Keywords

1 Introduction

Landslides are ubiquitous phenomena in the mountainous regions of Northern Pakistan (Said 1992; Kamp et al. 2008). Concomitant factors, like geology (the Himalaya represents one of the youngest and most prominent mountain system on the planet), tectonics (frequent and recurrent earthquakes), geomorphology (high relief energy and slope gradient), and environmental conditions (strong weathering processes) make landslide hazard a serious concern for the safety of life and property in the region (Sato et al. 2007; Owen et al. 2008; Khan et al. 2011). Besides earthquakes, heavy rainfall may cause landslides, especially during the monsoon period from July to August when the area receives the bulk of precipitation. Socioeconomic development and related human activities contribute considerably to increase the level of landslide risk. Formation of landslide dams on the streams and rivers is another significant issue in the mountainous areas of Pakistan (Fan et al. 2020). A well-known example is the Hunza lake that formed on 4 January 2010 when the Attabbad landslide blocked the flow of the Hunza River, with a risk for the population downstream in case of sudden out-bursting. However, due to the wide extension, the limited ground-truthing, and remoteness of the area, systematic studies have not been undertaken, while the boundaries and volume of landslide problem remain unknown. A detailed assessment of landslide distribution and characteristics has been advised (Sato et al. 2007).

In this context, satellite Earth Observations (EO) have shown their usefulness in landslide mapping (Sato et al. 2007). Satellite InSAR (Synthetic Aperture Radar Interferometry), in particular, can give a meaningful contribution thanks to high accuracy, wide-area coverage, the possibility of tracking the history of deformations back to early 90 s, the systematic data sampling, and higher benefit/cost ratio with respect of targeted ground-based monitoring campaigns (Tomás and Li 2017; Cigna 2018). The InSAR algorithm SqueeSAR (Ferretti et al. 2011) can provide accurate information on the possible displacement of a network of radar targets, including man-made structures (e.g., buildings, roads, and bridges) and natural features (e.g., rocky outcrops and bare soils). The SqueeSAR algorithm has been demonstrated to be a reliable technique for landslide investigations over wide areas and in natural terrain (Meisina et al. 2013; Raspini et al. 2016).

Despite more than 20 years of undisputed progress in satellite system imaging capabilities and advancement in processing and analysis of SAR acquisitions, only the launch of the Sentinel-1 constellation opened new possibilities for the operational use of satellite SAR data as a monitoring tool at the regional to national scales (Torres et al. 2012).

In this manuscript, the results obtained in the framework of the IPL project n.221 are presented. The potential of Sentinel-1 is exploited to detect and measure active movements in the Hunza-Nagar valley in North Pakistan. This is a remote area where field investigations need to be driven a by pre-screening phase to minimize efforts and prioritize further activities.

2 Description of the Study Area

The study area is located in the Hunza-Nagar valley of the Gilgit-Baltistan province, in the Karakoram Mountains, in northern Pakistan (Fig. 1). The elevation ranges from 1556 m to 7774 meters above sea level, with an average slope of 45°. This region is occupied by the high alpine humid tundra and includes the upper and middle ablation zones. It has an intense snowfall season spanning six to eight months, and a short summer. Glacial and cryogenic processes dominate, with mechanical weathering and rock fragmentation resulting in avalanches, rockfalls, and debris flows (Ali et al. 2019; Ur Rehman et al. 2020).

Fig. 1
A terrain map of northern Pakistan which marks the glaciers, unconsolidated superficial deposits of silt, sand and gravel, sedimentary and volcanic rocks, sedimentary rocks, metamorphic and volcanic rocks, limestone and Igneous and metamorphic rocks. Glacial and cryogenic processes dominate the region.

Location of study area in the Hunza-Nagar valley of the Gilgit-Baltistan province (Northern Pakistan). The pink polygon represents the area processed with Sentinel-1. Redraw of Geological Map of Pakistan (Abu-Bakr and Jackson 1964)

Full size image

Tectonically, the site is in the active collision zone of the Indian and Eurasian plates (Searle et al. 1999). Crustal shortening, active faulting, and subduction are continuing with convergence and uplift rates of ~4–5 cm/year (Searle et al. 1999) and ~ 7 mm/year (Zeitler 1985; Coward et al. 1986), respectively. The significant tectonic features responsible for brittle deformation in this area are the Main Karakoram Thrust (MKT) and Karakoram Fault (KF). Due to this brittle deformation, the rock mass is highly fractured and jointed. The MKT is the collision zone of the southern margin of the Eurasian plate, and extends into Baltistan through the Hashupa, Shigar, and Shyok valleys, respectively. MKT is a seismically active thrust fault that has a high angle along which many earthquakes have occurred (Verma and Sekhar 1986). The region has experienced many disastrous earthquakes in the past, about 300 (M > 5) and seven (M > 7) earthquakes events since 1980 (Kashmir Earthquake October 2005: M = 7.6; Afghanistan October 2015: M = 7.5) according to the USGS Earthquake Catalog (Ur Rehman et al. 2020). The Hunza-Nagar valley has a series of NE-dipping thrust faults and normal faults, producing foliation and fractures in the rocks.

The geology of the area comprises four formations, namely the Southern Karakoram metamorphic complex (SKm), the Hunza Plutonic Unit (HPU), the Shaksgam Formation (SF) and Quaternary (Q) Deposits (Searle et al. 1999). The SKm consists of paragneiss with interbedded pelites and amphibolites. The Shaksgam Formation is a subsection of northern Karakoram terrain (Searle et al. 1999) and includes Permian massive limestones. The HPU is a subsection of the Karakoram batholith and contains plagioclase, quartz, biotite and hornblende (Searle et al. 1999). Quaternary Deposits consist of alluvium deposits, unconsolidated conglomerate, sandstone, siltstone, and loess (Searle and Kahn 1996).

The geomorphology of the area comprises glaciofluvial fans, terraces, ancient glacial moraines, loose colluviated material on steep scree slopes, debris flows in debris cones, with talus deposits accumulate at the base of high cliffs in deeply dissected mountains (Hewitt 1998). Young, or recent glacial landforms play a significant role in the geomorphological landscape assemblage of the Karakoram Mountains. Lateral moraines consist mainly of supraglacial debris deposited from recent glacier margins, the transfer of englacial debris, or the pushing and redistribution processes of accumulated (pre-glacial) debris. Many active landslides occur in Karakoram ranges, especially in lateral glacial moraine deposits, where they pose severe dangers to roads, house, buildings and community infrastructures. Moraine deposits are unstable glacial landforms, characterized either by discrete shear plane failures due to steep slopes (> 30°), or by constant deformation resulting from weak mechanical properties of the ground and moisture (Su et al. 2022). Due to repeated glacier advances, older moraine cores are superimposed by younger supraglacial debris (Kamp and Haserodt 2004). These huge lateral moraines host many villages, and the inhabitants use them extensively for agricultural purposes, e.g., the Humarri village. The Humarri village is situated on the moraine terrace of the formerly glaciated region, and the village is located on the wall of the MKT. High-grade metamorphic rocks of the Karakoram block are exposed above the moraine ridges. The Hispar River flows 300 m below the left side of the hill. A local fault, called Humarri fault, passes near the Humarri Terrace in a NE-SW direction.

Humarri landslide is 2.800 m in length, 1.600 m in width, and about 300 m in maximum thickness. The steepness of the slope varies from 25° to 70°, facing the northeast direction. The sliding body consists of unconsolidated moraine materials, while the bedrock is predominantly ortho- and para-gneiss (Su et al. 2022).

3 Methods

3.1 Generation of Ground Deformation Maps and Time Series

C-band (center frequency 5.405 GHz and wavelength 5.6 cm) images from the ESA Sentinel-1 archive were processed by means of the SqueeSAR algorithm (Ferretti et al. 2011) to obtain the ground deformation map of the study area. A total of 85 images acquired along ascending geometry were used, covering the interval from February 2017 to January 2020.

The SqueeSAR algorithm represents the evolution of the Permanent Scatterer SAR Interferometry (PSInSAR) (Ferretti et al. 2001). The process is aimed at identifying radiometrically stable reflectors called Permanent Scatterers (PS) and exploiting whole stacks of SAR images. Every PS represents a pixel of a SAR image with high signal coherence, mainly corresponding to bedrock outcrops and man-made objects. The main drawback of this technique is the low density of PS that could be obtained in agricultural, natural and peri-urban areas (< 10 PS/km2 using C-band radar images), compared to the PS density in urban areas (> 100PS/km2 using C-band radar images).

The SqueeSAR algorithm was designed to overcome this limitation, defining a new type of PS points: the Distributed Scatterers (DSs), thus maximizing the spatial density of measurement points. A DS is a target corresponding to an area, where a certain number of neighboring pixels share similar reflectivity values and moderate interferometric coherence. By combining and merging the signal of these pixels, it is possible to extract a point-like feature with high interferometric coherence. DSs typically match with homogeneous areas such as deserts, debris, or uncultivated areas. The SqueeSAR approach jointly process PS and DS and considers their different statistical behaviors. This is possible through space adaptive filtering with DespecKS (Ferretti et al. 2011), which statistically averages homogeneous pixels (SHP) while preserving pointwise information associated with point targets. A filter based on the Kolmogorov-Smirnov statistical test (Stephens 1970; Kvam and Vidakovic 2007), was applied to all pixels within a certain estimation window, centred on the pixel under analysis (Ferretti et al. 2011). After that, it was possible to: (1) despeckle amplitude data; (2) filter interferometric phase values; and (3) estimate coherence values properly. The SqueeSAR analysis was designed to identify a dense grid of measurement points (MPs), either PS or DS, for which it would be possible to estimate the mean deformation velocity (usually measured in mm/yr) and the displacement time series (TS) (cf. Ferretti et al. 2011). Both of these parameters were estimated with millimeter accuracy (Colesanti et al. 2003) along the satellite Line of Sight (LOS).

3.2 Generation of Acceleration Maps

TS represent the most advanced SqueeSAR product, providing a deformation history over the analyzed time interval. As such, TS are basic tools for studying the kinematics of a given phenomenon, highlighting any potential changes occurred during the monitoring period (Zhang et al. 2018), such as sudden accelerations prior to a landslide failure (Intrieri et al. 2018; Carlà et al. 2019). More recently, TS were successfully exploited to set up a continuous monitoring system at the regional scale based on systematic imagery processing (Raspini et al. 2018) and identification of anomalies of movement (Raspini et al. 2019), i.e., points where a change in the dynamic of motion is occurring.

To fully exploit information contained within SqueeSAR outputs for the Humarri landslide, TS of each measurement point were analysed to identify any change in the deformation pattern. Displacement TS were analyzed year by year, to highlight kinematic variations (acceleration or deceleration, Fig. 2). This action reproduced, automatically, the manual process of identifying anomalous movement during visual inspection by radar interpreters.

Fig. 2
Six graphs of stabilization, lineal velocity and velocity increase, when away from the satellite or towards the satellite. The trends for the former case is increasing, while it is decreasing for the latter.

Screening and classification of displacement time series with respect to absence or presence of velocity variation

Full size image

4 Results

Figure 3 shows the deformation map of the area of interest exploiting the 85 Sentinel-1 ascending images, processed by the SqueeSAR algorithm. Each point on the map is classified according to its mean deformation velocity along the satellite LOS. Green colors indicate null to very low LOS deformation rates, usually between +2.0 and − 2.0 mm/yr (i.e., close to the sensitivity of the SqueeSAR technique), and reflects motionless areas. Point with LOS velocities < −2.0 mm/yr indicate surface deformation motion away from the satellite, while LOS deformation rates > +2.0 mm/yr reflect movements towards the satellite. The color scale gradations from yellow to dark red, and from light blue to violet represent increasing deformation rates.

Fig. 3
A deformation map of the 85 Sentinel-1 ascending images, processed by the SqueeSAR algorithm. Each point on the map is classified according to its mean deformation velocity along the satellite L O S. Green colors indicate null to very low L O S deformation rates, usually between plus 2.0 and minus 2.0 millimeters per year. The Humarri landslide is situated along the left bank of the Hispar River

Ground deformation map for the wide area of the Humarri landslide

Full size image

With almost seventy-one thousand points covering an area of approximately 200 km2, this map includes a wealth of information that can be exploited to scan wide areas, to spot unstable zones, and to reconstruct the deformation history of the observed scene back to 2017. This aspect is particularly relevant for inaccessible places where field inspection would be difficult, and for remote areas where ground-based monitoring is not unfeasible. SqueeSAR results indicate the presence of areas affected by active movements within the narrow and deep valleys incised by the Hunza River and its left tributary, the Hispar River. These movements are clearly related to slope instability phenomena, and as a consequence of scant to absent vegetation, the extent of the moving areas can be successfully depicted. Only the presence of snow cover and glaciers hinders the density of MPs.

The Humarri landslide (Ur Rehman et al. 2020) is situated along the left bank of the Hispar River and it is well depicted by the SqueeSAR results (Fig. 4) and by large fractures in the crown area, clearly visible within mid-resolution satellite imagery. Outside the landslide body stable points exhibit random oscillations of displacement values around zero, reflecting the absence of appreciable movements. A total of 7778 measure points (MPs) is extracted from within the landslide area, with a density of 463 MP/km2, and a mean velocity of −110 mm/year. The highest deformation rates (dark red points in Fig. 4) are located in the middle and lower part of the landslide where velocities higher than 300 mm/yr are recorded. Where the Humarri village is located, the velocity values are lower and range between −30 and − 80 mm/yr.

Fig. 4
A deformation map of the Humarri landslide depicts large fractures in the crown area, clearly visible within mid-resolution satellite imagery. Outside the landslide body stable points exhibit random oscillations of displacement values around zero, reflecting the absence of appreciable movements. The highest deformation rates indicated by dark red points are located in the middle and lower part of the landslide where velocities higher than 300 millimeters per year are recorded.

Ground deformation map for the Humarri landslide. On the right displacement time series, representative of different styles of motion have been selected and plotted

Full size image

Considering the acquisition geometry and the slope orientation, this landslide is moving towards north-east, i.e., towards the Humarri village and Hispar River valley, thus posing a very high risk for inhabitants, and a potential risk of a failure leading to the damming of the river. Different velocity patterns are identified for the different sectors of the landslide. The high variability of landslide kinematics, highlighted by different spatial patterns of deformation, is confirmed by the TS analysis of displacement, with measurement points exhibiting different temporal behavior even at a short distance (TS included on the right of Fig. 4). Despite a certain level of variability, TS within the Humarri landslide can be classified according four different styles of motion. A linear trend of deformation characterizes those points that have a constant rate of deformation and do not register any velocity variation during the monitored period (Pt. 1 in Fig. 4). Conversely, TS affected by trend changes denote a clear variation of their temporal behavior: Pt 2 and Pt 3 in Fig. 4 exhibit a single, or double acceleration during the monitored period that corresponds to slope instability with increasing displacement rate. Finally, Pt. 4 in Fig. 4 appears to indicate a progressive acceleration, contributing to an increasing strain rate in soil on the slope.

The early detection of accelerating areas, with particular attention to increasing strain rate, is the main purpose of any landslide monitoring system. As such, they can indicate a slope instability potentially leading to future landslide failure.

For this reason, to enhance information provided by SqueeSAR analysis following the manual inspection of the TS, displacement TS were automatically analyzed to determine the occurrence and magnitude of accelerations over the monitored period (2017–2019). This analysis uncovered further details, and extends spatially the results obtained with manual analysis, recognizing, year by year, those points with deformation rates higher than expected. For each area affected by accelerations during 2017 and 2019, a representative point has been selected and plotted (Fig. 5).

Fig. 5
1. Three acceleration maps for the Humarri landslide for 2017, 2018 and 2019. Point 1 and Point 2 are marked in the north-west and south-east. 2. A displacement time series plot is present on the right side with lines of decreasing trend.

Acceleration maps for the Humarri landslide. On the right displacement time series are plotted: point 1 shows a double acceleration in 2017 and 2019, while point 2 a single acceleration in 2019

Full size image

5 Discussion and Conclusions

The current focus of research is to exploit the maximum potential of Sentinel-1 to scan, detect, and quantify active movements in the Hunza-Nagar valley in North Pakistan. This is a remote area where field investigations require a pre-screening phase to minimize efforts and prioritize further activities.

The SqueeSAR algorithm applied to Sentinel-1 radar images successfully identified the active movements in the Hunza-Nagar valley. The advantage of Sentinel-1 images lies in its open access policy, the global coverage, and covering a time span of several years. The deformation map (Fig. 3) captures three main areas of deformation, detected in the narrow and deep valleys incised by the Hunza River and its left tributary, the Hispar River. In this pre-screening phase, priority for investigation are the lowest parts of the landslide on which the Humarri village is built. This landslide poses a very high risk for the inhabitants with a potential risk of river damming during failure.

The SqueeSAR results depict the body of Humarri landslide. Considering the acquisition geometry and the slope orientation, this landslide is moving towards the north-east. The density of measurement points is high due to the scarcity of vegetation on the slope. The highest deformation rates (over 300 mm/yr) are recorded in the middle and lower part of the landslide. Different velocity patterns are identified for different sectors of the landslide (Fig. 4). The analysis of displacement TS highlights the variability of landslide kinematics, with measurement points showing different temporal behavior, even at a short distance. Linear deformation trends, single or double acceleration, and progressive acceleration were all observed in the Sentinel-1 time series data. To determine the magnitude of accelerations within the monitored time period, we find that the manual inspection of displacement TS enhances the information provided by SqueeSAR analysis. For the Humarri landslide, we identified two main areas affected by accelerations. The former (Pt. 1 Fig. 5) was characterized by double acceleration in July 2017, and between July and August 2019. The latter (Pt. 2 Fig. 5) shows an acceleration between in September 2019. As expected, acceleration events correspond to the monsoon period (from July to August), when the area receives the bulk of the annual precipitation. The combination of slope steepness, scarce vegetation and heavy rains favors slope instability.

The successful application of InSAR analysis, coupled with the global coverage, and regular acquisition planning ensured by the Sentinel-1 constellation, allowed to deliver very precise and spatially dense information on ground motion. These qualities make it possible to scan wide areas, and to identify unstable zones, especially where remoteness, vastness, and climatic conditions make it difficult to perform field activities. We demonstrate the effectiveness of InSAR in the Hunza-Nagar valley in North Pakistan. Here, we identify all possible ground deformation with potential to directly impact on the community and we localize areas within the landslide where ground deformation is more intense for prioritized further investigations.