Conference Agenda

Anthropogenic activities and extreme climatic events increase landslide hazards and risks to human life, settlements and infrastructures worldwide. In-situ monitoring systems over landslide-prone slopes are often unavailable, making landslides challenging to detect and monitor in time and space. In this regard, global satellite missions, such as Sentinel-1 and -2, provide a huge amount of data which allows performing both retrospective and near real-time analyses for a better understanding of landslides cycle and external influencing factors. In this work, we combine results from Envisat, Sentinel-1, PlanetScope and Landsat in a multi-variable satellite remote sensing approach analysed with advanced statistical methods to characterise the whole deformation field of the Hoseynabad-e Kalpush landslide in Semnan province, Iran and investigate the role of meteorological and human factors that led to the catastrophic failure of this landslide in March-April 2019. The failure damaged more than 300 houses, of which 163 had to be evacuated due to the severity of the destruction.

PlanetScope 3-m resolution data (November 2018 and May 2019) were processed using Digital Image Correlation with Fast Fourier Transform (DIC-FFT) approach to assess the main failure mechanism. Multi-temporal InSAR observations from ascending and descending orbits Envisat ASAR (July 2003 to September 2010) and Sentinel-1 (October 2014 to December 2021) acquisitions were used to characterise the pre- and post-failure landslide kinematics. Principal Component Analysis (PCA) detected main ground displacement patterns over the landslide body. A hierarchical clustering algorithm was applied to the final cumulative map and digital elevation model to discretise the landslide sectors and extract average time series for further analysis. Optimised piecewise linear regression was applied to the ground displacement time series to decompose the signal into main trends and potential seasonality, which were then correlated to external factors, i.e. precipitation and reservoir water levels. The rainfall data set of The Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS) was used to obtain long-term monthly cumulative precipitation observations (2000-2022). The reservoir water level was derived utilising a GIS-based approach using Landsat-8 (April 2013 to August 2016), PlanetScope (August 2016 to December 2021) data and the Shuttle Radar Topography Mission (SRTM) 1 arc-second global digital elevation model.

The MT-InSAR results show that while previously stable, the landslide was reactivated during the water impoundment of a nearby reservoir in early 2013. The lower part of the landslide started to move at a horizontal rate of 3.5 cm/year, which accelerated up to 8.4 cm/year and propagated upslope in the following years as the impoundment of the reservoir continued. Exceptional precipitation hit the region in spring 2019, with 90% of the annual average (471 mm) only in the three months from January to March, culminating in the main landslide failure at the end of March. The main landslide failure started at the end of March and evolved mainly during April and partially in May, reaching the upper part of the final horizontal displacement offset of more than 40 m on the upper part of the landslide. In the aftermath, the landslide was still active, with trends in displacement rate comparable to the pre-failure phase, which decreased until a final stabilisation in the second half of 2021.

The complementary use of SAR and optical remote sensing techniques provided a good understanding of the pre-, co- and post-failure kinematics of the Hoseynabad-e Kalpush landslide and triggering factors for the reactivation and final failure. Our results suggest that the impoundment of a recently built reservoir reactivated a previously relict landslide and triggered a retrogressive destabilisation mechanism. During the pre-failure creeping, the landslide stability conditions permanently degraded. The exceptional precipitation of 2019 and the sudden increment of pore-water pressure were the final triggers of the landslide failure in March of that same year in a typical deep-seated failure mechanism. The outcomes of this study reveal the complex interactions between climate and anthropogenic interferences in influencing landslide kinematics and elevating the hazard of landslide reactivation and collapse. It is worth noting that only looking at Iran, over the last 4 decades, the number of major dams (locally called national dams) increased by 10 folds, from only 19 dams until 1978, with a total reservoir capacity of approx. 13 bcm (billion cubic metres) to more than 200 dams with a total reservoir capacity > 50 bcm by 2021 (from local news). Moreover, 1000s of smaller dams and embankments were built for irrigation, water supply and other purposes. The catastrophic slope failure in Hoseynabad-e Kalpush village highlights the importance of attention to landslides inventory maps of past and most recent landslides before and during the operation of such dam projects, especially when constructed in the vicinity of residential areas subject to high risk of damage and fatalities. Thus, investigating the Hoseynabad-e Kalpush landslide case is also relevant for other settings where artificial reservoirs have been built adjacent to relict landslide-prone slopes and where no or only limited in-situ monitoring data are available.

Eighteen years after the installation of the first experimental corner reflector (CR) network at Åknes in western Norway, more than 100 reflectors are now deployed at 25 sites for the operational monitoring of unstable rock slopes in Norway. These CR networks are mounted on masts and protected from snow, enabling year-round InSAR measurements that complement other in-situ measurement systems.

As part of the InSAR Norway service, developed by NORCE and operated by the Geological Survey of Norway (NGU), interferometric processing of Sentinel-1 data is performed on all CR networks with <24h latency. The resulting data are made available within the national landslide monitoring system operated by the Norwegian Water Resources and Energy Directorate, along with other in-situ measurements.

Due to the wide overlapping swaths of Sentinel-1 and the high latitude of Norway, most of the CR networks are imaged in multiple geometries, in ascending and descending paths. However, the recent loss of Sentinel-1B in December 2021 has halved the measurement frequency and removed the redundancy of the system. A potential weakness of using only Sentinel-1, especially with a 12-day revisit rather than 6, is the inability to track rapidly accelerating deformation. It is exactly this type of movement that we expect in the period leading up to a rock avalanche.

To address these challenges, we are currently exploring the feasibility of supplementing the monitoring with commercial satellites. In the summer of 2022, we acquired ICEYE Daily Coherent Ground Track Repeat images over the Lyngen peninsula in northern Norway, which is home to several high-risk unstable rock slopes with multiple in-situ instrument systems installed. We collected 58 Strip-mode images with a ground resolution of 3 meters between June 22 and September 29, totalling 99 days of coverage. The gaps were due to occasional orbital manoeuvres and commercial availability.

In this contribution, we will present the results of the initial analysis with ICEYE data, specifically:

• interferometric CR processing,

• comparison of the time series with measurements from in-situ instrumentation,

• detection capability for fast-moving rock slides using short temporal baselines,

• required adaptation of stack processing algorithms (SBAS and PSI).

We will summarise by outlining potential operational scenarios for complementing Sentinel-1-based monitoring with high-frequency, high-resolution commercial imagery.

Heavy rainfall events in mountainous areas can trigger thousands of destructive landslides, which pose a risk to people and infrastructure and significantly affect the landscape. Inventories of these landslides are used to assess their impact on the landscape and in hazard mitigation strategies and modelling. Optical and multi-spectral satellite imagery can be used to generate rainfall-triggered landslide inventories over wide areas, but cloud cover associated with the rainfall event can obscure this imagery. This delay means that for long rainfall events, such as the Indian Summer Monsoon or successive typhoons, landslide timing is often poorly constrained. This lack of information on landslide timing limits both hazard mitigation strategies and our ability to model the physical processes behind the triggered landsliding.

Synthetic aperture radar (SAR) data represent an alternative source of information on landslides and can be acquired in all weather conditions. The Sentinel-1 satellite constellation acquires SAR images every 12 days on two tracks globally, offering an opportunity to greatly improve the temporal resolution of triggered landslide inventories for long rainfall events.

First, new landslides that are well-constrained in time can be linked with specific periods of heavy rainfall, for example cloudburst events during the monsoon season. Physical models of soil water content can also be better calibrated when this timing information is available. We present a recently developed method of constraining landslide timing using Sentinel-1 amplitude time series. When tested on landslides of known timing (triggered by earthquakes or short, intense rainfall events), this method is able to detect the timings of around 30% of landslides in an inventory with an accuracy of 80%. When applied to landslides triggered during successive monsoon seasons in Nepal, this method reveals spatio-temporal clusters of landslides associated with cloudburst events and reveals how the earthquake in 2015 affected subsequent monsoon-triggered landsliding.

Second, SAR methods have the potential to detect multi-stage failure or reactivations on already denuded surfaces, for example the reactivation of a landslide that initially failed during the previous year’s monsoon season. With the exception of very-high-resolution imagery, this is usually not possible using multi-spectral satellites such as Sentinel-2. Previous studies in hyper-arid environments have demonstrated that coherence and amplitude time series are sensitive not only to the removal of vegetation but to erosion and deposition on unvegetated surfaces. This suggests that the detection of landslide reactivation should be possible, but in a more temperate environment, this signal will be complicated by changes in soil moisture, which also affect SAR time series. Using landslides in the Nepal Himalaya that are known to have reactivated during successive monsoon seasons as case studies, we will explore and present methods by which these two signals (soil moisture changes and landslide activity) can be separated. Such methods would allow us to better quantify landslide activity, with implications for both risk management and mass wasting volume estimates. The development of an integrated approach to landslide and soil moisture detection could be highly beneficial since soil water content is directly relevant to landslide triggering.

Landslides resulting from the failures of steep slopes in natural and artificial terrain pose significant threats to human life and infrastructure. The stability of these slopes is affected by a variety of factors including hydrologic activity, seismic activity, and changes in loading. Disastrous outcomes from catastrophic slope failures are unfortunately a recurring issue in a diverse range of industrial and natural settings, which has prompted considerable effort to monitor the stability of slopes through a variety of remote sensing and geophysical methods. Extreme weather resulting from progressing climate change will exacerbate environmental stresses on unstable slopes and increase the risk of failure necessitating further efforts to monitor these critical areas.

Ground-based interferometric synthetic aperture radars (InSAR) and geodetic prism systems have proven to be excellent methods to monitor near-real time deformation occurring on slopes [1, 2]; however, they are expensive and impractical to implement over large areas thus leaving some slopes inadequately monitored. Satellite InSAR time-series analysis is an effective approach to monitor deformation over large areas to complement the aforementioned techniques. The sparser spatial and temporal sampling of satellite InSAR compared with ground-based technologies is challenging and requires unique approaches to data analysis.

Inverse velocity analysis is a technique in ground-based slope deformation monitoring that has been shown to successfully forecast slope failures by fitting a laboratory-tested empirical model to measurements of the rate of surface deformation [3, 4]. The same technique is difficult to apply to satellite InSAR measurements due to much lower temporal sampling and higher noise levels which makes model fitting ambiguous. Recently, attempts at performing inverse velocity analyses of satellite InSAR datasets have produced accurate slope failure predictions in select case studies of historical slope failures [5-8]. However, the success of these studies is largely due to a priori knowledge of the location and timing of the slope failures that benefited the InSAR time series analysis and processing to highlight the deformation signature necessary for the inverse velocity analysis given the noise levels present in the data [i.e., 9]. Furthermore, there is controversy in the literature regarding the successful identification of slope failures in some case studies which highlights the imprecision of the applied methodologies and how the results of these analyses are interpreted [5-7, 10]. As such, the applicability of these approaches to successfully identify future slope failures in a variety of environments, from a variety of SAR sensors, and in a variety of acquisition geometries is questionable.

In this work, we present a novel statistical approach to inverse velocity analysis of satellite InSAR deformation time-series observations that leverages the large spatial coverage of phase measurements to constrain model estimates and provide actionable information for geotechnical hazard assessment of a given slope. Our approach is designed to work with any SAR sensor and to be site-agnostic, requiring only a priori information that is typically available for an InSAR deformation analysis (i.e a digital elevation model). We have generalized our method to apply to persistent-scatterer InSAR as well as distributed-scatterer InSAR and hybrid approaches. The fundamental hypothesis of our method is that the spatio-temporal characteristics of InSAR deformation measurements can be used to 1) automatically detect areas of unstable slopes and to 2) perform inverse velocity analysis on these regions to establish statistically significant bounds on failure predictions.

We validate our method against several well-characterized slope failures at open-pit mines and associated tailings storage facilities, which are among some of the most vulnerable infrastructure to slope failures. This presentation will describe our methodology, discuss our algorithmic assumptions, present results of the application to case studies, and assess the potential for this method to be applied in more general slope monitoring outside the context of our case studies.

References:

[1] G. J. Dick, E. Eberhardt, A. G. Cabrejo-Liévano, D. Stead, and N. D. Rose, “Development of an early-warning time-of-failure analysis methodology for open-pit mine slopes utilizing ground-based slope stability radar monitoring data,” Canadian Geotechnical Journal, vol. 52, Art. no. 4, 2015.

[2] N. D. Rose and O. Hungr, “Forecasting potential slope failure in open pit mines–contingency planning and remediation,” International Journal of Rock Mechanics and Mining Sciences, vol. 44, pp. 308–320, 2007.

[3] T. Carlà, E. Intrieri, F. Di Traglia, T. Nolesini, G. Gigli, and N. Casagli, “Guidelines on the use of inverse velocity method as a tool for setting alarm thresholds and forecasting landslides and structure collapses,” Landslides, vol. 14, Art. no. 2, 2017.

[4] T. Fukuzono, “A method to predict the time of slope failure caused by rainfall using the inverse number of velocity of surface displacement,” Landslides, vol. 22, Art. no. 2, 1985.

[5] S. Grebby et al., “Advanced analysis of satellite data reveals ground deformation precursors to the Brumadinho Tailings Dam collapse,” Communications Earth & Environment, vol. 2, Art. no. 1, 2021.

[6] P. Farina, V. Taurino, A. Ciampalini, and D. Colombo, “Spatially distributed and multi-temporal inverse velocity analysis: toward a proactive slope monitoring approach,” in International Slope Stability 2022 Symposium, Tuscon, AZ, USA, Oct. 17-21, 2022.

[7] T. Carlà et al., “Perspectives on the prediction of catastrophic slope failures from satellite InSAR,” Scientific reports, vol. 9, Art. no. 1, 2019.

[8] A. Thomas, S. Edwards, J. Engels, H. McCormack, V. Hopkins, and R. Holley, “Earth observation data and satellite InSAR for the remote monitoring of tailings storage facilities: a case study of Cadia Mine, Australia,” in Paste 2019: Proceedings of the 22nd International Conference on Paste, Thickened and Filtered Tailings, 2019, pp. 183–195.

[9] D. Holden, S. Donegan, and A. Pon, “Brumadinho Dam InSAR study: analysis of TerraSAR-X, COSMO-SkyMed and Sentinel-1 images preceding the collapse,” in Slope Stability 2020: Proceedings of the 2020 International Symposium on Slope Stability in Open Pit Mining and Civil Engineering, 2020, pp. 293–306.

[10] F. Gama, J. Mura, W. Paradella, and C. de Oliveira, “Deformations Prior to the Brumadinho Dam Collapse Revealed by Sentinel-1 InSAR Data Using SBAS and PSI Technique”, Remote Sensing, vol. 12, Art. no. 21, 2020.