Monitoring, prediction, and early warning using ground ... - CiteSeerX

Original Paper Landslides(2010) 7:291–301 DOI 10.1007/s10346-010-0215-y Received: 8 October 2009 Accepted: 21 April 2010 Published online: 29 May 2010 © Springer-Verlag 2010

Nicola Casagli I Filippo Catani I Chiara Del Ventisette I Guido Luzi

Monitoring, prediction, and early warning using ground-based radar interferometry

Abstract In order to define adequate prevention measures and to manage landslide emergencies, real-time monitoring is required. This paper presents two different applications of the remote sensing technique: the ground-based synthetic aperture radar interferometry, here proposed as a monitoring and early warning support for slope instability. Data acquisitions carried out through a ground-based synthetic aperture radar interferometer, operating in Ku band, installed in front of the observed slopes, are discussed. Two case studies, based on the use of the same apparatus (formerly developed by the Joint Research Center of the European Commission and by Ellegi-LiSALab srl), are reported: the first one concerns the monitoring of a large landslide, named Ruinon (Valfurva, Italy). The second one deals with the monitoring of the NW unstable slope in the Stromboli island aimed to implementing an early warning system. Acquired interferometric data are processed to provide displacements and velocity maps of the monitored area. The monitoring services ongoing on the Ruinon landslide and on Stromboli demonstrate the capability of this technique to operate in different operative settings (i.e., different phenomena and geological framework) and for different aims (monitoring for prevention, early warning, and emergency assessment). This methodology has also been proved by national and regional authorities of civil protection in order to provide a real-time monitoring for emergency management. Keywords Landslides . Interferometry . Ground-based SAR . Ruinon (Valfurva, Italy) . Stromboli (Italy) Introduction Landslides represent one of the major natural hazards causing fatalities and economic damage worldwide. In the countries threatened by this hazard, the local civil protection authorities are in charge of managing any slope instability emergency. Monitoring of slopes and dormant landslides is necessary to prevent disasters in unstable areas like mountains and hilly ranges or coastal cliffs. A real-time monitoring that considers the triggering factors and landslide kinematics is required in order to manage the emergency and also to define adequate prevention measures for the mitigation of loss of human life and of assets. An adequate early warning system must satisfy the requirements of data reliability and fast availability. This paper reports the updated experience of our research group on the ground-based synthetic aperture radar interferometry introduced in previous works (Antonello et al. 2004). In particular, we discuss the use of a ground-based synthetic aperture radar interferometer (GB-InSAR) not only as a monitoring system but also as a tool to obtain spatial information on the landslide displacements to be integrated with rainfall data. This technique can be incorporated in an early warning system for the detection of slope acceleration patterns indicating the upcoming occurrence of a slope failure.

Generally, the set-up of an early warning system for landslides is based on in situ instrumentation that provides measurements obtained in a few points of the monitored area, which are often not adequately representative of the whole unstable area. A remote sensing approach that allows obtaining the two-dimensional deformation maps can overcome these limitations. The large number of synthetic aperture radar (SAR) images collected within several international satellite missions—from the pioneer SEASAT to the very recent TerraSAR-X and COSMOSkyMed satellites—has led spaceborne SAR interferometry (InSAR) to play a key role within the remote sensing community and to be considered as a powerful tool in many application fields. Topographic mapping, the monitoring of ground displacements due to subsidence (Ferretti et al. 2000; Colombo et al. 2003; Farina et al. 2006a, 2007a), earthquakes and landslides (Kimura and Yamaguchi 2000; Canuti et al. 2004; Farina et al. 2006b, 2007b), and glacier motion (Kenyi and Kaufmann 2003) are some of the most investigated topics. Over the last years, spaceborne SAR interferometry has been broadly and exhaustively applied for the monitoring of mass movements showing the capability to provide deformation maps from space and at a basin scale with millimetric accuracy. Despite the large variability of landslide types and mechanisms, it has also been proved to be an efficient tool for the landslide inventory (Strozzi et al. 2005; Reidel and Walther 2008; Catani et al. 2005a; Righini et al. 2008) and monitoring (Berardino et al. 2003; Catani et al. 2005b; Corsini et al. 2006; Antonello et al. 2008). On the other hand, significant difficulties have been found to use this technique as an effective and operational tool in emergency management situations due to the acquisition parameters of the current satellite SAR missions, especially in terms of temporal coverage and incidence angle. A GB-InSAR exploits the same principle used in spaceborne InSAR but can provide images of a single slope in much shorter time intervals (about every 10 min), overcoming some of the limitations of satellite data acquisition. With respect to space-borne system, GB-InSAR offers a significantly higher image acquisition rate, coupled with the ability to provide displacement measurements over a limited area (up to a few square kilometers) with an optimal viewing geometry and a very high spatial resolution. Moreover, the GB-InSAR can operate in presence of steep slopes when satellite images perform unsuccessfully. These features show that the application of this technique provides main support for landslide monitoring and early warning systems, as tested in several cases (Tarchi et al., 2003a, 2003b; Antonello et al. 2004; Barbieri et al. 2004; Luzi et al. 2006; Antonello et al. 2008; Casagli et al. 2009). This work aims to illustrate the capability of this technique to operate in different operative settings by means of two study cases carried out by the Applied Geology Group of the University of Firenze. This work starts with a brief introduction of the basic Landslides 7 & (2010)

291

Original Paper principles of microwave interferometry and gives a concise description of the used apparatus. The two reported cases show the use of GB-InSAR both as a monitoring and as an early warning tool. The first one concerns the monitoring of the Ruinon landslide (Valfurva, Italy), one of the most hazardous slope movements in the Italian Alps. The second one concerns the application of a GB-InSAR as an early warning system for the North-Western flank instability of the Stromboli volcano. Concerning the Ruinon landslide, cumulative displacement maps are discussed demonstrating the usefulness of an imaging approach to identify unstable areas and to forecast the evolution of the landslide and its associated risk scenario. With reference to the Stromboli case, the capability of the proposed technique to provide the monitored slope deformation field in a quasi-real time and for warning purposes is shown. Methods and equipments Differential interferometry is based on the interpretation of the phase difference measured between two or more coherent acquisitions. In the usual spaceborne InSAR observations, different contributions merge into the measured phase. Firstly, the two satellite acquisitions used for the interferogram calculation are separated by a distance, the baseline, a geometrical contribution to the measured phase due to the different orbits. Secondly, a term coming from the atmospheric delay (Hanssen et al. 1999) and a noise term due to surface variations are present as well. Observing natural surfaces, coherence, the parameter that quantifies the degree of decorrelation (1 for no decorrelation, 0 for totally decorrelated images), decreases with time due to propagation and variations of scattering surface properties (Zebker and Villasenor 1992). When image pairs are taken exactly from the same position (zero baseline) and the decorrelation due to propagation and scattering phenomena can be considered negligible, the interferometric phase is directly related to the distance variation. Similarly to spaceborne radar application, ground-based radar images acquired at different times can be profitable for interferometry only if the decorrelation among different images is maintained low. The retrieval of the actual interferometric phase variation due to surface deformation is easier in a ground-based configuration for several reasons. GB interferometry can benefit from a high image acquisition rate reducing decorrelation. In addition, the observational geometry is fixed; hence, the above-mentioned zero baseline condition is always satisfied, and a single pair of images is sufficient to generate a topography-free interferogram. The spatial resolution of the interferogram is equal to those of the original SAR images; the resolution could degrade if spatial averaging is applied at some step of the processing chain. As for spaceborne interferometry, the attained displacement maps provide the component of the displacement along the line-of-sight (LOS) of the SAR system. Experimental tests carried out with several instrumentations demonstrated the capability of GB-InSAR in different application fields including the monitoring of slope stability (Leva et al. 2003; Tarchi et al., 2003a; Antonello et al. 2004). As far as the decorrelation due to propagation in atmosphere is concerned, for ground-based installations, this has been introduced and discussed in Luzi et al. (2004) and in Pipia et al. (2006). The availability of a large set of images due to the high GBInSAR measuring rate (about three to five images per hour) 292

Landslides 7 & (2010)

gives the opportunity of using averaging and specific statistical tool. Most of the GB-InSAR instruments are based on a continuous-wave step-frequency radar as microwave transceiver. A 2–3-m-long linear rail on which two antennas move with millimetric steps is used to form a synthetic aperture. The microwave transmitter produces, step-by-step, continuous waves around a center frequency. The transceiver moves along the rail, observing a portion of the slope determined by the field of view of the real antenna (Fig. 1a). The receiver can provide the amplitude and the phase of the microwave signal backscattered by the target. Range and cross-range synthesis of complex images are obtained by coherently summing signal contributions relative to different antenna positions and different microwave frequencies. The spatial resolution of the synthesized images in the described cases is about 2×2 m at 1-km range. The interferometric analysis of sequences of consecutive images allows us to derive the entire deformation field of the observed portion with millimeteric accuracy (Tarchi et al., 2003a, 2003b; Casagli et al. 2003). The apparatus used for the two campaigns was formerly developed by the Joint Research Center of the European Commission (JRC) and by Ellegi-LiSALab srl. Radar parameters as radiofrequency bandwidth, frequency step, mechanical step, and real antenna characteristics dictate the features of the final SAR images and interferometric maps such as the attainable spatial resolutions. The parameter values for the instrumentation used in the two measuring campaigns reported in this paper are resumed in Table 1. The center frequency determines the interferometric sensitivity, i.e., the minimum measurable displacement, usually largely smaller than the corresponding wavelength: in the described experimental campaigns, where the transceiver operates at Ku microwave band, a sub-millimetric sensitivity can be gained. According to the acquisition scheme depicted in Fig. 1b, the interferometric analysis of the resulting images time series provides the estimation of the LOS component of the velocity field. When LOS does not coincide with the direction of the displacement vector, there is an estimation error on the velocity due to the lateral viewing that may lead to an underestimation of the actual displacement. A monitoring case: Ruinon landslide (Valfurva, Italy) Test site description The first application of GB-InSAR technique deals with the monitoring of the Ruinon landslide, which is a 30×106 m3 active rock slide located on the North-East sector of the Valfurva valley (Fig. 2). The Valfurva valley is a NW–SE trending glacial valley deeply cut by the Frodolfo River (see map of Fig. 2b). This landslide represents a serious threat to human life and socioeconomic activities in the area. A potential risk from this landslide, which is suspended above the valley, is the possibility of a fast-moving rock avalanche affecting the important tourist road connecting Bormio to Santa Caterina Valfurva and the consequent development of a strongly unstable landslide dam on the Frodolfo River. The landslide that affects a steep slope made by schistose metapelites locally covered by slope debris (Agliardi et al. 2001;

by several authors (Agliardi et al. 2001; Crosta and Agliardi 2003; Tarchi et al., 2003a cum biblio). The Ruinon landslide occurs in pre-Permian metamorphic rocks belonging to the Austroalpine crystalline basement of the Campo Nappe, which is represented by the Bormio Phillites (Agliardi et al. 2001) and composed by a metapelitic unit. A schematic geological cross-section of the Ruinon landslide is showed in Fig. 3. The hydrological framework of the Ruinon landslide was carried out by Regional Environmental Agency (ARPA-Lombardia) in order to investigate the deep flowing. Results of this study testify a high permeability that causes a fast filtering (Dei Cas 2006). Since 1997 a permanent monitoring network, consisting of topographic instrumentation (total station and GPS receivers) and conventional in situ systems, such as inclinometer tubes and extensometers partially equipped for real-time data acquisition and transmission, has been installed by the ARPA. The first ground-based SAR interferometric measurement campaign about this landslide dates back to summer 2000 (Tarchi et al. 2003b; Antonello et al. 2004). In June 2006, a permanent GB-InSAR monitoring system was installed, on behalf of the local municipality and in cooperation with ARPA-Lombardia, for early warning purposes. With respect to the previous paper, which covered a very short time span (about 1 week), in this one, we present the results of one year acquisitions. The new data have permitted evaluation of the capability of the apparatus to detect the movement of the landslides’ surface and to integrate them with data coming from in situ instrumentations, allowing arrangement of early warning systems.

Fig. 1 a The GB-InSAR apparatus used in the Stromboli monitoring site. The radar moves along the rail (x axis) to perform synthetic aperture. b Scheme of the GBInSAR acquisition concept. A SAR power image projected on the Stromboli DEM is provided as an example

Tarchi et al., 2003a), reaches a total length of about 770 m and a width of about 410 m. The average gradient of the slope is about 35°. From a geomorphological point of view the landslide is characterized by two main scarps oriented northwest–southeast, parallel to the main fracture system recorded on the slope. The “Upper Scarp” is located at an elevation of about 2,100 m above sea level (asl), the “Lower Scarp” at about 1,900 m asl (Fig. 2c). Boreholes equipped with inclinometric probes locate main slip surfaces (27, 38, and 70 m from ground level; Dei Cas 2006). Geological aspects of Ruinon landslide were thoroughly analyzed

Data analysis In this section, some results of the GB-InSAR monitoring from June 2006 to June 2007, are briefly presented. Thanks to the high sampling rate of the GB-InSAR system, multitemporal maps showing the complete displacement field of the landslide have been generated and used to derive velocity–time diagrams for selected pixels. This kind of information is central to the early detection of changes in the landslide state of activity. Furthermore, the analysis of cumulative displacement maps in a quasicontinuous surface allows identification of landslide sectors characterized by different displacement patterns, possibly indicating partial activations or site-specific surface processes. In detail, the deformations are mainly focused on two sectors; the first one corresponds to the eastern part of the Upper Scarp, where the total estimated displacement along the line-of-sight for the entire period of monitoring (1 year between June 2006 and June 2007) is more than 630 mm, towards the sensor (negative values in the presented maps). The second one corresponds to the upper sector of the Lower Scarp where the total displacement along the LOS is more than 300 mm for the same period (Fig. 4). These sectors do not correspond to those indentified during a

Table 1 Radar parameter used into measurement campaigns

Test site

Central frequency (GHz)

Band width (MHz)

Synthetic aperture (m)

Max distance (m)

Theoretical range resolution (m)

Theoretical azimut resolution (m at 900 m)

Scanning time (min)

Ruinon

17–35

100

3–0

1800

1–5

2–6

14

Stromboli

17–45

100

2–8

1200

1–5

2–8

11

Landslides 7 & (2010)

293

Original Paper

Fig. 2 a–b Location of the Ruinon landslide. The Ruinon landslide developing in the Valfurva Valley on the hydrographic right of the Frodolfo river; c view of the monitored slope from the GB-InSAR installation site. The two main scarps of the landslide are indicated

previous GB-InSAR monitoring campaign carried out in 2000 (Tarchi et al., 2003a). In that campaign, the active sectors of the landslide were actually located in the south-eastern portion of the two scarps. The dynamics of the landslide can be better analyzed through the velocity vs. time plots retrieved from the pixels— selected as the most representative of different landslide sectors. The following four pixels have been selected (Fig. 4):

& & & &

LISA01 which belongs to the hydrographic left side of the Lower Scarp; LISA02 on the hydrographic right side of the Lower Scarp; LISA03 which refers to the hydrographic left side of Upper Scarp; and LISA04 on the hydrographic right side of Upper Scarp.

The selected points are chosen as close as possible to in situ sensors (mostly extensometers) location to improve the comparison between the two techniques. The cumulative displacement vs. time (Fig. 5a) and velocity vs. time plots (Fig. 5b) have shown a clear seasonality Fig. 3 Schematic geological crosssection of studied area

294

Landslides 7 & (2010)

with the maximum speed recorded in spring and the minimum recorded during the winter—a behavior influenced by snow melting. These plots demonstrate that the seasonal accelerating phase is generally rapid and followed by a slower deceleration phase. Furthermore, considering the mean velocity changes progressively over time, three different periods can be identified. In particular, from June to October 2006, the landslide moves, in its fastest sector (LISA03), at about 2.9 mm/day while from October 2006 to May 2007, the velocity, in the same sector, is about 1.1 mm/day (Figs. 5 and 6). From the middle of May, its velocity increases again after the snow melting. The results derived by GB-InSAR monitoring are in good agreement with the data obtained by in situ instrumentation (Dei Cas 2006) that show the same deformation trend (Casagli et al. 2008). The data recorded by the GB-InSAR show a good correlation between landslide acceleration peaks and rainfalls. The area is characterized by a typical continental-alpine rainfall regime (with rainy summer and autumn). Daily rainfall and cumulative rainfall are measured through the meteorological station operated by ARPA-Lombardia: acquired data are

shown in Fig. 7a, d in comparison with the velocity and the cumulated displacement time series. It is worth noting that, from the end of October 2006 to the middle of May 2007, a permanent snow cover was present on the landslide; in this period, the landslide shows lower velocities (less than 1 mm/day). The recorded acceleration peaks (see Fig. 7b–c) are mainly located after major rainfall events (Fig. 7a–d). Previously, other authors highlighted the strong relationship between the Ruinon landslide movements and precipitations (Crosta and Agliardi 2003). In this analysis, we define the single rainfall event as the occurrence of two o more rainy days separated by at least 2 days with rainfall