Introduction

Electrical resistivity tomography data (ERT) are sensitive to important hydrological properties of the subsurface such as water saturation and pore water salinity (Archie 1942; Mualem and Friedman 1991; Henry 1997; Ewing and Hunt 2006; Amidu and Dunbar 2007; Oldenborger et al. 2007). It is now well established that this method can image subsurface properties and processes associated with groundwater and unsaturated zone systems (Daily et al. 1992; Slater et al. 1997; Zhou et al. 2001; Binley et al. 2002; LaBrecque et al. 2004; Israil et al. 2004; Looms et al. 2008; Miller et al. 2008; Van Dam et al. 2009). As this geophysical method can provide information over large areas with minimum disturbance (of the hydrological system), the use of ERT to perform groundwater/hydrologic investigations has largely spread during the last decades.

For the peculiar case of landslide studies, the use of ERT proved to be successful in identifying sliding surfaces and preferential flow paths within unstable slopes (Griffiths and Barker 1993; Jongmans et al. 2000; Lebourg et al. 2005; Friedel et al. 2006; Van Den Eeckhaut et al. 2007; Yilmaz 2007; Jomard et al. 2007, 2010; Sass et al. 2008; Göktürkler et al. 2008; Marescot et al. 2008; Erginal et al. 2009; Schmutz et al. 2009; Chambers et al. 2011; Perrone et al. 2014; Uhlemann et al. 2015; Viero et al. 2015) and was also efficient in gathering information on the unstable slope hydrogeological regimes (Grandjean et al. 2006; Jomard et al. 2007; Lee et al. 2008; Niesner and Weidinger 2008; Piegari et al. 2009; Gance et al. 2016). Such information is critical to improve our understanding of landslide dynamics, since the circulation of fluids usually plays a key role in initiating movements, in relation with the build-up of pore water pressure (Flageollet et al. 2000; Guglielmi et al. 2005; Crosta and Frattini 2008; Bernardie et al. 2014; Vallet et al. 2015). The use of a geophysical imaging technique is highly relevant, since the distribution of pore water pressures within the unstable mass is often heterogeneous due to preferential flow paths (Bogaard et al. 2000; Malet et al. 2005; Montety et al. 2007).

Recent improvements of hydrogeological studies are linked to the increasing easiness to automate ERT measurements and, thus, to perform time-lapse surveys. When static surveys allow only to image subsurface properties, time-lapse measurements enable to image dynamic changes of these properties, providing insight into ongoing subsurface processes. Examples of successful time-lapse geophysical investigations in monitoring and understanding subsurface physical processes are numerous (among others, Ramirez et al. 1995; Day-Lewis et al. 2003; Singha and Gorelick 2005; Lebourg et al. 2010; Bièvre et al. 2012; Genelle et al. 2012; Luongo et al. 2012; Prokešová et al. 2013; Travelletti et al. 2012; Lehmann et al. 2013; Palis et al. 2016).

With the development of ERT time-lapse surveys, many research works dealt with the importance of accounting for the finite time required for data collection. For example, Kim et al. (2009) showed that it was necessary to perform a conjoint inversion of resistivity measurements accounting for their acquisition time to recover information typical of an ongoing infiltration process. Long-term studies focusing on properties changes at seasonal or even longer time scale are little concerned by this problem but should be designed to record at least one entire cycle of the hydrological process of interest. For these kinds of studies, few thorough investigations were carried out to compare different approaches that can identify resistivity changes correctly while minimizing the effects of environmental noise, data errors and other factors. Inversion process can add uncertainties because of the non-uniqueness solutions while resolving multiple iterations. This is why we chose to study here the signal given by the measured raw apparent resistivity data.

Up to now, ERT has been used in very few studies to characterize the long-term hydrological behaviour of an unstable slope (Lebourg et al. 2010; Palis et al. 2016; Gance et al. 2016), even though this knowledge is essential to understand the non-linearity of the slope response as the amount of deformation due to a rainfall event strongly depends on the hydrological state of the massif prior the rainy episode (Mikŏs et al. 2004; Lindenmaier et al. 2005).

In this paper, we propose to monitor the hydrogeological response of an unstable slope in Vence (south-eastern France) using ERT over a long period of time (9.5 years). The aim is to identify in the space and time domains direct and indirect variables that might be associated with the deformation process. In this regard, a new specific signal processing method is proposed to analyse apparent resistivity. A special attention was paid to data filtering as the method needs a good signal to noise ratio to be conclusive.

The field site, the “Vence” landslide

Landslide history

The “Vence” landslide, located in the Southern French Alps, is about 0.8 ± 0.2 million m3 and develops on a 250-m-wide and 350-m-long surface, with a mean slope angle varying from 12° to 14° (Lebourg et al. 2010, Fig. 1). The movement affects an Eocene sandy-clay layer that overlays a fractured and faulted Jurassic limestone (Mangan 1982, Fig. 1a). Both layers are part of an Oligocene syncline fold (Mangan 1982). The network of vertical faults affecting the limestone is organized according to regional directions: globally E–W and N–S. They correspond to an inherited Mesozoic fault or are related to the extensional tectonic event during the Lower Oligocene (Laurent et al. 2000). These faults are not visible on the experimental site as they are covered by the Eocene sandy-clayed layer, but can be observed on the limestone outcrops further from the landslide. In addition, these faults were mapped beneath the landslide during a previous geophysical campaign using ERT and drillings (Lebourg et al. 2010, Fig. 1b). There are many evidences of instabilities on the field such as tension cracks, scarps, disorders on buildings and the deviation of the Lubiane River at the foot of the slope (Fig. 1c). Since the 1980s, records show a strong relationship between water inflows (especially heavy rainfalls) and later accelerations and reactivations of the sliding process. This late statement is based on historical observations as disorders on houses appearing after particularly strong rainfalls in 2000 (250 mm rain amount in 2 days). Other catastrophic events indicate a strong relationship between water inflows and acceleration of the landslide. For example, the acceleration of the landslide kinematics induced the obstruction of the Lubiane River bed in October 1981 when an intense rainfall episode (250 mm in 24 h) generated mud flows.

Fig. 1
figure 1

Presentation of the Vence landslide. a Simplified geological cross section of the landslide along sliding direction (see b for the location). b Layout of the field observations and the instrumentation installed on site. c Photographs of the head scarp after winter 2000 event

Full size image

This landslide, located within an urban area (up to 26 houses constructed from the 1970s to the 2010s), has been subject to important issues and presents a real risk for local authorities. Today, a dozen houses are being destructed in the most active parts of the landslide. The interest in studying this active landslide has then been increased and understanding the processes leading to such destabilizations became a serious challenge for our team.

Field instrumentation

The response of the slope to water inflows is very complex. It consists in a combination of long-term seasonal behaviour and short-term response to rainfalls. Thus, we need to study the hydrological behaviour of the slope at long time scale in order to set apart and characterize these long and short trends. On that account, our team started a long-term monitoring in 2006.

The survey combines an electrical resistivity tomography (daily acquisition) and a rain gauge since 2006, as well as four boreholes monitoring groundwater level and temperature since 2009 (Fig. 1b). Three boreholes are also equipped with inclinometers (C6, C11 and C20; Fig. 1b).

The electrical profile uses 24 electrodes, with 5 m spacing. It is acquired daily in dipole-dipole configuration with a Syscal R1 Plus Switch 48 resistivity meter (developed by IRIS Instrument). The distance between electrodes pairs (injection or potential electrodes) ranges from 5 to 20 m (i.e. minimum and maximum quadripole lengths are 15 and 115 m, respectively). As this configuration is very sensitive to horizontal changes of resistivity, it is suitable to detect vertical structures like the faults affecting the calcareous bedrock.

Probes in the boreholes are located at few meters depth (C6 = 8 m, C7 = 10 m, C11 = 6.5 m, C20 = 7 m) and data are sampled every 10 min. Because of field issues, data from borehole C7 are available until late 2010, data from C11 until 2012, whereas boreholes C6 and C20 are still running.

The permanent ERT is installed on the border of the sliding mass, above a vertical fault affecting the limestone bedrock. The borehole with piezometer C6 is located few tens of meters from the ERT, close to the same vertical fault (Fig. 1).

All the devices are connected to a monitoring point, plugged into the 220 V outlet. Data are transmitted in near real-time thanks to an internet access on site and recovered in our lab.

Hydrological settings

Since there is a major influence of water inflows on the landslide dynamics, a field campaign using piezometers and ERT was performed in order to better characterize the hydrological settings (Lebourg et al. 2010). Previous field campaigns highlighted the coexistence of a deep karstic aquifer in the Jurassic limestone with a free water table in the Eocene sandy-clay layer, which roughly corresponds to the unstable mass (Fig. 2).

Fig. 2
figure 2

Conceptual hydrogeological scheme of the Vence landslide

Full size image

There is however a lack in the knowledge in the water transfers within the deep aquifer. It is indeed very hard to estimate the water supplies fluxes from upstream to downstream in depth. Besides, the network of vertical discontinuities affecting the limestone allows the transfer of fluids between the two reservoirs (Fig. 2). Some of these discontinuities were mapped using ERT and outcrops measurements (Fig. 1). The 8 years of piezometric measurements show the following:

  • a seasonal behaviour of the bottom layer with a gradual drying of the deep aquifer till the end of summer when it reaches its lowest level (about 3 m below ground surface), before returning to its highest level in a few days time in mid-autumn (about 2 to 1 m below ground surface, respectively, piezometers C20 and C6; Fig. 3).

  • a rapid response of the free water table to rainfalls when the piezometric level rises of several centimetres only few hours after the rainfall event and then starts to decrease 48 h later (Fig. 3).

Fig. 3
figure 3

Groundwater level at C6 and C20 (Fig. 1b) and cumulated rainfalls index (Helmstetter and Garambois 2010) with t c  = 1 day from December 2005 to January 2016. Narrow yellow rectangles highlight examples of rapid changes of groundwater level after a major rainfall event, while the large clear yellow rectangle highlights typical variations during a 1-year span

Full size image

Then, since the free shallow water table is directly feed by rainfall waters and the water supplies of the deep aquifer come from upstream rainfall, we will take into account the rainfall as the main source of water flowing into the landslide. To highlight the response of meteorological solicitations with the groundwater level variations, we use the cumulative rainfall index P c at time t i defined by Helmstetter and Garambois (2010) as:

$$ {P}_c\left({t}_i\right)=\sum_{j=0}^nP\left({t}_j\right) \exp \left(-\frac{t_i-{t}_j}{t_c}\right) $$
(1)

where the sum of the previous rainfalls decreases exponentially with time. This rainfall model takes into account the amount of rainfall water available for infiltration until a characteristic time t c set at 1 day.

Processing of ERT survey

Filtering of apparent electrical resistivity data

Electrical profiles were acquired daily over a 9.5-years period, i.e. a total of 3510 profiles were acquired between July 2006 and January 2016. Each of these profiles has 574 resistivity measurements. As often when geophysical instrumentation is deployed on the field, some acquisition problems were encountered (disruption of power supply, severed electrical wires, device failures, normal aging of the devices or destruction caused by animals). These problems are not linked to slope movements that are very limited at the ERT location (<10 cm in 10 years). A filtering procedure is then necessary to overcome these instrumentation and acquisition problems in order to extract scientifically exploitable data.

First, acquisition issues were detected: the measurements showing a wrong value of injection current intensity were removed (i.e. intensity lower than 5 mA and higher than 2 A). The filtering process was then achieved by the removal of apparent resistivity measurements non-compliant with some quality criteria. Apparent resistivities having a standard deviation greater than 5% were removed (data quality is improved by stacking 3 to 5 measurements for each quadripole and the standard deviation account for the repeatability of these measurements), as well as those with aberrant values given the site characteristics (i.e. resistivity measurements lower than 1 Ω m or greater than 400 Ω m). The lowest resistivity corresponds to fresh water; the highest resistivity corresponds to the highest apparent resistivities measured during low flow periods (excluding instrumental problems). Once this first rough filter is applied, acquisition days showing more than 15% data loss were removed. Then, each quadripole presenting more than 25% loss over the whole period was removed. We then applied a second filter on the dataset that kept the values within a confidence interval of 99% (μ − 3σ < valid data < μ + 3σ, where μ is the median value and σ is the mean strandard deviation) calculated over the largest stable period of acquisition for each quadripole. After this second filter, the days of acquisition with more than 15% data loss were removed from the study. Figure 4a gives an idea of the high variability of the raw dataset and the repartition of the filtered values.

Fig. 4
figure 4

Results of the filtering process. a Comparison between the repartition of the apparent resistivity data before (raw data) and after filtering (filtered data). b Daily percentage of available data (after filtering) and the total amount of acquisition (acquired data) for the survey duration. c Variation of the daily apparent resistivity median from July 2006 to January 2016. Values are linearly interpolated for period without data (grey line)

Full size image

After the data filtering, apparent resistivity measurements were still available for 2634 profiles (75% of the 10 years monitoring), with a high amount of “valid” measurements (>85%). Figure 4b shows the percentage of these “valid” measurements per profile for the survey duration. We linearly interpolated resistivity values for periods with data loss (see example on Fig. 4c).

Inversion of apparent electrical resistivity data

The complex link between changes in apparent resistivity and subsurface processes does not allow identifying different geological units with a quantifiable behaviour (e.g. drainage zones vs. impervious zones). Therefore, we inversed the data using the software Res2Dinv (Loke 1996-2014). The inversion routine is the smoothness constrained least-squares inversion with an L2-norm implemented by a quasi-Newton optimization technique (Loke and Barker 1996).

Two inverted resistivity profiles are shown in Fig. 5a. The same filtering process (see previous section) and the same inversion parameters (similar meshing and initial model, the same damping factor and number of iterations) were used to compare the two sections which variations in resistivity contrasts are due to changes in the medium and more precisely to water content. This groundwater level change is illustrated in Fig. 5b, where the piezometric levels of both C6 and C20 piezometers raise.

Fig. 5
figure 5

a Comparison of inverted resistivity profiles from the 10/12/2013 (left) and 20/01/2014 (right). b Evolution of the piezometric levels at C6 and C20 (see Fig. 1b for the location) with the rainfall amounts from 15/11/2013 to the 31/01/2014. The acquisition days of the two inverted resistivity profiles are highlighted thanks to vertical red bars

Full size image

The differences in resistivity contrasts in these two tomograms are clearly shown and are the result of the incoming water into the ground. The ground responds almost instantaneously (few hours) to meteorological forcing (rainfalls) as shown by the piezometers at shallow depth and still keeps a certain water level height during several days after the rainfall event (Fig. 5b). These variations can also be seen at greater depths thanks to daily ERT measurements, allowing an investigation depth of about 30 m (Fig. 5a).

On the ERT section of the 20/01/2014 (Fig. 5a), the bottom surface of the landslide body is clearly seen. The 10- to 15-m-thick conductive area on the top of the profile corresponds to the porous unstable zone. It overlays a resistive zone that corresponds to the limestone basement. This layer, supposed to be permeable, is cut by fractures and faults, which can be identified on the ERT section (yellow vertical discontinuity at 60 m from the beginning of the line). This image has been acquired while the groundwater level was high. The saturation degree, making the medium more conductive, helps the electrical current to travel in the ground, making the resistivity contrasts even higher. This produces an accurate image of the geological discontinuities. In the case of the ERT of the 10/12/2013, the groundwater level being lower, the dryer state makes the medium more resistively homogenous. This is why the discontinuities appear less sharp and only reveal the highly resistive areas as well as the highly conductive zones.

Cluster analysis method

To extract information relative to some typical hydrological behaviours of the slope from apparent resistivity data, we achieved a clustering analysis based on a hierarchical clustering algorithm (Jain and Dubes 1988) which constructs agglomerative clusters using a distance criterion. The distance criterion is calculated using the value of the Euclidian distance between each resistivity measurement for the whole acquisition period.

The distance d(ρ 1, ρ 2) between two resistivity measurements ρ 1 and ρ 2 can be expressed as follows:

$$ d\left({\rho}_1,{\rho}_2\right)=\sqrt{\sum_{i=1}^n{\left({\rho}_{1_i}-{\rho}_{2_i}\right)}^2} $$
(2)

where i is the acquisition day. The matrix of the Euclidian distances for all resistivity measurements is presented in Fig. 6a.

Fig. 6
figure 6

Clustering method of the apparent resistivity. a Distance matrix for the 574 resistivity measurements: yellow indicates a high similarity between two measurements along the whole studied period while dark green shows a high dissimilarity. b Complete dendrogram illustrating the agglomerative procedure used to constitute clusters of resistivity measurements using a distance criterion based on the Euclidian distance. c Example of resistivity measurements grouped into a cluster of 470 elements for the period 2006–2016 (cluster 1)

Full size image

The closest resistivity measurements (i.e. with a similar resistivity range for the studied period) present the most similar temporal variations and, thus, are clustered in the same subset. The closest subsets (i.e. subsets with the lowest average distance between the individuals of each subset) are grouped into bigger subsets. In other words, the distance between subsets increases with a decreasing number of agglomerate subsets. This procedure is repeated until obtaining only three clusters. Hence, the defined clusters represent sets of data showing similar apparent resistivity variations over the 9.5 years of survey (see example of cluster 1 in Fig. 6c). The dendrogram of Fig. 6b illustrates the agglomerative procedure used to constitute the clusters.

To summarize the results of the clustering method on this 9.5-year dataset of ERT monitoring, Table 1 shows the features of the three defined clusters. They correspond to specific resistivity ranges and each of them involves a different amount of measurements. The choice of number of clusters was driven by a will to simplify the most the ERT section and thus isolate broad specific resistivity behaviours.

Table 1 Summary of the clustering process. The 0.4% missing corresponds to the removed quadripoles during the filtering step
Full size table

Cluster analysis of apparent resistivities of ERT data

This section focuses on the evolution of apparent resistivity at the Vence landslide. Despite the complex link between apparent resistivity and subsurface processes, this study of raw data circumvents problems linked to data inversion such as the considerable increase of data uncertainties if the inverse problem is not adequately solved or the tendency to mask low amplitude signals owing to the smoothing effect of least square inversions (Ellis and Oldenburg 1994).

Spatial repartition and temporal variability of resistivity data

The apparent resistivities were grouped into three clusters using the method described in “Cluster analysis method” section (Fig. 6, Table 1). Looking to the spatial distribution of these three groups of resistivity ranges, we acknowledge that they correspond to specific resistivity areas in depth (Fig. 7a).

Fig. 7
figure 7

a Spatial repartition of the three defined clusters of apparent resistivity in the apparent pseudosection. b Inversed profile of the upper apparent section: the apparent resistivity values of each cluster correspond to their mean values (see Table 1)

Full size image

Moreover, constructing a simplified apparent pseudosection with three values of apparent resistivity corresponding with the mean clusters’ values of apparent resistivity (Table 1) and located as shown in Fig. 7a, we can process to an inversion of the simplified profile (Fig. 7b). The fact that this inversed resistivity profile shows broadly the same patterns as the real profiles (Fig. 5a) indicates that our clustering method of apparent resistivity is worthy.

Figure 8a shows the evolution of the daily median value of apparent resistivity of each cluster with time.

Fig. 8
figure 8

a Variations of the daily median values of the three clusters of apparent resistivity measurements for the period 2006–2016. b Accumulated rainfall index (Helmstetter and Garambois 2010) and groundwater level of C6 and C20 for the same period

Full size image

No clear link between the variations of these median values and rainfall was evidenced (Fig. 8). Indeed, it seems difficult to monitor a phenomenon usually lasting few hours with only a single ERT per day. Kim et al. (2009) showed that to properly monitor seepage/infiltration using ERT, it was necessary to consider the acquisition time of resistivity measurements to solve the inverse problem. However, the resistivity variations of the clusters show a clear correlation with the variations of the groundwater level (Fig. 8). The good correlation also observed between the two piezometers can reveal the generalized aspect the groundwater fluctuations. We then assume that the groundwater level’s variations recorded at one piezometer show the general groundwater level of the medium.

Link between apparent resistivity and groundwater level

In order to highlight distinct behaviours in depth, it was possible to plot the evolution of the daily cluster median values as a function of the daily piezometric level recorded with the C6 piezometer (Fig. 9):

  • the median resistivity values of cluster 1 covers a narrow resistivity range (about 10 Ω m of resistivity band width) and appears to be independent of the water level, but we note a slight variation from −1.5 m deep, where the values seem to increase with a decreasing piezometric level.

  • the median resistivity values of clusters 2 covers a wide range (about 30 Ω m of resistivity band width) and vary roughly with the groundwater level. However, no correlation can be detected.

  • the median resistivity values of clusters 3 covers a wider range (about 40 Ω m of resistivity band width) and seems to vary with the groundwater level with an increase of the median resistivity values when the piezometric level decreases.

Fig. 9
figure 9

Median values of apparent resistivity of the 3 clusters as a function of the groundwater level recorded at C6 (see Fig. 1b for the location), for the whole studied period

Full size image

The method applied here on apparent resistivity allows studying the spatiotemporal distribution in depth of areas with similar hydrologic behaviours. We then assume that these trends are linked to the hydrogeological properties of areas influencing the resistivity values of each cluster. Hence, we may suppose that

  • the apparent resistivity of cluster 1 is probably mainly influenced by areas with consistently high water content (e.g. saturated areas). Considering the local geology, it may correspond to the landslide body, containing highly conductive clay lenses or zones remaining under water at all times.

  • the apparent resistivity of clusters 2 and 3 suggests that their values strongly depend on areas whose resistivity decreases with increasing groundwater level. Thus, they would correspond to permeable zones located in highly resistive zones. It may correspond to well-drained zones of low porosity located beneath the groundwater level such as fractured limestone.

The dispersion of data on a larger range of resistivity for clusters 2 and 3 than for cluster 1 supports the hypothesis that these high resistivity measurements are influenced by areas whose water content varies importantly and strongly affects the measurement, increasing the electrical contrast. The water content of those areas may vary during rain infiltration through faults and fractures. Furthermore, the large variation of apparent resistivity under the same groundwater level in Fig. 9 reflect the long-term trend observed in Fig. 8 for these two clusters. Indeed, the broad decrease observed in the median values for clusters 2 and 3 implies changing values in time for the same groundwater level.

However, the dispersion of data for all clusters can be attributed to other phenomena such as the measurement noise, the slight change of resistivity due to temperature variations (the annual variation of temperature at 4 m below ground level is ±0.5 °C, for a mean annual temperature of 14.7 °C) or the effect of changes in water chemistry. Despite these drawbacks, the resistivity data provides significant information on the hydrogeological behaviour of the slope.

Finally, we found deep clusters with high resistivity that seem to depend on groundwater fluctuations. Despite the complex link between apparent resistivity and subsurface processes, this study shows that additional information can be retrieved from raw ERT data.

Lebourg et al. (2010) presented the results of a field campaign that was designed to understand the landslide local geology. Figure 10 reproduces the inversion results of ERT3 (acquisition in pole-pole array) that was acquired on the 10th of March 2009, near the actual location of the permanent ERT (see Fig. 1b for locations). Three main geological units were identified:

  • the landslide body composed of sandy-clayed material with low resistivities (mostly below 30 Ω m). The potential sliding surface is plotted as a dotted line in Fig. 9

  • the underlying limestone substratum with high resistivities (above 65 Ω m)

  • the water-filled faults affecting the limestone substratum (resistivities about 50–60 Ω m due to the smoothing effect of the inversion method)

Fig. 10
figure 10

ERT3, acquired in a pole-pole configuration, adapted from Lebourg et al. (2010) with the position of the permanent ERT (see Fig. 1b for locations). The limits of the sandy-clayey landslide is thought to affect the material above the fractured limestone substratum

Full size image

When looking at the location of the permanent ERT in Fig. 10, it appears that we highlighted different groups that correspond to different bodies with different resistivity proprieties:

  • knowing the geometry of the landslide, we associate the numerous amount of points of low resistive cluster 1 to the conductive and porous sandy-clayed sliding material, where the free water table develops. This area is assumed to be consistently saturated. This corresponds to the landslide body.

  • clusters 2 and 3 seem to describe the behaviour of the highly resistive underlying limestone substratum.

These assumptions are confirmed by the hydrological behaviours of these units. Cluster 1 varies very little, illustrating the conductive and permanently saturated state of the shallow body, while clusters 2 and 3 highlight a resistive body which resistivity varies highly with time. Taking into account its location on the apparent section, we assume that it corresponds to the fractured Jurassic limestone affected by at least one water-filled fault. The temporal variations of the apparent resistivity values of these two clusters may be explained by the water circulation through this fault under the slip surface.

Without making any assumption on the local geology, we were able to locate hydrogeological units displaying different temporal behaviours. The method presented here is the more robust when using longer monitoring period and shows that it is of high importance to study the hydrological response of a slope during pluri-annual surveys. As an example, one can see that resistivity variations are damped for the three clusters during winters (water saturation is high) and that their respective behaviours can be better set apart during summer as surface dries out. The observed changes in apparent resistivity during the dry season do not follow immediately the change of the groundwater level. Indeed, this change occurs after a certain delay that is probably characteristic of the bulk hydraulic conductivity of the medium.

Conclusions

In this paper, we studied the spatial and temporal variability of an unstable slope thanks to multi-parameters analysis, with a special focal on resistivity measurements. The aim of this time-lapse survey was to better understand fluid circulations within the unstable slope as historical records show a strong relationship between water inflows (especially heavy rainfalls) and later accelerations and reactivations of the sliding process. This study area, instrumented for 9.5 years, is a real natural experimental laboratory where we were able to develop interesting methods in the observation of complex natural system’s behaviour and improve the knowledge of such a risky zone.

Our results show that it is now possible to locate geological units displaying distinct hydrogeological behaviours without making any assumption on the local geology, using permanent ERT. The simultaneous analysis of such a large dataset allows obtaining robust results that are more characteristics of the long-term behaviour of the hydrogeological system.

The results of this work showed that thanks to a statistical analysis by agglomerative classification procedure, we were able to isolate at least three different areas in depth that react differently to water infiltration. This technique has been useful to detect and define different hydrological behaviours in the subsurface of an active landslide. We then identified a first group of low resistivity values that show very little variations, corresponding to a permanently saturated and conductive body that we interpreted as the landslide body itself. Then the two other clusters, showing higher resistivity values and higher variation with groundwater level changes, have been interpreted as a permeable body affected by water circulation. This was interpreted as the known faulted limestone underlying the landsliding mass. The location of this apparent resistivity bodies has been confirmed thanks to an averaged model section using these three clusters. Thus, this approach enables to gather qualitative information on the variability of the slope hydrogeological behaviour both spatially and temporally. This type of information is crucial to help improving the conversion of resistivity data into hydrologic quantities.

Indeed, one of the main challenges of future research using time-lapse ERT remains to translate electrical conductivity into hydrological quantities. Several researches achieved detailed studies of the complex effects of temperature, soil moisture and temporal variation of the ambient ionic concentration on the value of electrical conductivity (Waxman and Thomas 1974; Sen and Goode 1992; Rein et al. 2004; Hayley et al. 2010). Recent works all agree that the definition of petrophysical models to convert ERT measurements into hydrological measurements should be site-specific and take into account the spatial and temporal variability of the medium (among others, Singha and Gorelick 2006; Singha and Moysey 2006; Jayawickreme et al. 2010). Our work will probably help in selecting area where the hydrogeological properties are rather homogeneous and, thus, adapting petrophysical models. This might be a major step to improve the modelling of landslides where the distribution of pore pressure plays a key role on the kinematics.