usually less than 5 cm/year), spatial and temporal heterogeneity and strong impacts of atmospheric and snow-cover disturbances. We show how these issues can be handled through an integrated, targeted study approach combining field morpho-structural mapping, DInSAR analyses, and GPS data.
In this context, PSI data are very useful for a first-order characterization of slope activity and to retrieve its kinematics, but often fail to capture spatial segmentation, temporal trends and associated mechanisms for site-specific applications because of their point-like nature. DInSAR processing, targeted on multiple temporal baselines, allows us to successfully unravel the mechanisms, temporal trends of activity and forcing factors of heterogeneous DSGSDs characterized by discrete slope sectors possibly undergoing different evolution. Nevertheless, a sound interpretation of these remote sensing results requires detailed geological and morpho-structural field constraints.
Our observations depict a complex phenomenon characterized by nested sliding surfaces bounding discrete sectors that show different sensitivity to hydrological triggers and possibly undergoing different trends towards failure.