Applications: Subsidence and Landslides (3)

We discuss current and future opportunities, as well as challenges of PSI applications to slope hazard assessment and monitoring, with reference to the recent and upcoming radar satellite launches and the most recent literature in this field (e.g. Bianchini et al., 2014; Wasowski and Bovenga, 2014a,b). In particular, it is envisioned that by offering regular globe-scale coverage, improved temporal resolution (weekly or better) and freely available imagery, new radar satellite missions such as the ESA’s Sentinel-1 will guarantee an ever increasing and more efficient use of PSI in landslide investigations. These background missions are necessary for long-term, systematic mapping of unstable or potentially unstable slopes and regional scale assessment of landslide processes. The initial wide-area (reconnaissance) approaches relying on medium resolution PSI (e.g. ENVISAT, Sentinel-1) can be suitably integrated with high resolution PSI relying on the new radar sensors (e.g. COSMO-SkyMed, TerraSAR-X, RADARSAT-2), thereby providing most valuable information for the spatial and temporal analyses of slope deformation and a sound basis for derived products ranging from individual landslide monitoring to regional hazard identification.

The benefits of multi-sensor and multi-scale investigations (from regional to site-specific) are discussed by presenting PSI results concerning two regions (Central Europe and Western China) characterized by distinctly different topographic, climatic and vegetation conditions. It is shown that, with respect to the medium resolution PSI products (based on ERS and ENVISAT imagery in our case studies), the PSI products derived from high resolution imagery (Stripmap COSMO-SkyMed and TerraSAR-X) may not always lead to a significant improvement in terms of detection of unstable slopes or kinematic characterization of active landslides.

Therefore, for its most effective use PSI will have to be tailored to the specific region and site conditions, landslide types, depending on the primary objectives of the investigation. For example, in case of critical facilities at risk such as dams or bridges, of urbanized landslides or potentially unstable slopes in urban/peri-urban areas, as well as of slopes traversed by critical lifelines and engineering structures, the cost of acquiring and processing high resolution radar data can easily be justified. Conversely, in wide-area regional investigations, the use of medium resolution imagery will be more appropriate and the most cost-effective.

In general, thanks to the improving temporal and spatial resolutions of new generation radar sensors, significant breakthroughs are expected in detailed slope instability process modelling (e.g. kinematic and geotechnical models), as well as in the understanding of spatial and temporal patterns of landslide movement/activity and their relationships to causative or triggering factors (e.g. precipitation, seismic loading). The capability to provide, at regular intervals, long-term ground surface deformation trends offers an unprecedented opportunity for early detection and warning of potential slope instabilities, but further research is needed with focus on the integration of data from PSI and ground-based geotechnical monitoring. We consider this as one of the top applied research priorities.

Finally, we stress that PSI-derived results have yet to be fully explored, in particular those based on high spatio-temporal resolution data. Some of the landslide research and application areas that may benefit more from PSI include:

- Long-term behavior and climatic controls of very slow deeper landslides and  deep-seated gravitational slope deformations
- Numerical modeling of very slow persistent landslides and long term evolution of slopes
- Post-earthquake landslide activity and evolution of slopes
- Non-linear kinematics of landslides, maximum velocities and accumulated displacements

Acknowledgements

 ENVISAT, COSMO-SkyMed and TSX data were provided, respectively, by ESA under CAT‐1 project ID 2653, by the Italian Spatial Agency (ASI) within the COSMO-SkyMed AO Project ID 1820, and by DLR under TerraSAR‐X General AO Project ID MTH0432.

References

Bianchini S., Ciampalini A., Raspini F., Bardi F., Di Traglia F., Moretti S., Casagli N. 2014. “Multi-Temporal Evaluation of Landslide Movements and Impacts on Buildings in San Fratello (Italy) by Means of C-Band and X-Band PSI Data”. Pure Appl. Geophys. DOI 10.1007/s00024-014-0839-2

 Wasowski J., Bovenga F. 2014a. “Investigating landslides and unstable slopes with satellite Multi Temporal Interferometry: Current issues and future perspectives”. Engineering Geology 174: 103–138. http://dx.doi.org/10.1016/j.enggeo.2014.03.003

 Wasowski J., Bovenga F. 2014b. “Remote Sensing of Landslide Motion with Emphasis on Satellite Multitemporal Interferometry”. In T. Davis, J. Shroder eds. Landslide Hazards, Risks, and Disasters, Elsevier Inc., p. 345-403. http://dx.doi.org/10.1016/B9

Surface movements and landslides monitoring with satellite-based DInSAR techniques is becoming more and more widespread among research groups and scientific community. The SAR Differential Interferometry (DInSAR) is one of the most powerful devices for monitoring deformation processes on the Earth surface. Here, a dataset of TerraSAR-X StripMap imagery covering almost the whole Crotone province territory, located in the south of the Italian peninsula, has been selected. The time span goes from April 2008 to June 2010 and from August 2013 to October 2014. Crotone province is severely affected by landslide phenomena, due to the geological and geomorphological context, as well as to the physical and mechanical properties of the involved materials. According to an accurate bibliographic and newspaper research, in agreement with field surveys, as to find the most suitable cases for the use of the satellite interferometry, different case studies have been selected. Hence, the attention has been concentrated on several urban settlements. The most important are Papanice, Cutro, Santa Severina, San Mauro Marchesato, Cirò, Crucoli, in which, several landslide evidences, affecting roads and buildings, have been recognized.  Particular attention has been focused on slow and intermittent landslides, highly detectable by DInSAR techniques. In this work, two different multitemporal interferometry (MTI – Wasowsky et al., 2014) approaches and two different software packages have been used and compared in order to identify benefits/constraint of each MTI approach and each software. Such approaches are: a) the “permanent” (or “persistent”, or “point-like”) scatterers (PS - Ferretti et al., 2001) and the “small baseline subset approach” (SBAS - Berardino et al., 2002; Lanari et al., 2004) implemented both in the SARscape software, developed by the SARMAP team, and on SUBSOFT processor, this last based on the Coherent Pixel Technique (CPT - Mora et al., 2003, Blanco et al., 2008), developed by the Remote Sensing Laboratory (RSLab) group, from the Universitat Politècnica de Catalunya (UPC). For every technique, displacement time series have been obtained and compared. The interferometric analysis has shown good results, being able to identify displacement rates up to 4 cm/year, and also allowing to redraw landslide boundaries, previously identified in the Hydrogeological Setting Plan of the Calabria region (2001), updating their state of activity, as well. Moreover, being the available dataset antecedent to the landslides’ reactivations, the adopted DInSAR techniques have been able to recognize precursor stages of slope failures, hence confirming the reliability of SAR methods as powerful monitoring and prediction tools.

Conventional differential Interferometric Synthetic Aperture Radar (dInSAR) techniques are routinely used in accurate deformation mapping including landslide activities. However, in the case of very fast slope movements, due to the limitations of dInSAR with regard to the maximum detectable displacement (MMD), dInSAR techniques are not able to correctly measure these high landslide rates.

According to the literature and previous work of our colleagues, the large phase gradients induced by fast slope movements in Shuping landslide site could not be detected by dInSAR techniques even using 1m resolution TerraSAR-X Spotlight data, because the MMD in radians is π between two pixels. This resulted in an under-estimation of ground deformation magnitude and InSAR totally missing the actual active period of the Shuping landslide. Sub-pixel Offset Tracking (sPOT) has previously been applied to monitor glacier movements, volcanic activities and co-seismic tears in the solid earth resulting from severe earthquakes to address the technical defects and limitations of conventional dInSAR techniques, particularly their sparse coverage and the impact of dense vegetative cover. As an alternative method, sPOT has recently been employed to derive centimetre-level landslide rates in the Three Gorges Region of China with the availability of high-resolution (≤3m) space-borne SAR imagery. A previous study indicated that reliable deformation magnitudes can be measured from high-contrast surface features (e.g. artificial Corner Reflectors, rocks, buildings). However, there has been controversy concerning the interpretation of results on whether this method is able to accurately measure the “bare earth” landslide rates in densely vegetated terrain.

In this study, continuous landslide activity in the Shuping area was measured using the sPOT method applied using two nearly continuous annual time series of TerraSAR-X (TSX) 1m Spotlight data acquired from February 2009 – April 2010 and February 2012 – February 2013. The results were further validated by comparing the measurements derived using two independent sPOT tools provided respectively by COSI_Corr and SARScape. Results indicate that from a statistical standpoint, COSI_Corr and SARScape derive the same measurements of the Shuping landslide in terms of the landslide rates in addition to a consistent seasonal pattern showing a dramatic increase of landslip from May to August in both 2009-2010 and 2012-2013. This is found to be associated with changing water levels of the Three Gorges Reservoir, with landslide active periods coinciding with substantial changes of the reservoir water level during the flooding season. These results suggest a strong connection between the formation of landslides and the operation of the Three Gorges Dam as well as seasonal rainfall. In addition, a statistical analysis was carried out on the deformation measurements derived from manmade Corner Reflectors versus densely vegetated terrain. The results showed that natural scatterers show the same range of deformation magnitudes as those derived from Corner Reflectors, which illustrates that even sPOT measurements from areas suffering from lower coherence, are still able to correctly measure ground deformation range in densely vegetated terrain. This implies that such sPOT methods can be applied to the vast majority of regions for which no corner reflectors are available.

In the second analysis, with the availability of TSX Staring Spotlight (20cm azimuthal resolution) data, the effectiveness of dInSAR techniques will be re-assessed in this region as the much higher resolution increases the MMD of the dInSAR. The use of dInSAR and sPOT techniques in the densely vegetated terrain will be evaluated in terms of temporal de-correlation effect, accuracy, limitations, etc.

In future, it is intended to assess Staring Spotlight mode and local Terrestrial Lidar Scanning (TLS) observations to assess whether further “bare earth” observations can be obtained on landslip rates and deformation patterns.

The Three Gorges Project (TGP), constructed from 1994 and completed in 2009, benefits China a lot in electricity generation, flood control, navigation and irrigation. Increasing water level caused by the impoundment made many low places submerged into water. According to statics, 13 major cities, 140 towns and 326 villages were affected resulting more than 1.0million people had to move onto higher ground. The increased engineering activities for resettling people resulted to some extent of increasing unstable slopes threatening people living nearby. At the same time, the increased water level reactivated some preexisted old landslide such as Shuping landslide and Baishuihe landslide. According to investigations, there were about 560 old landslides and 36 unstable landslides in 1992-1995. But the number of unstable landslides significantly increased and about 311 of them were submerged into water since the water level reached to 175 m in 2008. Periodically filling and drawdown cycles in the reservoir areas interact with landslides and affect the stability of landslides. Therefore it is important to monitor and analyze the mechanisms of these unstable landslides for early warning and prevention purpose.

SAR images are sometimes the only available geodetic data in vast inaccessible mountainous areas. It is important to make full use of the SAR images and get the reliable measurements. Usually there are two kinds of methods for deriving deformations from SAR images: 1) the phase based methods, such as InSAR, PS-InSAR and SBAS, are suitable for slow-moving landslides monitoring with precision of millimeters. Phase discontinuities caused by temporal decorrelation would strongly affect the results in case of large deformation; 2) the amplitude based methods such as pixel offset tracking and point-like targets offset tracking (PTOT) which are more suitable for large displacement measurement with precision of more than 1/10 pixels if high correlations were guaranteed. High correlations can hardly be kept in vegetated areas and point-like targets should be made full use to improve the reliability of measurements.

34 TerraSAR-X StripMap (SM) images of approximately 3 m resolution and 36 TerraSAR-X High resolution Spotlight (HS) images of approximately 1 m resolution were collected for deformation extraction. Usually interferometric data pairs with moderate temporal baselines (a few months) and small normal baselines are suitable for the detection of slow moving landslides with C-band or L-band SAR data. Since X-band InSAR are more sensitive to temporal decorrelation, long temporal baselines make the differential phase very noisy. A differential interferogram with short temporal baseline and perpendicular baseline was firstly used to investigate potential active landslides by finding phase distortions. SBAS method which is more suitable for rural scenarios was then used to obtain the deformation histories of active slow moving landslide. Usually, underestimation on fast moving landslide will happen using phase based methods with sparse point-like targets. Point-like targets offset tracking method will then be applied to retrieve the deformation histories for cross comparing purpose. Then a risk level will be given according to the deformation rates measured by SBAS and PTOT.

After we retrieved the time series deformations on landslide, triggering factors of landslide will be analyzed. Water level fluctuation and rainfall are identified as the most two important factors affecting the landslides in Three Gorges areas. 94% of the landslides in this area were triggered by rainfall and water level fluctuation. Correlations between displacement and water level /rainfall will also be discussed. Based on our initial results, there were mainly two accelerating episodic processes during July 2008 and May 2010. The first process might be caused by consecutive rainfall while the second process might be caused by water level decline. Detailed analysis will be given next.

The accurate monitoring and characterization of active landslides represents one of the most interesting applications in Differential SAR Interferometry (DInSAR). Typically, for the characterization of active slopes different geotechnical technologies, such as inclinometers, extensometers, piezometers, or Differential Global Positioning Systems (DGPS) networks have been used. Despite providing very accurate information these in-field techniques generally provide scarce measurements in terms of coverage compared with DInSAR products and require some maintenance for long-term services. Moreover, they require the in-situ installation of instrumentation over the slopes, which may be a difficult task when the area of interest is not sufficiently serviced. In this framework, it becomes clear that DInSAR, and specially Persistent Scatterer Interferometry (PSI), present several competitive advantages compared with traditional geotechnical techniques for the effective monitoring of landslides.

 Landslides instabilities typically occur in natural environments largely affected by temporal decorrelation phenomena in which DInSAR performances generally decrease. This fact leads to a poor number of reliable high-quality measurement data, or persistent scatterers, and compromises the performance of PSI techniques. This issue is especially important in landslide monitoring applications where improving the density of reliable pixels as far as possible is a key factor to achieve a better understanding of the landslides' extension and dynamics.

During the last few years, the efforts of different research groups have been focused on increasing the number of persistent scatterers over natural environments. There are two main classical approaches in order to detect persistent scatterers: the coherence stability and the image amplitude analysis (amplitude dispersion index or SCR). The choice of the pixel selection approach depends on the nature of the area of interest and the number of images. The coherence stability pixel selection method requires performing a multi-look of the interferograms, which allows the reduction of the noise (achieving higher SNR), but also yields to a reduction of the spatial resolution. This option is more suited over natural environments with predominance of distributed Scatterers like pixels or when working with reduced data-stacks. Contrarily, the amplitude dispersion pixel selection approach preserves the full-resolution of the SAR images presenting a good performance in the detection of deterministic point-like scatterers over urban environments. There is hence a limitation when the processing requires exploiting both types of scatterers. In this context, different de-noising filtering strategies, typically employed in image processing applications to reduce additive Gaussian noise, have been adapted for InSAR and DInSAR applications, showing an excellent performance for multiplicative speckle noise. The rationale is to improve the interferometric phase quality without losing resolution in order to detect both, the deterministic point-like and the natural distributed scatterers, within the area of interest, with the objective of obtaining reliable displacements’ estimations by increasing the density of points and by keeping the coherent scatterers towards time when producing long-term observation data.

The work proposed in this paper shows an advanced PSI processing based on the use of the NL-InSAR filtering technique presented in [1] for landslide monitoring applications jointly with a more suitable temporal selection of the interferograms. The DInSAR processing carried out in this work has been addressed employing the SPN technique, the advanced PSI processing chain developed by the company Altamira-Information [3]. This methodology allows varying the number of looks pixel-by-pixel, as it depends on the number of similar patches found in a search window around the pixel under analysis, preserving the resolution over point-like scatterers and averaging homogenous regions with predominance of distributed scatterers [2]. In addition, adaptive coherent threshold selection criteria are also used to select the optimum InSAR pairs to improve the detection of coherent scatterers in long-term stacks of data. The combination of the two solutions is extremely important to improve the final density of measurement points in highly decorrelated environments and, therefore, provides  more valuable short and long-term solutions over active slopes.

Two studies will be presented in order to demonstrate the feasibility of the technique proposed for providing motion measurements over active slopes with different requirements. On the one hand, a TerraSAR-X data set corresponding to the period from August to November 2013 over a well-instrumented large-scale fast moving landslide will be deeply analyzed and compared with the in-filed measurements available. On the other hand, a large stack of ERS and ENVISAT data set corresponding to the period from 1993 up to 2008 over a slow moving ancient landslide. The results obtained by means of the proposed solutions will be carefully compared in terms of final density and quality of measurement points with the so-called classical PSI approaches showing the significant improvements achieved for landslide characterization.

[1] C.-A. Deledalle, L. Denis, and F. Tupin, “NL-InSAR: Nonlocal Interferogram Estimation,” IEEE Transactions on Geoscience and Remote Sensing, vol. 49, no. 4, pp. 1441–1452, Apr. 2011.

[2] D. Albiol, R. Iglesias, F. Sánchez, and J. Duro, “Improved characterization of slow-moving landslides by means of adaptive NL-InSAR filtering,” in Proc. SPIE: SAR Image Analysis, Modeling, and Techniques XIV, Amsterdam (Netherlands), Sep. 2014.

[3] A. Arnaud, N. Adam, R. Hanssen, J. Inglada, J. Duro, J. Closa, and M. Eineder, “ASAR ERS interferometric phase continuity,” in IGARSS 2003. 2003 IEEE International Geoscience and Remote Sensing Symposium. Proceedings (IEEE Cat. No.03CH37477), vol. 2, no. C. IEEE, 2003, pp. 1133–1135.

The Just landslide in Southern Poland is an active, creeping landslide of up to 16 mm/yr displacement. It causes the significant and persistent damage to the national road no. 75, and number of individual houses and farm buildings. Due to important road infrastructure the landslide is monitored by Polish Geological Institute-National Research Institute (PGI-NRI). Repeated in-situ measurements are carried out in inclinometer boreholes and geodetic benchmarks. Additionally groundwater level changes and precipitation data are collected. The landslides was also mapped and investigated with geological drillings and geophysical methods. Usually, these observations are sufficient for the general hazard assessment. However, it does not allow to estimate a 3D displacement field of deformation, and give a prediction of the future activity on surface. Such knowledge is essential for the proper and operational road reinforcement.

Detailed information about spatial patterns of small and local surface displacement could be derived from SAR interferometric data using the so-called multi-geometry approach. InSAR when/if validated with in-situ, and combined with subsurface geology and geophysical data may significantly contribute to the modeling of 3D dynamics of the entire colluvial mass and be used for the road, and other line type of infrastructure, management.

Since the landslide covers relatively the small area (40.5 hectares) the project is focused to evaluate SAR interferometric performance with TerraSAR-X archival StripMap, and newly acquired  Spotlight and Staring SpotLight (ST) data. SAR interferometry with spaceborne data at this spatial and temporal resolution level has not been available before. Time series analysis of archived ERS and Envisat data allowed to detect just small number of Persistent Scatterers that does not guarantee a reliable information about the spatial deformation pattern of the landslide.

Results of processing of 20 TerraSAR-X StripMap scenes show acceleration of landslide movement in its central part. InSAR observations strongly correlate with the inclinometer data, that shows a high rate of displacement occurred after heavy and long lasting rains in June 2010. The StripMap analysis will be extended with the super-resolution SAR stacks (spotlight and staring spotlight) that will allow of mapping of the deformation patterns on a high spatial and temporal scale.

If properly interpreted and validated with in-situ data it can help to fully understand variations of landslide velocity over the time and its triggering factors. Regarding to SAR data properties with such high resolution as provided by ST data, it is expected that more object will act as individual targets and thus different proportion between distributed targets and individual scatterers will be observed.