Original Paper Landslides DOI 10.1007/s10346-016-0701-y Received: 23 December 2014 Accepted: 15 March 2016 © Springer-Verlag Berlin Heidelberg 2016 Simon Loew I Sophie Gschwind I Valentin Gischig I Alexandra Keller-Signer I Giorgio Valenti Monitoring and early warning of the 2012 Preonzo catastrophic rockslope failure Abstract In this paper, we describe the investigations and actions taken to reduce risk and prevent casualties from a catastrophic 210,000 m3 rockslope failure, which occurred near the village of Preonzo in the Swiss Alps on May 15, 2012. We describe the geological predisposition and displacement history before and during the accelerated creep stage as well as the development and operation of an efficient early warning system. The failure of May 15, 2012, occurred from a large and retrogressive instability in gneisses and amphibolites with a total volume of about 350,000 m3, which formed an alpine meadow 1250 m above the valley floor. About 140,000 m3 of unstable rock mass remained in place and might collapse partially or completely in the future. The instability showed clearly visible signs of movements along a tension crack since 1989 and accelerated creep with significant hydromechanical forcing since about 2006. Because the active rockslide at Preonzo threatened a large industrial facility and important transport routes located directly at the toe of the slope, an early warning system was installed in 2010. The thresholds for prealarm, general public alarm, and evacuation were derived from crack meter and total station monitoring data covering a period of about 10 years, supplemented with information from past failure events with similar predisposition. These thresholds were successfully applied to evacuate the industrial facility and to close important roads a few days before the catastrophic slope failure of May 15, 2012. The rock slope failure occurred in two events, exposing a compound rupture plane dipping 42° and generating deposits in the midslope portion with a travel angle of 39°. Three hours after the second rockslide, the fresh deposits became reactivated in a devastating debris avalanche that reached the foot of the slope but did not destroy any infrastructure. The final run-out distance of this combined rock collapse–debris avalanche corresponded to the predictions made in the year 2004. Keywords Rockslide . Early warning . Monitoring . Acceleration . Threshold . Evacuation Introduction In the Central European Alps, large rock slope failures with volume over one million m3 occur with an annual probability of approximately 3–5 % (Heim 1932). Such catastrophic landslides typically cause tens to hundreds of casualties per year (Schuster 1996; Sidle and Ochiai 2006), because many of these rock slope failures develop into very fast moving rock avalanches. In Switzerland, the frequency of intermediate size rock slope failures has been assessed by Dammeier et al. (2016) for the period 2000– 2014. They show that in the Swiss Alps, rock slope failures with volumes ranging between 50,000 and a few 100,000 m3 take place about once per year. Exact numbers cannot be given due to several undetected events in remote locations of the Alps. In the period 1972–2007, 32 persons lost their lives due to landslides in Switzerland (Hilker et al. 2009). However, most of these fatal landslides were shallow soil slides and 16 of these fatalities were caused by failure of a concrete protection wall in Gondo in the year 2000. This implies that fatalities caused by larger rock slope failures are very rare in Switzerland, which is mainly due to adequate long- and short-term risk mitigation measures. Not all large rock slope instabilities lead to catastrophic failure and rapid landslide motion—many remain slow, moving in a continuous or intermittent manner over long distances. This is why initial stages of rock slope investigations need to identify carefully the possible and most probable behavior type. Hungr and Evans (2004), for example, relate the failure behavior of large landslides to the predisposition, i.e., to the rock type (weak or strong) and structural control including lateral constraints. Hungr et al. (2005) update and refine this classification and relate ranges of movement velocity to kinematic mode. While rotational slides tend to have low to moderate velocities, translational slides and collapses develop extremely rapid (>5 m/s) motions after a period of slow and accelerating displacements. However, there are many exceptions from these rules and many Bstrong^ unstable rock masses contain narrow layers of weak rock such as localized shear zones or schist/phyllite bands. For this reason, many rock slope instabilities are monitored even if they never fail catastrophically. In addition to the understanding of geological predisposition and rockslide1 kinematics, hazard analysis of potentially catastrophic rockslides requires thorough understanding of the mechanisms driving slope movements until failure is inevitable (e.g., Loew et al. 2012). The immediate cause of a slope failure—or trigger—must be separated from the long-term mechanisms that weaken rock mass strength and promote failure—so-called preparatory or driving factors (e.g., Gunzburger et al. 2005). Typical examples of triggers are strong earthquakes (Mw > 6) or rain storms, while examples of preparatory factors include deep weathering or pore pressure cycling. However, there are many cases of catastrophic failure, for which a trigger could not be identified, such as the Randa failures of 1991 (Schindler et al. 1993). In this paper, we describe a unique example of a catastrophic slope failure at Preonzo in the Swiss Alps, where the predisposition, kinematic modes, driving factors, and failure scenarios have been carefully recorded over several years prior to the accelerating motion that ultimately led to slope collapse. The results of these investigations were used as basis to design and install a costeffective early warning system and to define alarm thresholds. The system was successfully operated, and appropriate mitigation actions were taken by the cantonal authorities at the right time before about 210,000 m3 of rock collapsed on May 15, 2012. The goal of this paper is to discuss the investigations related to hazard and risk reduction, their results, and the derived mitigation In this paper, the term rockslide is used for all kinds of rock slope instabilities independent of the actual kinematic mode, i.e., this term includes planar, circular and compound slides, topples, spreads, and collapses following the definition of Hungr et al. (2014). Landslides Original Paper actions. The paper describes the geological predisposition, landslide history, and expected failure scenarios at Preonzo (BGeological predisposition and landslide history^), the observed spatial displacement pattern and kinematics (BSpatial displacement monitoring^), the monitored displacement time series, preparatory factors and expected time to failure (BDisplacement time series, driving factors, and failure time prediction^), the design and operation of the early warning system and some key observations of the slope failure of May 15th, 2012 (BEarly warning thresholds and May 2012 slope collapse^). These investigations are not only presented to show a success story but also to demonstrate the purpose, outcome, and limitations of the applied methods and tools. The paper focuses on what was considered critical and financially feasible by the cantonal officials and their experts. Detailed scientific investigations of the monitoring data and underlying mechanisms are underway and are not the focus of this paper. Geological predisposition and landslide history The Preonzo slope failure of May 2012 is part of a series of retrogressive rockslides and rock avalanches (Fig. 1). The analysis of past failures also substantially contributes to the understanding of the geological predisposition, as well as future slope failure scenarios. The instability complex is situated above the village of Preonzo in the Rivera valley of southern Switzerland (Canton Ticino) and located at Alpe di Roscioro, 1520 m a.s.l. (Fig. 2). The detailed geological and hydrogeological conditions in this area are described by Loew et al. (2004), Willenberg et al. (2009), and Signer (2010). The geological model is primarily based on surface outcrop mapping and remote mapping with high-resolution aerial images. In the entire slope to the west of the Rivera valley, the foliated high-metamorphic crystalline rocks dip 20–25° into the 208° direction (nearly normal to the mean slope orientation). The main lithologies in the instability complex are banded amphibolite, Augen gneiss, and banded biotite gneiss below 1370 m (Fig. 3). The main systematic joints sets are parallel to foliation and compositional banding (S0, average dip direction/dip angle 208/21), and three steeply dipping sets (S1 090/77, S2 345/75, S3 260/85). Steeply dipping listric faults running parallel to S1 form local rockslide scarps. The most prominent historical failure from this instability complex occurred on February 22, 1702, after a period with increasing rock fall activity (noted already in 1697). It destroyed parts of the ancient village of Preonzo, including the church BOratorio Madonna della Cintura.^ The deposits from this rock avalanche are still visible at the foot of the slope, and amphibolite boulders found in the center of Rivera valley might be related to this event. On August 22, 1725, a large debris flow mobilized parts of the rockfall debris, destroyed several houses and resulted in 17 deaths. In 1989, local farmers noticed a new tension crack on the flat alpine meadow of Alpe di Roscioro 50 to 100 m behind the historical head scarp (Fig. 4a) forming the uphill boundary of a currently active rockslide with an estimated volume of about 350,000 m3 (Fig. 5). In May 2002, 150,000 m3 of rock failed from a ridge located to the southeast (outside) of the active rockslide following heavy rainfall. Finally, on May 9, 2010, again with notable antecedent rainfall, approximately 25,000 m3 of rock was released from the northern sector of the active rockslide. All these events formed an oversteepened cliff (55°–65°) from about 1200 to 1500 m a.s.l. (Fig. 4b), followed by a debris-covered slope (35°) and debris-flow channels down to the main Rivera valley at an elevation of about 250 m a.s.l. (Fig. 2). For many decades, a large industrial facility has been located at the foot of the slope and two important highways occupy the Rivera valley at distances of only 200–500 m from the foot of the slope (Figs. 1 and 2). As the new tension cracks on Alpe di Roscioro continued to grow laterally, the aperture increased since the early observations in 1989, and the failure type was expected to be catastrophic, the new rockslide was considered a large risk for the industry and roads, and detailed investigations started in the year 2002 (Willenberg et al. 2009). Spatial displacement monitoring The first goal of surface displacement monitoring was to delineate the active rockslide boundaries, to estimate rockslide failure volumes, and to infer possible kinematic mechanisms at depth. The main tension crack system on Alpe die Roscioro is composed of two fracture sets termed S1 and S2, with branching and progressive crack propagation in the NNW and SSE directions (Figs. 4a and 5). First manual displacement measurements along the tension cracks Fig. 1 Photographs of slope between Alpe di Roscioro and industrial area of Sgrussa at three different years Landslides Fig. 2 Map of the Preonzo 2012 rockslide release, transit, and deposition areas, including objects at risk (industrial area of Sgrussa, cantonal road, A2 highway). Location of Preonzo within Switzerland Fig. 3 Schematic geological section with prefailure and postfailure topography of Alpe di Roscioro were made in 1990, followed by the installation of five rod crack meters in 1999 (see extensometers EST label in Fig. 5). These linear displacement sensors are SLS190 hybrid track potentiometers with variable body length manufactured by Penny and Giles. They have a typical linearity of 0.1 %, a repeatability of less than 0.01 mm, and an operational temperature between −30 and +100 °C. These crack meters were anchored with flanges between rock walls of the main tension crack in direction normal to the crack walls (Fig. 6b, c). Based on field observations of fracture signatures on opposite sides of these S1 tension cracks, the local displacements are oriented normal to fracture strike and plunge with small angles 16°–18° toward the valley. Cumulative displacements before the failure of 2012 ranged between 60 cm in the south and 240 cm in the north. A first local and small-scale geodetic network (Local Network, Fig. 5) was installed on Alpe di Roscioro in 2003, with four reflectors on the unstable mass east of the main tension crack (points N1 to N4 in Fig. 5) and one stable reference point (N5) on the western side of the main tension crack. In May 2004, additional geodetic reflectors (not labeled in Fig. 5) were installed on the face of the old rock scarp and monitored twice a year from a stable location on the Alpe Cher (Fig. 2). In July 2010, a robotic total station (Leica TM30) was installed at the foot of the slope on a stable pillar (Figs. 2 and 6a), measuring the distance every 20 min to 14 reflectors covering the entire instability complex (this network is named Regional, R1–R10 in Fig. 5). Monitoring and data processing (distance along line-of site) was performed with the Leica GEOMOS system, and the results were automatically forwarded to an FTP server in the Forestry Department. The Leica TM30 is a high-accuracy total station with angular Landslides Original Paper Fig. 4 Photographs of instabilities at Alpe di Roscioro before (a, b) and after (c–f) May 2012 collapse. a Ortho-image with tension cracks in 2010. b Frontal view of instability complex as of 2010. c View into lateral ESE scarp with secondary rupture surfaces, brown amphibolitic rocks overlying white augengneiss. Red rectangle in b marks region of d with 2012 release plane. d Frontal view of 2012 release surface. e View downslope from headscarp with rupture place, deposits, industrial area, and roads. f View from WNW onto basal rupture plane measurement accuracy of 0.5″ (not used for early warning), prism distance measurement accuracy of 0.6 mm + 1 ppm, and a longrange automatic target recognition over 3000 m with 7 mm accuracy. Before the May 2012 failure, the absolute displacements of the active rockslide on Alpe di Roscioro showed relatively uniform displacement directions oriented normal to the main strike of the tension crack (55°–70° displacement direction), with geodetically derived plunge angles between 20° and 40° (blue arrow labels in Fig. 7). Displacement rates of the eight new reflectors located in the active rockslide area were systematically higher than the opening rates shown by crack meters, suggesting that other undetected Landslides active fractures might have existed downslope of the main tension crack. Between 2005 and 2007, eight differential ground-based radar interferometry (GB-SAR) campaigns were carried out with a synthetic aperture radar system (LiSALab srl, Tarchi et al. 2003) mounted on a stable foundation at the foot of the slope (range 1350–1570 m, line-of-sight azimuth S64°W). The resulting displacement field showed the greatest displacements at the top of the existing scarp, which then systematically decreased downslope until they became statistically insignificant at an elevation of about 1390 m a.s.l. (Fig. 7). In the summer of 2007, the highest displacement rates of 0.3 mm/day were measured in the northernmost Fig. 5 Monitoring locations, bedding orientation, previous failure scarps, and suggested boundary of the active rockslide area. Also visible is the ridge with strongly disintegrated materials in the SE of the instability complex (2002 release area with reflectors R7 and R10) sector, which failed in May 2010. The initial displacement velocities measured with the robotic total station in July/August 2010 are also plotted in Fig. 7, scaled for the same measurement duration of 57 days as the GB-SAR measurements from May/July 2007. Even though the measurements had been carried out in different years, the displacements measured along the identical line-of-sight (Fig. 2) are similar and supported the delineation of the active rockslide boundaries shown on Figs. 5 and 7. From the spatial displacement distribution and the local extensional dislocations along the tension cracks, it was concluded that the kinematics of the instability was likely controlled by compound sliding along a poorly defined sliding zone. Significant internal deformation contributes to the observed absolute surface displacements, and stronger downslope displacements in the north as compared to the south indicate a clockwise rotation. This kinematic model, combined with the geological predisposition, supported the initial assumption that future movements of the active rockslide would lead to catastrophic failure. The measured absolute displacement vectors and local relative displacements along the tension cracks, which suggested compound sliding, could not be supported by stereographic kinematic analysis of the systematic joint sets measured on Alpe di Roscioro. The volumes that may potentially be released from Alpe di Roscioro are a key parameter for run-out predictions, hazard, and risk analyses. Based on all displacement data and mapped landslide features, the lateral boundaries of the active rockslide were clearly visible in the west and north and delimited as shown on Figs. 5 and 7. On the other hand, the boundaries at the bottom (eastern boundary on Figs. 5 and 7) and in the south remained uncertain. In addition, the estimated volume of future failure scenarios depended on the uncertain basal rupture plane geometry and potential compartments of the active rockslide failing as individual events. Therefore, as of 2009, two failure scenarios of 260,000 and 680,000 m3 were explicitly considered in run out simulations (Willenberg et al. 2009). Displacement time series, driving factors, and failure time prediction A second objective of displacement monitoring was to investigate driving factors and potential failure time from high-resolution time series of displacements and environmental factors. Continuous in situ and remote displacement monitoring at Preonzo demonstrated increasing long-term average displacement velocities since the year 1999. As shown in Fig. 8, the main tension crack showed coherent opening rates, with a tendency for higher rates in the north (crack meters 3 and 5) compared to the south (crack meters 1 and 2) since 2010 and especially since 2011. This long-term velocity increase of the reflector displacements and crack openings was overprinted by short-term accelerations Landslides Original Paper Fig. 6 Total stations installed on concrete pillar in air-conditioned cabin at slope toe near industrial area of Sgrussa (a). Crack meters (SLS190 hybrid track potentiometers) installed in main tension crack on Alpe di Roscioro (b), anchored with flanges between rock walls of the main tension crack in direction normal to the crack walls (c) triggered by heavy rainfall (Figs. 8 and 9). These short-term accelerations returned to nearly steady displacement rates after a few weeks. However, in this period, every strong rainfall even brought the quasi-steady rock mass movements to a higher average velocity. In the period 2002–2011, the 10-day averaged baseline velocities increased from 0.01 to 0.1 mm/day (Fig. 10). The local failure of May 9, 2010, occurred after 9 days of rain (about 200 mm) and a synchronous acceleration of all measured crack openings. At the daily scale, short-term accelerations coincided directly with the period of intense rainfall. This indicated that local rainfall (and snowmelt) infiltrated directly into the open Landslides fracture system at Alpe di Roscioro, increasing the level of a phreatic groundwater table and pore pressures within the unstable rock mass. These observations suggested that hydromechanical loading was a main driving factor of the displacement accelerations, similar to many other hard-rock slope instabilities (e.g., Gröneng et al. 2011; Crosta et al. 2014). Several researchers have developed empirical methods to predict the time of failure of rock slopes from displacement records. One of the most popular approaches has been initially described by Fukuzono (1985) and Voight (1988), which is based on an inverse displacement rate plot and the assumption that rock slopes show a stage of Baccelerated creep^ prior to failure. The method suggested by Voight (1988) assumes that the displacement acceleration can be described by a power law function of the displacement velocity. The constants of this power law are assumed to be dependent on time invariant material properties, such as stress corrosion index, and external loading conditions, such as pore pressure or temperature (Cornelius and Scott 1993). For real rockslides like Preonzo, the trend of the computed inverse displacement rate versus time is very complex, because short-term rainstorms or snow melt periods—correspond to variations in external loading conditions—generate strong short-term fluctuations of displacement velocity (Figs. 8 and 9). These shortterm fluctuations create strong variations in displacement velocity or inverse displacement velocity as illustrated for crack meter 1 in Fig. 10. This figure was created from daily crack opening rates, calculating the moving velocity over time intervals of 10, 100, and 500 days. It can be seen from this figure that velocities and inverse velocities calculated over short and long time intervals show very large variations and are not directly usable for failure time prediction. Also, velocity averaging or ignoring extreme values does not lead to improvements in longer term failure time prediction. On the other hand, the long term trend of all crack opening velocities shows deceleration (inverse velocity acceleration) until the beginning of 2006 followed by systematically increasing crack opening rates (visible in a systematic inverse velocity decrease) until the end of 2009. Therefore, the beginning of 2006 corresponds to a transition from deceleration (presumably since the failure of 2002) to accelerated creep that ended with the catastrophic failure of May 2012. Usually, the adverse impact of short-term fluctuations related to rain storms and snow melt on failure forecasts is reduced by fitting a model through the raw (or filtered) data (Crosta and Agliardi 2003). Thus, characteristic velocity curves for the rock mass are obtained. The most pragmatic and most commonly applied approach is to fit a linear function to the inverse velocity time series. Sättele et al. (2015) also observed seasonal scattering for the Preonzo displacement data and used a least-squares fit to a simple inverse velocity model representing the accelerated creep stage. In this study, we investigate an alternative model to fit longer term records of observed displacements time series following the three-phase creep curve often observed for landslide displacement behavior (Crosta and Agliardi 2003). It is the S-shaped curve proposed by Xiao et al. (2009) to describe damage behavior of laboratory samples during fatigue tests. This curve is characterized by an initial deceleration phase, a phase of constant rate and an acceleration phase, and was initially proposed for landslide applications by Gischig (2011). The equation adapted from Xiao et al. (2009) is Fig. 7 Spatial displacement rate distributions along line-of-sight nearly parallel to real displacement vectors. Shown are radar displacements from the interferogram between May and July 2007 (57 days) and geodetic reflector displacements from July 14 to August 13, 2010, upscaled for 57 days Δd ¼ α β −1 β−Δt p1 Δd = d(t) − d0 is the displacement change with respect to the initial reading d0 at time t0. Note that the number of cycles n used by Xiao et al. (2009) is here replaced by the time Δt = t − t0 passed since the time t0 of the initial data reading. α, β, and p are constants that have to be fitted to observations of d(t). The reasoning behind using this curve are the following: (1) its shape resembling the three-stage creep curve, and (2) similarly as shown in the laboratory experiment by Xiao et al. (2009), cyclic load changes (from annual temperature or pore pressure variations) may also be the major cause for long-term rock mass damage and rockslide displacements (Gischig et al. 2011a, b). Ideally, the curve is fit to a displacement time series that contains all three creep phases. Figure 11a shows the relative displacement of crack meter 1 in the period 2002–2011. The aforementioned curve was fitted to the data using a nested grid-search approach minimizing the root-mean-square (RMS) value of the difference between the observed and calculated displacements. The best-fit parameters were α = 213.1 mm, β = 12.3 years, and p = 1.36, and the corresponding RMS = 0.26 mm. Also included in Fig. 11a is a bundle of curves that fit the observations with and RMS 10 % higher than best-fit one (i.e., curves with 2.6 < RMS < 2.86) to illustrate the uncertainty of the curvefitting. The corresponding parameters range α = 191–264 mm, β = 11.45–13.9 years, and p = 1.15–1.56 and depend on each other as shown in the inset figure. As can be deduced from the above equation, the parameter β determines the time when Δd tends to infinity and thus the last possible time when catastrophic failure would occur. Based on our best-fit values of β, the time of failure would correspond to September 2014 (ranges November 2013–April 2016). Figure 11b shows displacement velocities from daily mean displacements averaged over 10 days together with the same best-fit curves translated into velocities. The same approach was applied to displacement data from reflector R4 (with α = 1583 mm, β = 5.73 years, p = 0.85) located in the northern part of the instability complex (Fig. 12). Comparing Figs. 11 and 12 shows significant differences in displacements and velocities for the overlapping time window since July 2010. For example, the accelerations of fall 2011 are much more pronounced for reflector 4 (in the north, 5 mm/day) as compared to crack meter 1 (in the south, 1 mm/day). Assuming that the failure time is close to the date when daily displacements are above 1 cm (e.g., Crosta and Agliardi 2003; Blikra 2008), the extrapolations of Figs. 11 and 12 would suggest a possible failure time between spring 2012 (mainly reflector R4) and 2015 (mainly crack meter 1). Landslides Original Paper Fig. 8 Long-term opening displacements of main tension crack at Alpe di Roscioro between May 2002 and May 2012 as derived from five crack meters. Also shown are daily precipitation since 2008 and total apertures normal to opening direction as measured in April 2012 Figure 13 shows the same crack aperture data fitted to the creep curve of Xiao et al. (2009) and used as basis for inverse velocity extrapolation. Comparing Figs. 10 and 13 shows that fitting the data to such a mathematical function leads to an improved long-term extrapolation power. However, because the selected three-stage creep model approaches infinite velocities asymptotically, the predicted failure time can be delayed with respect to the actual time of the slope failure. In addition, due to smoothing, a short-term triggered failure cannot be predicted with such a function. Fig. 9 Daily rainfall and geodetic reflector displacements along line-of-sight between July 2010 and May 2012 Landslides 1 1000 0.9 Velocity 10d 900 0.8 Velocity 100d 800 0.7 Velocity 500d 700 0.6 Aperture 600 0.5 500 0.4 400 0.3 300 0.2 200 0.1 100 0 Crack Opening Inverse Velocity EST (day/mm) b Crack Aperture (mm) Crack Opening Velocity EST 1 (mm/day) a 0 40 35 Inverse Velocity 10d 30 Inverse Velocity 100d 25 Inverse Velocity 500d 20 15 10 5 0 Fig. 10 Observed displacements, calculated velocity (a), and inverse velocity (b) of crack meter 1. Velocity calculated from moving 10-, 100-, and 500-day observation periods Early warning thresholds and May 2012 slope collapse Based on hazard assessment as described in Loew et al. (2004) and Willenberg et al. (2009), the cantonal officials decided in 2010 to implement precautionary measures, including a disaster management plan and an early warning system to manage the risks of a larger catastrophic rock slope failure with a volume of several 100,000 m3. The early warning system was based on the five crack meters and one rain gauge on Alpe di Roscioro and the robotic total station measuring 19 reflectors from a stable point at the foot of the slope. Both sensor types measured displacements in intervals of 20 min and had continuous data transmission and processing by the software called OASI that integrates multidomain information systems and was originally developed for data collection, management, and analysis of noise-, air-, and traffic-related data (Andretta et al. 2004). The definition of alarm levels was based on the velocities and short-term accelerations observed since 2002 and compared with other failed rockslides in similar geological setting (e.g., Schindler et al. 1993; Krähenbühl 2006). Due to the limited duration of rainfall-triggered accelerations (inset Figs. 8 and 9), the 10-day averaged velocities of acceleration periods are smaller than the hourly velocities. The strongest observed accelerations between 2002 and 2011 reached values of 4–5 mm/day when averaged over 10 days (Figs. 11b and 12b) and of up to 0.3 mm/h when averaged over 1 h. Due to the smaller absolute displacements, the rainstorminduced velocities of crack meter 1 and 2 (in the south) amount to only about 50 % of those for crack meters 3, 4, and 5 (in the north). Alarm levels were specified for prealarm, official alarm, and evacuation alarm released by the cantonal officers. In addition, a threshold value for automatic alarms of the crack meters was set up. These alarms were only sent by SMS to selected landslide experts. The alarm levels range between 1 mm/day and 5 mm/h and are summarized in Table 1. Prealarms include velocities observed during heavy rainstorms not leading to catastrophic rock failures. In order to avoid false official alarms due to short-term rainfall triggers, these thresholds are above previously recorded values. During April 2012, displacements of all crack meters and most reflectors started to deviate clearly from the modeled creep curves (Figs. 11, 12, and 13). The onset of this deviation initially resembled the acceleration of May 2010 leading to a smaller collapse of 25,000 m 3. Therefore, a prealarm was released on April 24. Alarms, daily rainfall, and hourly displacement velocities between May 1 and 15, 2012, are shown in Fig. 14. On Thursday, May 3, when Landslides Original Paper Fig. 11 a Displacement curve for crack meter 1 positioned in southern part of main tension crack (bold black line), fitted with a three-stage creep curve (red line) of Xiao et al. (2009) and a range of model curves for 10 % variation around the best fit parameter set (see text for details). b Displacement rates from daily mean displacements averaged over 10 days (bold black line) and extrapolated velocity models from fitted creep curve Fig. 12 a Displacement curve for geodetic reflector R4 located in the northern part of the active rockslide about 10 m from failure scarp (bold black line), fitted with a three-stage creep curve (red line) of Xiao et al. (2009) and a range of model curves for 10 % variation around the best fit parameter set (see text for details). b Displacement rates from daily mean displacements averaged over 10 days (bold black line) and extrapolated velocity models from fitted creep curve Landslides 1200 16 Fi ed Creep Model 1000 Modeled Velocity Crack Opening (mm) Modeled Inverse Velocity 14 12 800 10 8 600 6 400 4 200 2 Velocity and Inverse Velocity (mm/day) Measurements EST 1 0 0 Fig. 13 Measured displacements of crack meter 1, fitted creep curve (Xiao et al. 2009) at the end of 2011, velocity, and inverse velocity calculated from fitted creep curve and example estimation of failure time based on inverse velocity extrapolation, using displacement data until end of 2011 the hourly crack aperture velocities increased for the first time above 1 mm/h, the cantonal officials decided to evacuate the industrial facilities at Sgrussa, the foot of the slope. The following significant rainfall period of May 5–6 accelerated the rockslide even stronger and the hourly velocities reached higher peaks of up to 6 mm/h, and mean values of about 2.5 mm/h after termination of the rainy days (Fig. 14). From Monday, May 7, to Friday, May 11, workers in the industrial facilities were allowed to resume their work. Between May 10 and 11, the slope displacement velocities started to accelerate without any precipitation (Fig. 14). This change in behavior—as compared to previous short-term rainfall triggered accelerations—was considered very critical. Therefore, a public alarm was released and the cantonal road was closed on Saturday, May 12. On Monday, May 14, the transnational A2 highway was closed and finally the slope collapsed in the early morning of May 15, 2012. The slope collapse took place in two events at 2:45 am and 4:32 am. The first event could not be directly observed because of darkness, and the second event was recorded both by seismic stations and direct observations. At 07:31 am video cameras captured a large secondary debris avalanche from the rockslide deposits, which does not document the primary slope failures. The primary deposits from the failure only reached to the midslope elevation of about 750 m a.s.l., not the protection berm at 300 m a.s.l. as predicted for a similar size event in 2009 (Willenberg et al. 2009). The deposits filled the flatter part of the slope above Sgrussa (Fig. 2), resulting in a travel angle of 39°. The secondary debris avalanche destroyed the forest in the lower slope section and—with a travel angle of 36°—nearly reached the industrial area as predicted in 2004 (Loew et al. 2004) and published in 2009 (Figs. 1, 2, and 4e). The failure destroyed a large number of reflectors and crack meters in the newly exposed release area (Fig. 4c–f). The decisions to reopen the cantonal and national roads and industrial facility were supported by visual inspections and GB-InSAR measurements. The exposed failure surface was subsequently mapped by Lidar and photogrammetry. In NE-SW sections, the basal rupture shows a planar or compound geometry with a mean dip angle of 42° (Figs. 3 and 4f) and daylights between 1390 and 1420 m a.s.l. (Fig. 15). This implies that the base of the failed mass was located less deep and was smaller than initially assumed from GB-InSAR measurements (compare with Fig. 7). The upper boundary is formed by a 20 m deep subvertical tension crack (Fig. 3), the lateral boundaries of the release surface are formed by scarps (Fig. 4c, d, f). From the entire active rockslide as defined by prefailure displacements, the May 2012 rockslide incorporated about 210,000 m3, and about 140,000 m3 remained in place. A detailed analysis of the underlying mechanisms controlling the accelerated creep stage since 2006 and the final collapse of May 2012 is underway and will be published in a subsequent paper. Potential key mechanisms controlling this behavior could be related to pore pressure modulated nonlinear propagation of critically stressed crack tips through initially intact rock bridges, and rate- and state-dependent friction along preexisting discontinuities within the instability. Table 1 Definition of alarm levels as of December 2010 Alarm level Displacement velocity Explanation Prealarm 1–3 mm/day Police and community officials are informed. Automatic alarm 2 mm/h Crack meters automatically send SMS to scientists involved in decision making when a single measurement differs by more than 2 mm/h from previous measurement. Official alarm 3–5 mm/h Media and the public are informed. Evacuation 5 mm/h Evacuation of industry at slope toe and closure of cantonal and local roads Landslides Original Paper 6 20 Pluviometer Extensometer 2 Extensometer 4 Extensometer 1 Extensometer 3 Extensometer 5 Highway Closure May 14th 5 15 Pre-Alarm April 24th Occupancy Sgrussa May 7th 10 Evacua on Sgrussa May 3rd Public Alarm and Evacua on May 12th 4 3 2 Precipita on [mm/h] Crack Aperture Velocity [mm/h] a 5 1 0 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 8.5 9.0 9.5 10.0 10.5 11.0 11.5 12.0 12.5 13.0 13.5 14.0 0 Days since May 1st 2012 10 8 6 Pluviometer Extensometer 2 Extensometer 4 Extensometer 1 Extensometer 3 Extensometer 5 5 4 6 3 4 2 2 Precipita on [mm/h] Crack Aperture Inverse Velocity [h/mm] b 1 0 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 8.5 9.0 9.5 10.0 10.5 11.0 11.5 12.0 12.5 13.0 13.5 14.0 0 Days since May 1st 2012 Fig. 14 Hourly displacement velocities (a) and inverse velocities (b) of all crack meters, hourly rainfall and decisions taken by cantonal officials. Days since May 1, 2012 Discussion and conclusions Retrogressive rockslope movements at Alpe di Roscioro above the village of Preonzo in southern Switzerland started to become visible in the late 1980s. Continuous measurements across an open tension crack, which was located about 50–100 m behind a historic scarp, have been carried out since 1999. The opening velocity of this tension crack varied along strike and showed clear signs of acceleration since 2006 and a strong dependence and temporal correlation with major rainstorms events, facilitated by the convergent slope morphology prior to failure. Field investigations lead to the conclusion that the active rockslide had a volume of several 100,000 m3 which was expected to develop into a catastrophic failure. The main reasons for this assumed behavior were the brittle nature of the foliated crystalline rocks dipping into the slope and an assumed—but not kinematically verified—compound sliding mechanism. The potentially released volume could not be exactly determined and the run-outs for two scenarios (260,000 and 680,000 m3) were simulated with DAN 2D and 3D several years before the failure (Loew et al. 2004; Willenberg et al. 2009). The run-out predicted for the smaller volume was expected to reach the Landslides industrial facility and cantonal road located at the foot of the slope. Therefore, more detailed investigations of the unstable volume were carried out including eight ground-based radar interferometric campaigns with a synthetic-aperture radar. Periodically, repeated inverse velocity extrapolations from observed displacement records fitted with a three-stage creep curve suggested a failure time between Spring 2012 and 2015 (as of November 2011). A more precise prediction of failure time was only possible since the beginning of April 2012, when the slope movements started to deviate significantly from the past creep behavior. An early warning system was installed in 2010, based on five crack meters distributed along the main tension crack and a robotic total station with 19 reflectors installed at the foot of the slope. The purchase and installation costs of the system installed at Preonzo are in the order of 100,000 Swiss Francs (or US$). In comparison to other early warning systems installed at similar high-risk locations in developed countries (e.g., Blikra 2008; Michoud et al. 2013), the system installed at Preonzo has only few surface-based displacement monitoring sensors. Other highrisk sites reviewed in Michoud et al. (2013) include borehole-based highway was closed. One day later, a large part (210,000 m3) of the active rockslide collapsed in two events at 2:45 and 4:32 am. At 7:31 am, the deposit of these failures, which was accumulated in a part of the slope with reduced inclination, became remobilized as a debris avalanche reaching the foot of the slope. This final run-out distance matches well with the predictions made in 2004 for a scenario of similar volume. In summary, the detailed monitoring of displacements since 1999 and the corresponding analysis of the active rockslide predisposition, volume, and kinematics, as well as the forcing factors allowed to assess correctly the catastrophic failure behavior, and to design an early warning system with adequate alarm thresholds. This costeffective system turned out to be reliable and effective for risk prevention at one of the high-risk rock slope instabilities in the Alps. Acknowledgments We would like to thank M. Franzi (Dipartimento del Territorio/ Sezione Forestale Ticino) for providing data from Preonzo and many fruitful discussions. Several colleagues from ETH Zurich (H. Willenberg, M. Ziegler, R. Seifert, A. Kos) and UBC Vancouver (E. Eberhardt, O. Hungr) provided important contributions to the studies presented in this paper. We would like to acknowledge special support of M. Ziegler for postfailure photogrammetric analysis. We also thank three anonymous reviewers for their detailed comments. References Fig. 15 Difference in ground elevation between prefailure and postfailure topography of May 2012 rockslide event displacement and pore pressure sensors, continuous radar interferometric investigations, and substantially more surface sensors and sensor types. Even though the system installed at Preonzo only included the most important sensors, the early warning system proved to be effective and reliable (see also discussion in Sättele et al. 2015). The alarm thresholds were derived from a detailed analysis of 10 years of displacement and climatic monitoring data, observed short-term accelerations during rainstorms, and rock slope failures with similar geological predisposition and kinematic mode. A prealarm was issued on April 24, 2012, when the rockslide displacements started to deviate clearly from the accelerated creep behavior observed since 2006. On May 3, when the hourly displacement rates of crack meters increased for the first time above 1 mm/h and rainfall was predicted for the subsequent days, the cantonal officers decided to evacuate the industrial area at the foot of the slope, even though the thresholds for evacuation, as defined in summer 2010 with 5 mm/h, were not reached. After the heavy rainfall period, which triggered the previously known accelerations, the work in the industrial area was resumed on May 7. On May 12, when the tension crack aperture velocities reached several millimeters per hour without any rainfall trigger—suggesting a fundamental change in rockslide behavior—a general public alarm was released and the industry and cantonal road were evacuated again. On May 14, when crack aperture velocities reached more than 10 mm per hour, the nearby transnational A2 Andretta M, Bernasconi G, Corti G, Mastropietro R (2004) OASI: an integrated multidomain information system. Proc. International Conference on Complexity and Integrated Resources Management. International Environmental Modelling and Software Society iEMSs, Osnabrück, Germany, p 6 Blikra L.H (2008) The Aknes rockslide: monitoring, threshold values and early-warning. In: Chen, Z. et al. (Eds): Landslides and engineered slopes: from the past to the future. 10th International Symposium on Landslides and Engineered Slopes. Xian, China, 1 & 2; 1089–1094 Cornelius RR, Scott PA (1993) A materials failure relation of accelerating creep as empirical description of damage accumulation. Rock Mech Rock Eng 26:233–252 Crosta GB, Agliardi F (2003) Failure forecast for large rock slides by surface displacement measurements. Can Geotech J 40:176–191 Crosta GB, di Prisco C, Frattini P, Frigerio G, Castellanza R, Agliardi F (2014) Chasing a complete understanding of the triggering mechanisms of a large rapidly evolving rockslide. Landslides 11:747–764 Dammeier, F, Moore JR, Hammer C, Haslinger F, Loew S (2016) Automatic detection of alpine rockslides in continuous seismic data using Hidden Markov Models. J Geophys Res Earth Surface 121: 351–371 Fukuzono, T. 1985. A new method for predicting the failure time of a slope. In Proceedings of the 4th International Conference and Field Workshop on Landslides, Tokyo, Tokyo Universit Press, p 145–150 Gischig V (2011) Kinematics and failure mechanisms of the Randa rock slope instability (Switzerland). PhD Thesis ETH Zurich No. 19730: p 204 Gischig V, Moore JR, Evans KF, Amann F, Loew S (2011a) Thermo-mechanical forcing of deep rock slope deformation—part I: conceptual study of a simplified slope. J Geophys Res 116(F04010):18 Gischig V, Moore JR, Evans KF, Amann F, Loew S (2011b) Thermo-mechanical forcing of deep rock slope deformation—part II: the Randa rock slope instability. J Geophys Res 116(F04011):17 Gröneng G, Christiansen HH, Nilsen B, Blikra LH (2011) Meteorological effects on seasonal displacements of the Åknes rockslide, western Norway. Landslides 8:1–15 Gunzburger Y, Merrien-Soukatchoff V, Guglielmi Y (2005) Influence fo daily surface temperature fluctuations on rock slope stability: case study of the Rocher de Valabres slope (France). Int J Rock Mech Min Sci 42:331–349 Heim A (1932) Bergsturz und Menschenleben. Fretz and Wasmuth Verlag, Zurich, p 218 Hilker N, Badoux A, Hegg C (2009) The Swiss flood and landslide damage database 1972–2007. Nat Hazards Earth Syst Sci 9:913–925 Landslides Original Paper Hungr O, Evans SG (2004) The occurrence and classification of massive rock slope failure. Felsbau 22(2):16–23 Hungr O, Corominas J, Eberhardt E (2005) Estimating landslide motion mechanisms, travel distance and velocity. In: Hungr, Fell, Couture and Eberhardt (eds). Landslide Risk Management. Proc. of an Int. Conference, Vancouver. Taylor and Francis Group, London Hungr O, Leroueil S, Picarelli L (2014) Varnes classification of landslide types, an update. Landslides 11:167–194 Krähenbühl R (2006) Der Felssturz, der sich auf die Stunde genau ankündigte. Bull für angewandte Geologie 11(1):49–63 Loew S, Willenberg H, and Eberhardt E (2004) Hanginstabilität und Gefahrenanalyse der Alpe di Roscioro (Preonzo, TI). Geologischer Bericht Nr.: ETH 3465/45, ETH Zurich, Geologisches Institut. Professur für Ingenieurgeologie Loew S, Gischig V, Moore J, Keller-Signer A (2012) Monitoring of potentially catastrophic rockslides. In: Eberhardt, E., Froese, C., Turner, K. and Leroueil, S. (eds). Landslides and engineered slopes: protecting society through improved understanding. Proceedings of the 11th International and 2nd North American Symposium on Landslides and Engineered Slopes, Banff, Canada, Taylor and Francis Group, London: 101–116 Michoud C, Bazin S, Blikra LH, Derron MH, Jaboyedoff M (2013) Experiences from sitespecific landslide early warning systems. Nat Hazards Earth Syst Sci 13:2659–2673 Sättele M, Krautblatter M, Bründl M, Straub D (2015) Forecasting rock slope failure: how reliable and effective are warning systems? Landslides. doi:10.1007/s10346-015-0605-2 Schindler C, Cuénod Y, Eisenlohr T, Joris CL (1993) Die Ereignisse vom 18. April und 9. Mai 1991 bei Randa (VS): ein atypischer Bergsturz in Raten. Eclogae Geol Helv 86(3):643–665 Schuster RL (1996) Socioeconomic significance of landslides. In: Turner AK, Schuster RL (eds) Landslides: investigation and mitigation. National Academy Press, Washington, pp 12–35 Landslides Sidle RC, Ochiai H (2006) Landslides: processes, prediction, and land use. Water resources monograph 18. Amer. Geophys. Union, Washington, p 312 Signer A (2010) Structural and kinematic analysis of the Roscioro rock slide, Preonzo. Switzerland. MSc Thesis ETH Zurich, Zurich, p 92 Tarchi D, Casagli N, Moretti S, Leva D, Sieber AJ (2003) Monitoring landslide displacements by using ground-based synthetic aperture radar interferometry: application to the Ruinon landslide in the Italian Alps. J Geophys Res 108(B8):2387 Voight B (1988) A method for prediction of volcanic eruption. Nature (London) 332:125–130 Willenberg H, Eberhardt E, Loew S, McDougall S, Hungr O (2009) Hazard assessment and runout analysis for an unstable rock slope above an industrial site in the Rivera valley, Switzerland. Landslides 6:111–116 Xiao J-Q, Ding D-X, Xu G, Jiang F-L (2009) Inverted S-shaped model for nonlinear fatigue in rock. Int J Rock Mech Min Sci 46:643–548 S. Loew ()) : S. Gschwind : V. Gischig Department of Earth Sciences, ETH Zurich, 8092, Zurich, Switzerland e-mail: simon.loew@erdw.ethz.ch A. Keller-Signer ETH Zurich, now at GEOTEST AG, Promenade 15, 7270, Davos-Platz, Switzerland G. Valenti Dipartimento del Territorio, Sezione Forestale Ticino, Bellinzona, 6500, Switzerland