A Study on the Factors Controlling the Kinematics of a Reactivated and Slow-Moving Landslide in the Eastern Liguria Region (NW Italy) through the Integration of Automatic Geotechnical Sensors

Abstract

This paper deals with the investigation of factors influencing the movement patterns of a reactivated slow-moving landslide situated in the eastern Liguria region (NW Italy) through the analysis of extensive ground-based hydrological and geotechnical monitoring data. Subsurface horizontal displacement and pore water pressure data were acquired simultaneously by means of automatic sensors positioned at pre-existing and localized failure zones. The joint examination of field measurements enabled us to explore the connections between rain, pore water pressure, and displacements. The results of continuous displacement monitoring showed that the landslide kinematics involved phases of extremely slow movements alternated with periods of relative inactivity. Both stages occurred prevalently at seasonal scale displaying similar durations. The slow-motion phases took place at relatively constant pore water pressure and were ascribed to mechanisms of viscous shear displacements along failure surfaces. Inactive phases entailed no significant deformations, mostly corresponding to prolonged dry periods. The two motion patterns were interrupted by episodic sharp deformations triggered by delayed (preparation periods from 4 to 11 days) rainfall-induced pore water pressure peaks, which were ascribed to sliding mechanisms taking place through rigid-plastic frictional behaviour. During these deformation events, hysteresis relationships between pore water pressure and displacement were found, revealing far more complex hydro-mechanical behaviour.

1. Introduction

Slow-moving landslides are widespread in many geological and geomorphological environments, affecting a wide range of slope-forming materials and developing through several types of movement (e.g., slide, spread, flow), often with various styles of motion (e.g., complex, composite or multiple) [1]. This group of slope instabilities typically consists of soil/rock volumes moving along single or multiple pre-existing zones of weakness (i.e., reactivation stage sensu [2]), showing displacement rates ranging from some mm/yr up to several m/yr [3,4]. Furthermore, their motion can be markedly heterogeneous, showing the alternation of multiple stages of movement (e.g., acceleration and deceleration phases, rest periods), following both regular and irregular trends over different time scales (e.g., monthly, yearly, or seasonal) [5,6,7,8,9,10,11,12,13]. Nevertheless, these ground instabilities are often the source of serious social and economic impacts due to their effects in terms of damage to the built environment [14,15].

The mobility of reactivated and slow-moving landslides can be controlled by predisposing (i.e., time-invariant conditions), preparatory (i.e., time-dependent processes exhibiting cyclical changes or trends), and triggering (i.e., transitory processes developing over very short time periods) factors [1,16,17]. As known from both the theoretical and experimental mechanical behaviour of soils and rocks, rainfall-induced pore water pressure changes play a key role in conditioning normal effective stresses along existing sliding surfaces, consequently preparing for the onset of slope instability mechanisms [18,19]. The kinematics response of reactivated slow-moving landslides to precipitations can be time-delayed; namely, acceleration phases are most observed after several days and even months [9,20]. Both the timing and magnitude of hydrologically driven pore water pressure variations at depth depend on the extent of rain or snowmelt water infiltration from the ground surface [21,22,23,24]. In large landslides, typically showing complex hydrogeological settings, other conditions can modify pore water pressure regimes, such as groundwater circulation through different aquifers or the origination of local excess pore pressures due to the spatial variability of hydraulic conductivity across the landslide-forming materials [25,26,27,28,29,30,31,32]. More rapid groundwater recharges can be related to preferential flow circuits developing through highly fissured moving landslide masses [33,34].

Despite the close connections between pore water pressure changes and the development of movements in slow landslides, several studies revealed the inherent complex nature of both the mechanical and hydrological processes governing their kinematics [1,5,9,10,35,36,37]. In some cases, non-linear relationships between the piezometric surface oscillation and the movement rate were documented [33,38], while in others, different motion rates were observed depending on whether pore water pressure increases or decreases, revealing a hysteretic behaviour [10,23,25,36]. In this regard, several geomechanical models describing the stress–strain–time relationships in geomaterials (e.g., viscous-plastic, rate strengthening, and softening behaviours) have been used to interpret the spectrum of motion characteristics observed in the field [39,40,41].

The displacement trends in slow landslides that are driven by pore water pressure changes can be investigated using different approaches, such as numerical modelling [39,41], laboratory experiments [37,42,43], geotechnical field monitoring [7,10,33,36,44], and remotely sensed data analysis [9,45,46]. In the framework of field-based measurements, surface displacement data [10,36,47] or techniques that do not always imply a direct connection between measured displacements and landslide movements, such as wire extensometers [48,49,50], are often used. In contrast, the employment of displacements directly and continuously measured at localized sliding zones can be relatively challenging, often due to technical or economic constraints. In this regard, the exploitation of automatic inclinometers can be very useful in gaining knowledge about the preparatory factors controlling the temporal trends in displacement [51,52,53,54]. This piece of equipment helps to overcome some of the limitations of traditional ones, such as the low temporal resolution of measurements, enabling a more accurate examination of the kinematics behaviour of slow landslides, especially if displacement data are coupled with other continuously monitored controlling factors.

As claimed by some authors [10,40], thorough examination of representative case studies provides valuable insights for deciphering general principles on the basis of the patterns of movement of slow-moving landslides. In light of this, the main goal of this work is to investigate the kinematics controlling factors of a reactivated slow-moving landslide located in the eastern Liguria region (north-western Italy) through extensive field-based measurements. Such a landslide is partially urbanized, and in recent years, significant accelerations of the movements have resulted in extensive damage and the evacuation of several residential buildings. The main engineering–geological and kinematics characteristics of the landslide were outlined by [55]. In this study, based on analysis and comparison of new and previous monitoring data, further aspects on the preparatory and triggering factors conditioning the patterns of landslide activity are presented. To fulfil this study’s goal, a detailed analysis of continuous hydrological and geotechnical monitoring data series has been performed. Specifically, displacement and pore water pressure trends have simultaneously been obtained by means of fixed-in-place inclinometric probes and a pressure transducer, respectively. All these sensors have been positioned at the main deformation zones known from previous geotechnical monitoring campaigns performed through mobile inclinometers, enabling the examination of the connections between pore water pressure and deep displacements over a period of two and a half years.

2. General Setting of the Study Area

2.1. Geological, Geomorphological, and Climatic Framework

The landslide under investigation is located in north-western Italy, along the eastern coastal side of the Liguria region and the easternmost portion of the Genoa Province, in a site called Fontane di San Bernardo, which is placed in the hills immediately behind the town of Sestri Levante (Figure 1).

From the geological point of view, this sector of Liguria belongs to the north-western segment of the Northern Apennine, a fold-and-thrust belt characterized by terrains referred to as oceanic, transitional, and continental paleogeographic environments. During the Apennine orogenesis, these successions were piled-up to constitute the following main tectonic units (bottom to top): the Tuscan Nappe (i.e., Tuscan Domain), the Canetolo Unit (Sub-Ligurian Domain), the Ottone Unit (External Ligurian Domain), and the Bracco-Val Graveglia, Gottero, and Lavagna units (Internal Ligurids Domain). The latter are included in the so-called Vara Supergroup succession, which is dominantly outcropping in the study area and is characterized by an ophiolitic sequence. This stratigraphic series begins with an ultramafic oceanic basement of the Jurassic Age made up of serpentinized peridotites intruded by gabbros and basalts and is followed by a volcanic sedimentary cover, spanning from late Jurassic to early Cretaceous, consisting of basalts, chaotic breccias, and cherts. The top of the sequence comprises carbonatic-siliceous pelagic sediments underlying thick bodies of siliciclastic turbidites deposited from the late Cretaceous up to Paleogene [56].

The geo-structural setting results from the complex tectonic history of the Northern Apennine that, in the study area, is largely dominated by geologic structures connected to the uplift phases, which have progressively thrusted the oceanic successions onto the transitional and continental domains. These tectonic stages were responsible for the formation of a sequence of kilometre-scale recumbent folds and overthrusts [56]. This geologic framework has been further complicated by the post-orogenic evolution associated with the formation of the Ligurian Sea that involved distensive phases, active since the late Miocene age and throughout the Plio-Quaternary, producing systems of high-angle normal faults [57].

The geomorphological outline is largely controlled by litho-structural factors [57,58,59,60]. The landscape morphology, in both inland and coastal sectors, along with the pattern of the drainage network, strongly depends on morpho-selective erosion, chiefly connected to the combined role of widespread outcrop of both strong and weak rocks and the arrangement of tectonic structures. In this regard, landforms and deposits are predominantly related to fluvial- and runoff-related processes and mass movements. Nevertheless, weak formations composed of clayey rocks typically produce gentler topography and thicker slope deposits and are generally affected by large landslides [12,57,61].

On a more detailed scale, the rocks outcropping at the Fontane di San Bernardo site, and in the surrounding areas, are related to two turbiditic formations belonging to the Internal Ligurian Domain [56], namely the Monte Gottero Sandstones Fm. (GOT—Gottero Unit, Late Campanian—Paleocene) and the Forcella Banded Shales Fm. (FBM—Lavagna Unit, Late Campanian) (Figure 2a,b). The former is represented by well stratified and thick quartz-feldspatic sandstone banks with interbedded thin shale layers; the total thickness of the formation ranges between 600 and 800 m. The FBS Fm., also known as Scisti Zonati Fm., consists of the irregular alternation of thinly bedded and polychrome clayey, silty–arenaceous, and marly shales and its maximum thickness exceeds 250 m. In the study area, GOT formation is superimposed on the FBS and both are involved in a wide recumbent antiform and NE-vergent fold, with approximately N10° axis direction [58] (Figure 2c). Locally, the contact between the above-mentioned formations can be partly masked by both landslide and slope deposits, as at the Fontane di San Bernardo site. The landslide under investigation is located on the left side of the small Staffora catchment (1.6 km²). This basin is N-S oriented and drained by a 3 km long main watercourse, which flows southwards to its confluence with the Gromolo stream. The maximum elevations are around 495 m a.s.l. and progressively decrease southwards until they cross the small alluvial coastal plain hosting the urban centre of Sestri Levante. The basin flanks are moderately steep to steep (slope gradient between 15° and 35°) and generally deeply incised by short and straight fluvial stems. The slopes are widely mantled by poorly sorted eluvial-colluvial deposits, up to some metres thick, locally anthropic reworked for agricultural purposes.

The climate is the hot-temperate form of the Mediterranean type according to Köppen–Geiger classification [62] and is characterized by mild winters (lowest mean monthly temperature about 8 °C in January) and hot summers (highest mean monthly temperature about 23 °C in July). The average annual rainfall is around 1100 mm, according to rainfall records available since 2003 at the Sara rain gauge, which is located in the centre of Sestri Levante. However, precipitation is predominantly concentrated during autumn and winter (highest monthly rainfall about 170 mm), while in spring and summer rainfall levels are significantly lower (lowest monthly rainfall about 38 mm).

2.2. Geomorphological and Engineering–Geological Setting of the Landslide

The considered landslide affects a SW-facing slope, along which the small hamlet of Fontane is settled (Figure 3). Previous studies described this landslide as an ancient and reactivated complex movement according to the scheme proposed by Cruden and Varnes [63], albeit a dominant rotational sliding component has been supposed [61].

The landslide, which extends between 360 and 95 m a.s.l. with a length of approximately 650 m and a maximum width of approximately 250 m, can be separated into two main zones (Figure 3). The upslope part encompasses both the main landslide crown and scarp along with an upper portion of the principal head. The main crown develops along the southern side of the M. Castello ridge. Below it, a mostly emptied sector, with a general concave shape and an average steepness of 25°, occurs (Figure 3). This sector is characterized by sporadic rock outcrops alternating with remoulded landslide debris and it is almost completely terraced for agricultural activities.

Since neither historical information nor significant geomorphological indicators of movement are known or visible, this zone is considered inactive. In this work, only the downslope part of the landslide has been taken into consideration, since it is characterized by active kinematics. Based on previous information from [55,61] and local inhabitants, abrupt accelerations, interspersed with active or dormant phases, are known to have occurred from this sector of the landslide in the last sixty years. The two most important reactivations date back to 1957–1958 and 2014. The active portion of the landslide, which basically includes the displaced material, has an extension of about 5.5 hectares, developing from approximately 220 m a.s.l. up to 95 m a.s.l., with an average slope gradient of 20° (Figure 3). Laterally, it is bordered by the flanks of two secondary ridges, approximately NE-SW oriented, while the toe is currently eroded by the Staffora stream. Moreover, this part of the landslide is incised by two short and linear streams. Its geomorphological layout is somewhat irregular owing to the progressive effects of principal and minor mass movements, subaerial erosion processes, and human activity. The morphology is thus uneven, showing zones of depletion alternating with bulging areas (Figure 3).

Pepe et al. [55] outlined the main geometrical characteristics of the accumulation zone through combined analysis of data collected from stratigraphic boreholes and seismic refraction surveys deployed by local authorities and private owners at different times between 2008 and 2016 (Figure 4).

Based on such study, engineering–geological cross-sections in the longitudinal and transverse directions were developed, showing that the thickness of the landslide deposit is extremely spatially variable. Specifically, the landslide thickness was found to progressively increase in the downslope direction, from approximately 9 m at the upper edge to approximately 27 m at the landslide foot. However, much more thickness fluctuations were observed laterally, at short range as well. This was pointed out by concave-shaped low-velocity anomalies of seismic waves and, subsequently, by the similar geometrical configuration of low-resistivity anomalies (<100 Ohm m) detected by means of electrical resistivity tomography (ERT) surveys, which were performed along the former seismic alignments (Figure 5a,b). These anomalies mainly correspond to depressions filled by low-stiffness and porous landslide deposits [55]. Some minor discrepancies between the seismic and ERT sections are likely due to the different response of the two distinct physical parameters to the same subsoil variations in porosity, lithology, and degree of alteration. The lateral changes in the geometry of the landslide accumulation depend on the complex architecture of the underlying bedrock, and they were correlated to buried hollows/channels likely inherited from the ancient pre-failure slope morphology (Figure 5c).

The core samples extracted from stratigraphic boreholes along with geotechnical laboratory analyses showed that the landslide-forming materials are heterogeneous in terms of grain size [55]. The landslide deposit is broadly made up of clayey sands and clayey–silty sands (SM-SC and SC soil groups according to the USCS classification), irregularly alternated with clayey–silty gravels and clayey gravels (i.e., GC and GM-GC groups). Coarser soils typically consist of gravel contents ranging between 30 and 45% and sand percentages approximately between 25 and 35%, while the percentage passing through sieve number 200 (i.e., particle size < 75 μm) ranges between 25 and 45%. Fine-grained soils consist of gravel and sand percentages as high as approximately 15% and 55%, respectively, with silt and clay amounts varying between 30 and 50%. Soil materials are characterized by low plasticity (plasticity index < 15%), whereas the bulk unit weight determined on partially undisturbed specimens is between 19.4 and 21.5 kN/m³.

The landslide accumulation overlays the firm bedrock that consists of rocks belonging to the FBS Fm. Drilled boreholes reveal that the rock mass is lithologically heterogeneous and sometimes intensely fractured, resulting in a very poor to fair geomechanical quality. Rock materials can include clayey shales, clayey–arenaceous shales, silty–clayey shales and marly shales, typically thinly bedded (from a few centimetres to decimetres thick). Owing to the intrinsic fissility of lithotypes and the tectonic disturbance, the RQD index ranges between 0 and 55%. The outcropping rock masses can be described as either structurally complex (A2 and B2 groups) or heterogeneous and tectonically disturbed (rock mass Type VIII), according to [64,65], respectively. Rock core samples result from grey-black to brownish coloured according to the weathering degree and exhibit a weak strength. The average unconfined compressive strength is 15 MPa while the natural unit weight is around 26 kN/m³. Bedrock horizons directly underlying the landslide deposit can be highly weathered, weakened, and fractured, sometimes locally crushed and de-structured, with litho-relicts immersed in a soil matrix.

3. Materials and Methods

3.1. Previous Monitoring Information

A first monitoring campaign of subsurface deformation was set up in 2006 and consisted of one vertical borehole drilled approximately in the middle portion of the landslide accumulation that was equipped with an inclinometric casing tube (Figure 3). During a biennium of measurements (i.e., from October 2006 to October 2008) executed through a removable inclinometric probe, any significant deformation was detected: the cumulative horizontal displacements measured at the tube head were lower than 1.5 mm.

A second monitoring configuration was defined after significant acceleration of the movements occurred between January and February 2014, following a particularly wet period of several months [66,67]. Due to the considerable displacements, some deep tension cracks developed through the landslide body (see Figure 3), while some buildings of the Fontane hamlet suffered major damage and had to be evacuated. Some months later, in June 2014, subsurface displacement measurements started along flexible inclinometer casings that were installed into four boreholes (i.e., FON1, FON2, FON3, and FON4) drilled along the longitudinal direction of the landslide accumulation (see Figure 3). Their lengths below the ground level were 14, 19, 17, and 29 m, respectively. Two tubes, namely FON1 and FON4, were installed at the upper edge and foot of the active portion of the landslide, respectively. Instead, the FON2 and FON3 tubes were in the middle section of the landslide, approximately at the same elevation and 50 m apart from each other. After the sharp reactivation that took place in February 2014, the landslide remained active (Figure 6a), showing an extremely slow to very slow mobility according to [63]. In all inclinometric tubes, the shape of displacement profiles is characterized by pronounced and localized horizontal movements. Along the verticals FON3 and FON4, no significant profile perturbations were detected above the main levels of concentrated deformation. As far as the FON1 and FON2 tubes, slightly more irregular displacement profiles were obtained, probably because of minor internal deformations or flaws connected to the casing installation (e.g., grouting). As shown by the vectors of cumulative horizontal displacement measured at the tube guide heads (Figure 6b), the movements were consistent with both the dip direction of the slope and the pattern of tension cracks developed during the 2014 reactivation. However, both the displacement magnitude and average azimuth were slightly different at the section of the landslide foot (i.e., FON4 tube).

3.2. Monitoring Network and Equipment

The investigation of the factors controlling the kinematics of the landslide was guided by the antecedent monitoring results described above. Therefore, in June 2015, fixed automatic inclinometric probes were installed inside the four boreholes drilled after the reactivation occurred in winter 2014. This was carried out with the purpose of increasing the temporal resolution of subsurface displacement measurements through the acquisition of continuous data time-series. The automatic probes examined in this study were placed at the same depths as the main sliding levels detected during the previous monitoring phases conducted by means of traditional measurements. By considering the shape of prior horizontal subsurface displacement profiles, these unstable zones were assumed to be reasonably representative of the mobility across the whole thickness of the active portion of the accumulation. Specifically, the inclinometric sensors were positioned at a depth of 6 m in FON1, 12 m in both FON2 and FON3, and 24 m in FON4. According to the available stratigraphic logs (Figure 7), in the FON1 tube, the installed probe acquired deformations developing in a horizon of the landslide deposit consisting of clayey gravels (i.e., GC). In the FON2, FON3, and FON4 tubes, the sensors were placed in the proximity of the interface separating the landslide accumulation, consisting of either clayey sands (i.e., SC) or silty–clayey gravels (i.e., GM-GC), and the most weathered and de-structured bedrock levels (Figure 7).

The automatic probes provide the amount of horizontal displacement by means of Hall element sensors that measure the inclination of the casing with respect to the vertical direction. The deviation from the vertical axis of the casing tube produces a change in magnetic flux density, which is detected from the Hall sensor in the form of voltage. The accuracy of the tilt angle readings is 0.5% in the ±10° full-range scale, producing errors in displacement measurements of approximately ±0.1 mm [68]. The temporal resolution of the displacement measurements was set to be hourly. During the examined monitoring period, the four verticals were also surveyed through a mobile biaxial inclinometric probe to detect deformations both above and below the continuously monitored levels. Therefore, four traditional measurements were also performed by temporarily removing the permanent automatic probes. During the measurements, the probe was lowered to the bottom of the casing tube and then raised incrementally in 1 m intervals. The accuracy of the traditional inclinometer equipment is ±2 mm per 30 m of casing tube [69].

To monitor the pore water pressure, an automatic piezometric sensor was installed within the core destruction borehole P3, which was drilled close to the FON2 inclinometer tube. The pore water pressure sensor was positioned at a depth of 12 m below the ground surface, which corresponds to the location of the most marked displacement previously identified along the nearby FON2 inclinometer. The installed sensor consists of a piezoelectric element bonded to a flexible diaphragm; the deflection produced on the latter by the change in water pressure is detected by the transducer and converted into an electrical signal directly proportional to an absolute value of pore water pressure [70]. In the measurement section, the piezometric sensor was placed in a small sand-filled pocket of the drilled borehole, which was subsequently sealed-off with packed bentonite. The accuracy of the electric piezometer is 0.25% of full scale, which was 200 kPa in this case, giving an error of ±0.5 kPa [71]. Each inclinometric and piezometer sensor was connected to a datalogger for measurement acquisitions. All monitoring systems are powered by a photovoltaic panel and include a wireless device for transferring measured data.

Starting from June 2015, rainfall levels on the landslide were measured by a tilting bucket rain gauge installed at the landslide head (Figure 3). Nonetheless, some technical problems and malfunctions affected precipitation recording during the considered monitoring period which resulted in some data gaps (86% data coverage). Therefore, rainfall data collected from the Sara rain gauge (100% data coverage) were used. This station is located approximately 2.7 km south-east from the landslide under investigation.

3.3. Framework of Monitoring Data Analysis

Geotechnical and hydrological monitoring data, simultaneously acquired from June 2015 to December 2017, were analyzed to explore the patterns of movement of the Fontane di San Bernardo landslide and to investigate both the factors and mechanisms regulating its kinematics. The investigation has been developed following different phases. Rainfall and piezometric trends have first been taken into consideration to examine both the hydrological and pore water pressure regime throughout the monitoring period. In this regard, rainfall levels over different time spans (i.e., yearly, monthly, and daily) have been determined, while sub-daily pore water pressure data have been averaged into daily values and their fluctuations have been examined through the definition of minimum, maximum, and mean values. Moreover, pore water pressure change descriptors, namely daily variation and deviation from daily mean, have been examined. Subsequently, the relationships between rainfall and pore water pressure have been considered to investigate the hydrological response of the landslide.

In order to identify the patterns of movement, the displacement time-series acquired by automatic inclinometric probes were used. Hourly displacement readings were averaged over 24 h to obtain daily mean values. When uniform trends in movement from the time series were recognized, the least-squares linear fitting method was applied to daily cumulative displacements. The slope of the best fitting line was deemed as indicative of the mean displacement rate, while the standard error of the estimate was defined as a measure of the precision of the assumed linear model. The obtained mobility rates were considered statistically significant if the displacement gradient was larger than the standard error of the gradient at the 95% confidence level.

To explore in detail the relationships between rainfall, pore water pressure, and subsurface displacement, monitoring data acquired by sensors installed in FON2 and P3 verticals were used. These two sensors were positioned close to each other (see Figure 3) and at the same failure depth, which was detected in advance through traditional inclinometer measurements, and were therefore considered to be adequately representative of the characterization of deep displacement mechanisms.

4. Results

4.1. Rainfall Data

Rainfall data recorded over the monitoring period are shown in Figure 8, while their most important statistics are summarized in

Table 1. Summary of statistics of rainfall data measured by the Sara rain gauge over the investigated period.

Time Interval	2015	2016	2017	2015–2017
Data coverage (%)	100	99.7	100	99.8
Total cumulative rainfall (mm)	591.2	1139.8	720.0	2451.0
Maximum daily rainfall (mm)	87.8	130.4	107.0	130.4
Maximum monthly rainfall (mm)	338.4	335.6	218.6	338.4
Mean daily rainfall (mm)	2.9	3.1	2.0	2.6
Mean monthly rainfall (mm)	89.4	94.2	57.2	78.8

.

Both daily and cumulative rainfall levels followed trends consistent with the local hydrological regime, revealing seasonal cyclic variations with alternating drier and wetter periods. The rainiest phases generally coincide with autumn and winter months, whereas the driest ones coincide with the summer season (Figure 8a). This is more clearly visible from the pattern of the cumulative deviation of monthly rainfalls from the monthly mean for the entire period (Figure 8b).

The most pronounced positive variations, namely where monthly rainfalls were greater than the mean, typically occurred in October and November or in January and February. However, the cumulative rainfall amounts, along with the monthly and daily average values, reveal that 2016 and the second half of 2015 were wetter than 2017 (Table 1). The total rainfall measured in 2017 was far lower than the local mean annual rainfall (i.e., 1100 mm) and this was chiefly correlated to a dry period covering most of the spring and summer seasons. It can be noted that the rainiest months were October in 2015 (338.4 mm), followed by February in 2016 (335.6 mm) and December in 2017 (218.6 mm). Similarly, the maximum daily rainfall levels were recorded in February 2016 (130.4 mm), December 2017 (107.0 mm), and October 2015 (87.8) and correspond to the most important rainstorms that occurred during the monitoring period (Table 1 and Figure 8a).

The rainfall events that occurred in October 2015 and December 2017 were characterized by rainfall concentrated over no more than 2–3 days that were followed or preceded by periods, ranging from 10 to over 20 days, with little or no rain (Figure 8a). Conversely, the rainy phase of February 2016 was characterized by two main rainstorms separated by a 20-day wet period with accumulated rainfall levels equalling 108.8 mm over 11 rainy days; also, rainfall continued for some days after.

4.2. Pore Water Pressure Data

The temporal distribution of pore water pressure measurements covers 94% of the monitoring window, with three short data gaps caused by either some malfunctions in data acquisition or power supply interruptions (Figure 9a).

In general, at the measuring depth, the pore water pressure data showed no significant long-term variations, with a mean value of 14.3 kPa and a standard deviation of approximately ±2.5 kPa over the acquisition period (Figure 9a). The highest pore pressure value (26.8 kPa) was recorded in early March 2016, whereas the lowest one (7.0 kPa) was recorded in mid-March 2016. Conversely, short-term changes were more frequently observed, albeit typically small in magnitude. The largest pore water pressure oscillations were recorded between February and early March 2016 during the multi-day rainy period characterized by the concomitant occurrence of extremely high and moderate rainfall levels (Figure 9a). As outlined by the diagram showing the daily variation in pore water pressure, following single-day extreme precipitation, no relevant pore water pressure build-up developed (Figure 9b). For example, in December 2017, more than 100 mm of cumulated daily rainfall produced a slight increase in water pressure (i.e., 2.0 kPa).

The general pattern of the piezometric data was consistent with the graph of the deviation of the daily pore water pressure from the mean of the observation period (Figure 9c). Indeed, the acquired time-series of pore water pressure data were largely characterized by small fluctuations around the mean value. The overall piezometric trend was further pointed out by the shape of both the histogram and the boxplot of the deviation of daily values from the mean pore water pressure: the data distribution was unimodal, with 75% of the deviations between −1.4 and 1.7 kPa, and approximately normally distributed, showing a slight right asymmetry and a positive kurtosis (Figure 10). Nevertheless, some high outliers corresponding to major positive deviations were visible (Figure 10b).

These anomalous deviations arose from two pore water pressure pulses generated during the rainy phase that occurred between February and early March 2016 (Figure 9a). These two pressure fluctuations appeared to be characterized by relatively similar behaviour: the maximum values achieved at the ascending limbs were comparable, namely 26.0 and 26.8 kPa, respectively, and both were followed by descending limbs indicating a progressive dissipation of excess pressure (Figure 11a).

However, the pore water pressure evolution following the first peak was only partially detected because of technical problems that affected the data acquisition system. After the second peak, the pore water pressure dropped to a lower value than before the start of precipitation (Figure 11a). Interestingly, the pressure peaks were not coincident with the highest daily rainfall levels, since they were attained 11 and 4 days after, respectively (Figure 11a). It follows that, at the considered depth, the responses of pore water pressure to single high-intensity rainfall episodes were relatively slow. However, during these two multi-day spans, characterized by both light and moderate precipitations, progressive water pressure growth took place. This would indicate that the generation of pore water pressure peaks was likely correlated to the attainment of specific thresholds in cumulated rainfall. In this regard, based on the time interval elapsed during the formation of the first water pressure wave, the 14-day cumulated rainfall was considered. However, as shown in Figure 11b, the obtained outcomes are ambiguous since the observed maximum values of water pressure did not correspond to a unique level of cumulated rainfall. Indeed, the two peaks occurred when 14-day cumulated rainfall levels were approximately 210 and 125 mm, respectively (Figure 11b), revealing that the relationship between rainfall and pore water pressure response was not straightforward.

4.3. Deep Displacement Data

Despite some brief data gaps, the time-series of cumulative displacement acquired by automatic inclinometer probes were sufficiently continuous over the observation period (Figure 12). The data coverage was from 93% to 95% for FON2, FON3, and FON4 probes; a lower data coverage (86%) was obtained for the probe installed inside the FON1 tube. The breaks in data continuity were mainly caused by either electrical malfunctions or power supply interruptions. The trends in daily averaged cumulative displacements appeared to be slightly noisy, likely because of the wide range of errors commonly affecting data acquisition of automatic inclinometer probes [52,53]. It is known that the performance of automatic equipment is influenced by disturbances in the electrical signal, deriving from noise sources along with systematic and random errors affecting both the probe itself and the acquisition system, which can produce measurement spikes [52,53]. Moreover, as observed in the collected data-series (Figure 12), these disturbances also include decreases in cumulative displacements, which in turn result in negative values of daily velocity. However, these values were approximately less than or equal to the accuracy of the installed probes and thus they were assumed negligible for the purposes of the analysis.

Deep displacement data allowed us to outline noticeable features of the landslide mobility. The probes installed within the verticals located in the central part of the active landslide sector (i.e., FON2 and FON3) achieved the largest total cumulative displacements throughout the monitoring period (Figure 12). Smaller displacements were measured from the probe placed inside the tube located at the landslide foot (i.e., FON4). Conversely, the time-series of displacement acquired by the FON1 probe did not show significant movements at the landslide head. During periodical measurements with removable probes, no significant deformations were detected above and below the levels monitored continuously, testifying that they represent the most important failure surfaces within the landslide body (Figure 13a). Cumulative displacement data acquired by automatic inclinometers were quite consistent with those obtained from mobile probe, although the low temporal resolution of the latter did not allow for reliable comparisons in terms of motion trends (Figure 13b). The largest discrepancies were noticed for displacements measured inside the FON1 tube. However, it should be noted that the FON1 automatic probe did suffer from technical problems, which may have affected its functioning. For these reasons, data acquired along this vertical were not considered in the following analyses.

The displacement trends obtained at the other three surveyed failure zones appeared to be consistent and comparable, suggesting similar kinematics behaviour. Based on the time-series of average cumulative daily displacement, three main types of kinematics stages were distinguished, which were defined as slow (SM), rapid (RM) and inactive (IM) motions (Figure 12). The SM and IM stages included periods showing relatively uniform or constant trends in the time-series of displacement. At first glance, these motion patterns appear to follow seasonal cycles and thus are likely influenced by the hydrological regime. Overall, the SM stage was observed during wet periods, typically comprising the autumn and winter months, whereas the IM stage was observed during intervals characterized by very low rainfall levels, namely approximately from early spring to late summer (Figure 12). In contrast, the RM stage entailed sharp increases in cumulative displacement. This stage was episodic and associated with particularly severe rainfall conditions. More detailed descriptions of each discerned kinematics stage are reported in the following subsections.

4.3.1. Rapid-Motion Stage

This motion stage was recorded between February and early March 2016, with a duration of approximately one month, and consisted of two main events of movement. It was detected at all three monitored failure depths, albeit with some slight differences in terms of both magnitude and the timing of deformations (Figure 14).

The total displacements measured by the FON2 and FON3 probes were comparable, showing two simultaneous positive gradients in cumulative displacement (

Table 2. Summary of the most important duration and mobility data regarding the rapid-motion (RM) stage.

Probe Id	Period	Daily Data Records (Days)	Total Displacement (mm)	Max. Displacement Rate (mm/Day)
FON2	6 February 2016–10 March 2016	31	5.4	0.9
FON3	6 February 2016–10 March 2016	19	5.2	1.7
FON4	6 February 2016–10 March 2016	28	1.9	0.4

and Figure 14a). However, it should be noted that some interruptions in data acquisition occurred due to power outages. The longest lack of data affected the FON3 probe during the first part of the motion stage, with a duration of 14 days. Instead, the magnitude of deformations was lower at the landslide foot (i.e., FON4 probe) and roughly within the margin of accuracy of the sensor during the initial phase of the stage; here, only one small step in cumulative deep displacement occurred, which was, however, synchronous with the second one recorded by the probes within the FON2 and FON3 tubes. Some interesting outcomes were also noted in terms of daily displacement rate (Figure 14b and Table 2).

Overall, the maximum rates of displacement recorded during the whole RM stage can be classified as “very slow” according to [63]. More specifically, during the first phase of motion, a sequence of short duration and small magnitude increases and decreases in displacement rate were generally recorded in FON2, while in FON3 a remarkable acceleration, outlined by an isolated peak in velocity, occurred (Figure 14b and Table 2). The second phase of motion was characterized by a noticeable acceleration followed by a deceleration. Interestingly, this marked increase in displacement rates was simultaneous across the various sectors of the landslide, but did not appear to be directly triggered by the single severe rainstorms.

4.3.2. Slow-Motion Stage

Over the observation period, three SM stages were discerned from the displacement time-series (Figure 12). These periods were characterized by very similar durations, on average from 4 to 5 months, and a similar hydrological regime, suggesting a seasonal coincidence. Indeed, the cumulative rainfall levels were approximately between 410 and 465 mm, with a mean daily rainfall between 3.3 and 3.7 mm (Figure 15 and

Table 3. Summary of the most important duration, mobility, and hydrological data regarding the slow-motion (SM) stage.

Probe Id	Motion Stage	Duration (Days)	Data Coverage (%)	Average Displacement Rate (mm/Year)	Standard Error at 95% (mm/Year)	Cumulative Rainfall (mm)	Mean Daily Rainfall (mm)
FON2	SM1	139	100	2.3	±0.1	466	3.4
	SM2	135	100	5.2	±0.1	449	3.3
	SM3	110	66	4.1	±0.1	412	3.7
FON3	SM1	139	96	3.0	±0.1	466	3.4
	SM2	135	100	3.1	±0.1	449	3.3
	SM3	110	56	2.7	±0.2	412	3.7
FON4	SM1	139	98	3.5	±0.1	466	3.4
	SM2	135	84	4.4	±0.2	449	3.3
	SM3	110	88	4.0	±0.1	412	3.4

).

As denoted by the linear trends fitted to daily cumulative displacement data, these kinematics stages can be assumed to be characterized by uniform motion over time (Figure 15). The modelled average displacement rates were rather comparable among the different periods and across the various zones of the active landslide; also, all values can be considered statistically significant at the 95% confidence limit (Table 3). The computed displacement rates ranged between 2.2 ± 0.1 and 5.2 ± 0.1 mm/year (Table 3) and can be assumed as “extremely slow” according to Cruden and Varnes’s [63] classification.

4.3.3. Inactive-Motion Stage

SM stages were anticipated or followed by periods in which the monitoring data indicated that the landslide gradually ceased motion (Figure 12). This is denoted by virtually flat stretches of average cumulative daily displacement time-series for all the automatic probes (Figure 16).

Also, these kinematics stages showed similar durations for most of the monitoring period, typically six to seven months (

Table 4. Summary of the most important duration, mobility, and hydrological data regarding the inactive-motion (IM) stage.

Probe Id	Motion Stage	Duration (Days)	Data Coverage (%)	Mean Displacement Rate (mm/Year)	Standard Error at 95% (mm/Year)	Cumulative Rainfall (mm)	Mean Daily Rainfall (mm)
FON2	IM1	104	100	0.1	±0.1	185	1.8
	IM2	187	100	−0.8	±0.1	273	1.5
	IM3	222	100	−1.0	±0.1	281	1.3
FON3	IM1	104	100	0.0	±0.1	185	1.8
	IM2	187	100	−1.4	±0.1	273	1.5
	IM3	222	100	−1.5	±0.1	281	1.3
FON4	IM1	104	100	0.2	±0.1	185	1.8
	IM2	187	100	−1.1	±0.1	273	1.5
	IM3	222	95	−1.3	±0.1	281	1.3

). Moreover, these periods occurred cyclically during phases of scarce precipitations (Table 4): the highest cumulative rainfall was around 280 mm, falling approximately over seven months, with a mean daily rainfall of 1.3 mm (Figure 16 and Table 4).

The fitting of linear trends to daily cumulative displacement data revealed that movements referring to some IM periods, within the limits of the sensor accuracies, can be considered negligible and not statistically meaningful at the 95% confidence level (Table 4). Indeed, the estimated mean rates of displacement were close to the corresponding standard errors at the 95% confidence limit. However, in some cases, some slightly negative gradients were obtained (Figure 16 and Table 4), which may be explained as a result of the effects of sensitivity of automatic probes to instrument drift. Therefore, based on both monitoring and computed motion data, during these kinematics phases the landslide motion was interpreted as substantially at rest.

4.4. Relationships between Rainfall, Pore Water Pressure, and Subsurface Displacement

The relationships between rainfall, pore water pressure, and deep movements have been investigated separately for each distinguished kinematics stage through the coupled analysis of monitoring data recorded by the sensors installed in the FON2 and P3 verticals (see Section 3.3). The displacement time-series pointed out that both the SM and IM stages showed a certain degree of correlation with rainfall distribution at the seasonal scale (see Section 4.3). However, the comparison between the trends in pore water pressure and cumulative displacement revealed that water pressure did not oscillate significantly at the examined sliding depth throughout these periods (Figure 17). This was testified by (1) the computed daily mean values of pore water pressure, which were very similar for all kinematics stages and consistent with the mean of the entire monitoring period (compare

Table 5. Summary of the most important pore water pressure statistics computed for each slow-motion (SM) and inactive-motion (IM) stage identified.

Probe Id	Motion Stage	Pore Water Pressure Data
		Daily Value				Daily Change
		Mean	Max	Min	St. Dev.	Max	Min
P3	SM1	12.5	15.0	10.0	1.3	2.0	−1.0
	SM2	14.0	17.0	11.0	2.0	2.0	−1.0
	SM3	15.7	18.0	12.0	1.6	2.0	−1.0
	IM1	13.8	16.0	12.0	1.1	3.0	−1.0
	IM2	13.5	17.0	7.0	2.2	3.0	−2.0
	IM3	15.3	19.0	12.0	2.8	1.0	−1.0

and Figure 9a), and (2) by the statistical indices describing the variation in water pressure (Table 5), which indicated moderate fluctuations on both a seasonal and daily scale (see Section 4.2).

Regarding the RM stage, daily rainfall and daily pore water pressure trends were analyzed in conjunction with those of daily cumulative displacement and daily displacement rate (Figure 18a,b). It was observed that the two positive gradients of cumulative displacement matched with the pore water pressure peaks (Figure 18a), which in turn required a preparatory period of several days to be attained following the highest levels of daily rainfall (see Section 4.2). In general, these two phases of greater mobility appeared to follow rather similar styles. Indeed, cumulative displacements tended to increase simultaneously with pore water pressure growth; once initiated, motions progressively continued approaching pore water pressure peaks and then slowed down as the pore water pressure dropped (Figure 18a). However, the two motion sequences showed rather different behaviours in terms of displacement rates (Figure 18b).

The first motion phase was characterized by a fairly regular series of accelerations and decelerations of similar magnitude and duration, as shown by the serrated profile in the graph of daily displacement rate. Interestingly, these changes in movement velocity developed despite the pore water pressure trend showing a general increase (Figure 18b). Conversely, the second phase of motion was characterized by a main sharp acceleration developed simultaneously with the increase in pore water pressure (Figure 18b). These two different motion dynamics may be correlated to the magnitude in pore water pressure variations. As can be noticed from the plot of daily variations in pore water pressure (Figure 18c), the onset of the most marked acceleration coincided with the highest daily change in water pressure (i.e., around 7 kPa). Overall, the movements tended to cease only when the pressure began to decrease significantly (Figure 18b). Nevertheless, the mechanisms of deep movements in response to variations in pore water pressure appeared to be complex. In fact, the relationship between cumulative displacement and pore water pressure was strongly non-linear (Figure 19a). By comparing the graphs depicted in Figure 19, it can be noticed that, although the peak values of pore water pressure achieved during the two movement events (i.e., RM1 and RM2) were essentially the same, the corresponding displacement rates were different.

Moreover, it can be noted that the displacement rates observed during the descending limbs of water pressure tended to be lower than those of ascending limbs, especially for the second movement event (Figure 19c). In the latter case, it is also interesting to note that, after an initial rapid increase in displacement rate induced by pore water pressure growth, the motion rate continued to increase and then to decrease despite a nearly constant value of water pressure. It follows that the movements along the slip surface could show either acceleration or deceleration at the same value of water pressure.

5. Discussion

The general framework that has emerged from the whole monitoring period of the Fontane di San Bernardo landslide is consistent with the typical movement characteristics of the movement of reactivated and slow-moving landslides [2,3], as well as with previous information on the landslide dynamic itself [55]. The outcomes of both traditional and automatic inclinometer monitoring indicated that the most pronounced deformations develop at pre-existing and localized failure horizons, where the landslide materials can be assumed to be at their residual shear strength [19]. The continuous measurement of deep displacements showed that landslide kinematics have intermittent phases of extremely slow displacements and relative inactivity that can be interrupted by episodic accelerations. In the frame of this research, the identified motion patterns were defined as slow (SM), inactive (IM), and rapid motion (RM), respectively (Figure 12). These kinematics patterns are like those detected in previous studies through multitemporal in situ geotechnical monitoring of numerous reactivated and slow-moving landslides [7,10,11,35,48,72]. Unlike various examples from the literature, in this study, automatic inclinometer probes specifically installed at the main failure surfaces, detected in advance through a manual probe, were used. Furthermore, at a representative sector of the landslide, deep displacement readings were coupled with continuous pore water pressure measurements at the sliding surface. Thereby, based on the joint analysis of rainfall, pore water pressure, and displacement data, interesting aspects concerning the framework of dynamic factors that influence the landslide hydrological response and its mobility were found.

The outlined patterns of movement were identified across different sectors of the unstable volume, also showing consistent behaviour, suggesting that they could represent an intrinsic attribute of the landslide (Figure 12). Such similarities in behaviour were meaningful especially in terms of the initiation and duration of each discerned motion pattern (Figure 14, Figure 15 and Figure 16). Instead, some differences were observed concerning the magnitude of deformations, mostly during sharp motions. In this regard, the extent of deformations measured at the middle section of the unstable volume was greater than that measured at the landslide foot (Figure 14). These observations are in agreement with previous information from geotechnical monitoring, as well as with evidence of damage to buildings and infrastructures [55,61], and they could be related either to the different morphology of the two landslide sectors (e.g., [7,10,73]) or to the bedrock geometry (e.g., [74]), which is highly spatially variable for the landslide in question [55].

From monitoring data, it was observed that the SM and IM phases repeated cyclically, showing seasonality consistent with the local rainfall regime. But, unexpectedly, no significant seasonal fluctuations in pore water pressure were found (Figure 9 and Figure 17). Specifically, the SM patterns consisted of uniform and extremely slow movements developing with no statistically significant variations in pore water pressure along the failure surface (Figure 17a and Table 5). In light of this, these movements may be supposed to occur at relatively constant effective stresses and therefore attributable to viscous shear displacements. In the literature, viscous deformations were already found to have an important role in the motion of deep-seated and slow landslides [42]. Instead, during the IM periods, although the pore water pressure regimes were essentially the same as those recorded during the SM stages, no statistically significant displacements were measured (Figure 17b and Table 5). Consequently, these monitoring observations were interpreted as rest phases of the landslide that may depend on other control factors that were neither monitored nor investigated in this study.

Differently, the RM pattern showed an episodic nature linked to multi-day rainy periods characterized by both intense and moderate precipitations which led to pore water pressure peaks that in turn induced sharp displacements (Figure 18). Interestingly, time lags were observed between the occurrence of rainfall peaks and the corresponding peaks in both water pressure and recorded movements. Particularly, it was found that the development of water pressure peaks required a preparatory period of several days after the most severe rainstorms, indicating a delayed pore water pressure response along the failure surface. Other studies on reactivated and slow-moving landslides reported similar evidence [51]. In this research, the pore water pressure response to rainfall at the failure depth was found to be complex and probably related to the combined role of rainfall magnitude, duration, and frequency rather than to isolated and very intense precipitations. This would also be consistent with the largest acceleration events occurring in historical times [55]. However, reliable thresholds of cumulative rainfall levels antecedent to pore water pressure peaks were not identified, probably because of the insufficient number of significant rainfall events that occurred during the period of measurement, along with the complex hydrogeological setting of the landslide body. In fact, several factors such as the heterogeneity and anisotropy of the landslide materials, as well as their initial hydrologic conditions, can influence the rainfall infiltration processes and thus groundwater recharge and pore water pressure accumulation [25,27].

The RM phases were responsible for the largest magnitude of deformations and were driven by pore water pressure growth along the failure surface (Figure 18). This pointed out a connection between the pore water pressure-induced reduction in effective stresses and residual shear strength decay, resulting in accelerated movement along existing sliding surfaces [10,35,72]. However, no obvious relations between landslide velocity and pore water pressure were found after the onset motion, making it difficult to reconstruct the possible deformation mechanisms involved. Like in other documented cases [10,23,38,75], similar values of pore water pressure were found to correspond to different values of velocity, suggesting hysteretic behaviour. Therefore, the relationship between changes in pore water pressure and residual shear strength appeared to be far more complex than might be expected. In the literature, these field observations have often been interpreted and modelled assuming viscous behaviours [26,33,48]. For the landslide under investigation, phases of increasing pore water pressure tended to be associated with higher displacement rates than phases of decreasing water pressure (Figure 19). Moreover, during one rapid movement event, displacements tended to accelerate at a constant value of pore water pressure, probably due to an increased gradient of pore water pressure variation (Figure 19). These outcomes indicate that the resistance against sliding changes after the initiation of the movement [40,43,46]. Some authors interpreted these styles of motion behaviour assuming rate- and state-dependent frictional models [39]. Accordingly, for the landslide in question, it seems reasonable to assume that during RM stages, the motion is likely initiated as rigid-plastic frictional sliding mechanisms which subsequently may evolve as deformation processes regulated by rate- and state-dependent frictional factors. Eventually, it is worth noting that since this interpretation relied on a single measurement point, further measurements in other sectors of the landslide along with analyses of future monitoring data will contribute to improve knowledge.

6. Conclusions

This study was aimed at providing a further contribution to the knowledge on reactivated and slow-moving landslides. To this purpose, a joint examination of continuous hydrological and geotechnical monitoring data series, acquired in the framework of the long-term monitoring of the Fontane di San Bernardo landslide, was performed. In particular, pore water pressure and subsurface horizontal displacement were measured by means of automatic sensors installed at the positions of dominant movements taking place along the known failure surfaces of the landslide. The study outcomes provided new insight into the kinematics complexity of the investigated landslide and can be so summarized:

-

Over a monitoring period of two and a half years, the landslide activity revealed phases of slow displacements alternated with relative inactivity, both occurring with somewhat regular frequencies, prevalently on a seasonal scale, and characterized by similar durations. These two motion patterns can be interrupted by episodic sharp accelerations.

-

The slow-motion phases involved progressive deformations developing at average displacement rates classifiable as “extremely slow” (cf., [63]). These movements occurred at a relatively constant pore water pressure regime and were ascribed to viscous shear displacement mechanisms along the failure surface. Inactive phases entailed no significant displacements and mostly corresponded to prolonged periods of dry conditions.

-

The acceleration stages produced the largest deformations, occurring with maximum motion rates that can be indexed as “very slow” (cf., [63]). The fastest displacements were triggered by pore water pressure peaks driven by multi-day periods of significant rainfall and were associated with sliding mechanisms taking place through a rigid-plastic frictional behaviour.

-

A relatively slow hydrological response was observed along the failure surface as pore water pressure peaks took several days to form. Moreover, hysteresis relationships between pore water pressure and landslide displacement were found. These latter outcomes revealed complex hydro-mechanical behaviour of the landslide that should be explored with further investigations and monitoring.

In conclusion, this study provided compelling evidence of the usefulness of geotechnical equipment installed in direct contact with sliding zones to outline the factors that can prepare to the mobility of reactivated and slow-moving landslides. Furthermore, this study can provide a valuable contribution to set up and calibrate hydro-mechanical models aimed at simulating kinematics evolution of the Fontane di San Bernardo landslide and, therefore, helps to develop an early warning system for civil protection purposes.