Rockfalls constitute a major hazard in most steep natural rock slopes. The growing number of residential buildings and transport infrastructure in mountain areas has progressively increased the exposure to such processes, making the development of reliable detection systems crucial for early warning and rapid response (Stähli et al., 2015). Local geological and geomorphological conditions are the main pre-disposing factors affecting the sizes of failing rock blocks, the falling dynamics, as well as the total runout distances (Corominas et al., 2017). Different triggering agents (mainly earthquakes and/or meteo-climatic variables) also have an impact on slope failure processes, which can range from a single block fall scenario to large and more complex rock avalanches. In addition, increase in rockfall activity has been observed in areas affected by large and deep-seated slope instabilities prior to catastrophic failure events (Rosser et al., 2007).

Accurate catalogues (including event location, time, and magnitude) are essential to understand and forecast rockfalls, as well as other landslide processes (Kirschbaum et al., 2010). Usual approaches to build catalogues are based on chronicles and observations of past events; however, catalogues may lack completeness, as the information is often qualitative and constrained to limited time windows and/or specific locations. This is especially true for small- to medium-size rockfall events (Paranunzio et al., 2016). For this reasons there is an increasing focus on more quantitative monitoring approaches, which can provide accurate and unbiased datasets.

As rockfall phenomena also induce seismic waves (Dammeier et al., 2011; Dietze et al., 2017), seismic instruments can be installed directly on the unstable rock face to catch precursory signs of rock failure (Arosio et al., 2009), or at relatively large distances to detect a rockfall event occurrence and its propagation (Manconi et al., 2016). In particular, seismic sensors present a significant number of advantages as they are (i) compact and relatively low-cost sensors, (ii) highly adaptable to difficult field conditions, and (iii) can provide reliable information in their flat-response frequency range on a broad spectrum of mass wasting processes occurring in relatively large areas (Burtin et al., 2014; Coviello et al., 2015; Vouillamoz et al., 2018). Consequently, in recent years the seismic signature of rock-slope failure phenomena has been investigated by several authors in different environments and monitoring set-ups (e.g. Helmstetter and Garambois, 2010; Zimmer and Sitar, 2015; Fuchs et al., 2018). The results have shown how it is possible to derive information to characterize rockfalls from the seismic signals, with different levels of accuracy depending on signal sampling rates, distances between the sensors, and the event, as well as the network density (Hibert et al., 2017; Provost et al., 2018). High-resolution, dense seismic networks are expensive to install and need resource-intense maintenance: one high-resolution seismic station costs in the order of tens of thousands of US dollars to build and equip, including sensors, on-site data acquisition systems, telecommunications, and back-up power. Thus, low-cost solutions are becoming more and more attractive to increase the capability of detection and investigation of seismic activity (Cochran, 2018). Moreover, low-procurement, as well as limited installation and maintenance efforts, is envisaged in case of the deployment of seismic networks including tens (or even hundreds) of sensors. In this scenario, a recently developed low-cost seismic sensor, i.e. the Raspberry Shake (RS) seismometers, provides an interesting low-cost plug-and-play solution. The RS devices have become more and more popular, mainly for home use, educational purposes, and outreach. However, their potential for seismic monitoring in challenging environmental conditions is still unexplored. In this work, we show the results of a 1-year pilot test performed in the Swiss Alps, deploying a network of three Raspberry Shake seismometers to monitor rock-slope failure events associated with a large, deep-seated slope instability. In the following sections, we provide a short technical description of the sensor, introduce the study area selected, and provide details on the performance of the Raspberry Shake.

The RS is an all-in-one plug-and-go solution for seismological applications. Developed by OSOP, S.A. in Panamá, the RS integrates geophone sensors, 24-bit digitizers, period-extension circuits, and computer into a single enclosure (see details in the Supplement). Currently, available RS versions (V6+) measure ground velocities with one (1-D, vertical component) or three (3-D, one vertical and two horizontal components) geophones (4.5 Hz Racothech RGI-20DX) and sampling rates are adaptable up to 100 Hz. Moreover, combination of geophones with other devices like MEMS and omnidirectional pressure sensors are also available. The power supply is 5 V (2.5 A supply) and consumption is estimated in 2.8 W at start-up time and 1.5 W during running time. Data are saved on a local SD card (default 8 Gb, but larger cards can be installed), and the estimated data amount per channel is below 10 Mb day-1 (∼2 years of local storage). Local storage can be thus adapted depending on the SD card mounted, the number of sensors available, and sampling rate selected. By default, time synchronization is based on NTP (Network Time Protocol); however, a GPS module can be connected via USB for situations where internet connection is not available. We refer the reader to the Supplement and to the web page https://raspberryshake.org (last access: 10 December 2018) for additional technical details on power consumption and communication issues. At the moment of our procurement (January 2017) only the RS-V4 was available on the market, and thus the results and performance assessments presented below refer to the 1-D version (vertical component 4.5 Hz geophone, with 50 Hz sampling rate). Recently, Anthony et al. (2018) provided the results of systematic lab and field tests to assess the performance of RS-4Ds and suggested that they are suitable to complement existing networks aimed at studying local and regional seismic events.

The Great Aletsch Glacier region (Swiss Alps, see Fig. 1) has undergone several cycles of glacial advancement and retreat, which have deeply affected the evolution of the surrounding landscape (Grämiger et al., 2017). In this region, the effects of the current climate change are striking, as the Aletsch glacier (blue shading in Fig. 1) is experiencing remarkable retreat with rates in the order of 50 m every year (Jouvet et al., 2011). In particular, a deep-seated slope instability located in the southern slope of the Aletsch valley, more specifically in the area called Moosfluh, has shown evidence of progressive increase in surface displacement during the past decades (Kos et al., 2016; Strozzi et al., 2010). In the late summer 2016, an unusual acceleration of the Moosfluh rockslide was observed, with maximum velocities locally reaching up to 1 m per day (Manconi et al., 2018). Such a critical evolution caused the generation of deep tensile cracks, and resulted in an increased number of rock failures at different locations of the landside body.

In this scenario, we have installed a local network composed of three RS V4 sensors. RS-1 (installed on 19 May 2017) and RS-2 (installed on 27 June 2017) are co-located within precedent monitoring infrastructure and exploit the necessary power from them (solar panels and batteries) and the internet connection (GSM) necessary for real-time data transmission (Loew et al., 2017). RS-3 (installed on 3 July 2017) is located in the basement of the Moosfluh cable-car station, and depends on existing power and Internet connection facilities. The coupling between the station and the ground is established through an aluminium plate (10 mm × 180 mm × 280 mm) screwed directly on the rock face by means of three M10 bolt anchors. The standard RS enclosure provided is made of plastic plates (5 mm thickness) and classified as IP10 (see directive IEC 60529, Edition 2.2, 2013, for IP coding). Due to the expected harsh conditions at our monitoring locations, especially during winter periods, we assembled the RS on a polycarbonate enclosure (180 mm×75 mm×180 mm, IP67, model PC 175/75 HG – https://www.distrelec.ch/, last access: 10 December 2018) to isolate the sensor and the electronic parts from direct effects of external agents (rain, snow, wind, dust, animals, see also details in the Supplement). IP67 enclosures are currently available to buy from the RS shop (not available at the time of our procurement). Data acquired from RS-1 and RS-2 are transmitted in real time to the ETH Zurich servers via cellular network through a mobile access router (AnyRover, see details at https://www.anyweb.ch/, 10 December 2018). Instead, the RS-3 data are stored locally and also forwarded (optional feature in the RS configuration) to a Winston Wave Server (Wave INformation STOrage Network, developed by the Alaska Volcano Observatory to replace the Earthworm Wave Server, data resides in an open source MySQL database).

In this work we show the performance of a network of three Raspberry Shake (RS) during a 1-year pilot project aimed at testing such low-cost seismic sensors (developed for home use) to study rockfall activity in alpine environments. Our results highlight that, despite installation on the rock surface and only moderate protection from the expected harsh environmental conditions, the RS seismometers provided continuous waveforms during the 1-year observation period, without any further intervention after the installation. Continuous seismic monitoring for rockfall detection is of high relevance in alpine areas, where the use of other instruments can be hindered due to environmental conditions, logistics, and/or high costs. We also show that low background-noise levels at our RS stations allowed for the detection of local, regional, and distant earthquakes, as well as large mass wasting at relatively large distances. Currently, visual interpretation of the waveform properties in time and frequency domains allowed us to discriminate between rockfall events associated with the evolution of the slope instability, e.g. rockfall phenomena of different size and runout, and seismic events, such as regional earthquakes and teleseisms. Future work is aimed at developing automatic detection and discrimination, as well as at attempting location of seismic signals due to rockfalls. During the design of this pilot study, we aimed at retrieving the number of rockfall events that occurred and use the event amplitudes and durations as a proxy to classify their size. However, as demonstrated in this work, the performance of the RS in alpine environment look better than expected, and the use of higher sampling rates, as well as 3-D ground velocity records instead of 1-D vertical components only, might further enhance the capacity of better describing rockfall events. We thus foresee that due to their good performance and low cost, RS will be more and more adopted in research studies.