Water infiltration and recharge processes in karst systems are complex and difficult to measure with conventional hydrological methods. In particular, temporarily saturated groundwater reservoirs hosted in the vadose zone can play a buffering role in water infiltration. This results from the pronounced porosity and permeability contrasts created by local karstification processes of carbonate rocks. Analyses of time-lapse 2-D geoelectrical imaging over a period of 3 years at the Rochefort Cave Laboratory (RCL) site in south Belgium highlight variable hydrodynamics in a karst vadose zone. This represents the first long-term and permanently installed electrical resistivity tomography (ERT) monitoring in a karst landscape. The collected data were compared to conventional hydrological measurements (drip discharge monitoring, soil moisture and water conductivity data sets) and a detailed structural analysis of the local geological structures providing a thorough understanding of the groundwater infiltration. Seasonal changes affect all the imaged areas leading to increases in resistivity in spring and summer attributed to enhanced evapotranspiration, whereas winter is characterised by a general decrease in resistivity associated with a groundwater recharge of the vadose zone. Three types of hydrological dynamics, corresponding to areas with distinct lithological and structural features, could be identified via changes in resistivity: (D1) upper conductive layers, associated with clay-rich soil and epikarst, showing the highest variability related to weather conditions; (D2) deeper and more resistive limestone areas, characterised by variable degrees of porosity and clay contents, hence showing more diffuse seasonal variations; and (D3) a conductive fractured zone associated with damped seasonal dynamics, while showing a great variability similar to that of the upper layers in response to rainfall events. This study provides detailed images of the sources of drip discharge spots traditionally monitored in caves and aims to support modelling approaches of karst hydrological processes.
Karst regions provide drinking water for a quarter of the world's population (Ford and Williams, 2007; Mangin, 1975). In a changing world, improving the management of vital resources is a key problem, as highlighted in Hartmann et al. (2014). Achieving enhanced management calls for a better understanding of superficial water movements, which are known to be strongly heterogeneous in karst areas. The autogenic recharge of the phreatic zone of karst aquifers is driven by water infiltration through the vadose zone (White, 2002). The thickness of this vadose zone varies from one karst system to another but is commonly described as two entities: (i) its uppermost layer, the soil joined with the so-called epikarst which is characterised by high weathering and porosity of carbonate rocks, overlaying the (ii) infiltration zone. The hydrological function of both layers differs from one type of karst to another (e.g. Mediterranean or humid, young or mature karst landscapes; Klimchouk, 2004). While rainfall can directly feed the infiltration zone through sinkholes or open cracks in the epikarst, a part of meteoric water remains delayed in the epikarst (Bakalowicz, 2005). Locally, water can be stored in perched saturated pockets because of strong permeability contrasts with regard to lower layers. Such epikarst storage was proven to be sustainable enough to host aquatic biota (Sket et al., 2004) or to induce strong dilution of rainwater isotopic signatures (Perrin et al., 2003). In some regions, especially in China, such storage in the subsurface is expected to be great enough to sustainably provide water to populations (Williams, 2008). Nevertheless, these water reservoirs are likely to be seasonally influenced and laterally heterogeneous, interacting with the soil and biosphere through evapotranspiration, while seeping under gravity, or by overflow after intense rainfall events (Clemens et al., 1999; Goldscheider and Drew, 2007; Sheffer et al., 2011). Such leakage down to the infiltration zone therefore ranges from very slow seepages within the carbonated matrix porosity to quick flows through fractures and cracks in the carbonate rocks (Atkinson, 1977; Smart and Friederich, 1987).
All models describing karst hydrology agree on the dichotomy of matrix and conduit recharge processes (Hartmann et al., 2014). Karstification is expected to act on the porosity of the bulk rock and therefore on its hydraulic conductivity (Kiraly, 2003). Permanent storage in the vadose zone, responsible for perennial dripping recorded in cave networks, has been confirmed in several case studies (e.g. Arbel et al., 2010). However, compared to the epikarst, the role of the infiltration zone itself in delaying the infiltration and potentially storing groundwater in the matrix porosity remains an open question. In dry periods, dripping with unvarying low volume discharges is only explained by infiltration via low capacity routes or perched aquifers slowly releasing water into the underlying layers (Smart and Friederich, 1987). Hartmann et al. (2013) modelled the recharge of matrix reservoirs in the vadose zone by lateral exchange with saturated conduits. This confirms the possibility for these processes to occur at several levels within the infiltration zone, making it possible for groundwater to be stored not only in the epikarst, but in several subsystems of the entire vadose zone.
To support hydrological models, investigation techniques commonly consist of tracer tests or spring flow monitoring, mainly applied to the characterisation of the saturated zone but also tested in the vadose zone for the monitoring of stalactites drip discharge (e.g. Pronk et al., 2009). In particular, such experiments can provide evidence of variable transfer types. Natural caves provide great opportunities to study the vadose zone hydrodynamics from the inside with punctual and direct measurements and/or monitoring. Hydrographs or hydrochemical monitoring are often a valuable source of information. Although novel promising approaches for building dense cave drip discharge monitoring networks are rising (e.g. Mahmud et al., 2016, 2018), strong heterogeneities of karst areas often make it challenging to build robust networks that adequately capture groundwater storage variations in the vadose zone. Karst subsurface remains poorly known and not often instrumented or monitored. In particular, very little has been achieved to image and monitor perched reservoirs.
Geophysical methods provide non-invasive and integrated tools that can strongly improve karst hydrological knowledge. Hence, numerous studies have been conducted to characterise karst subsurface (see Chalikakis et al.; 2011, for a review). In terms of hydrological monitoring, Valois et al. (2011) and Deville et al. (2012) highlighted the signal of epikarst storage variations in gravity anomalies of repeated gravity measurements. Fores (2016) supported similar measurements with seismic noise monitoring.
In parallel, ERT (electrical resistivity tomography) monitoring methods have proved to be highly efficient, especially in hydrogeophysics (e.g. Coscia et al., 2012; Kuras et al., 2009; Revil et al., 2012) and in engineering and geotechnics for monitoring landslide areas (e.g. Chambers et al., 2013; Uhlemann et al., 2016a), contaminated sites (e.g. Caterina et al., 2017; Kuras et al., 2016; LaBrecque et al., 1996a) or permafrost regions (e.g. Supper et al., 2014). The strength of such methods resides in their effectiveness to track changes in the electrical properties of the subsurface, reflecting variations in moisture content, groundwater content, temperature or chemical properties. Binley et al. (2015) identify ERT monitoring as a key technique in the advancing of hydrogeophysical methods applicable for investigating subsurface processes. A few studies have already used repeated ERT surveys to track hydrological changes in karst areas. Recently, Xu et al. (2017) investigated time-lapse ERT data to define subsurface characteristics near the Lascaux cave (France). Carrière et al. (2016) successfully used time-lapse ERT and magnetic resonance sounding (MRS) to identify the role of the porous matrix in regulating water infiltration from epikarst structures, previously identified by ground-penetrating radar (GPR) and ERT surveys in southern France (Carrière et al., 2013). Meyerhoff et al. (2012) applied repeated time-lapse ERT measurements to visualise variations in karst saturated conduits' conductivity, assessing the mixing of matrix water and surface water. In parallel, Kaufmann and Deceuster (2014) have demonstrated the applicability of using ERT to image the porous matrix associated with karstification processes. Altogether, these studies demonstrate the applicability of such techniques with regard to hydrological purposes in karst, although they spotted real challenges: the heterogeneity of the subsurface making the interpretation of resistivity models more complex and the difficulty of practically ensuring proper contacts for electrodes, especially in the presence of outcropping limestone (Chalikakis et al., 2011).
To the best of our knowledge, this paper presents the first attempt of long-term, permanently installed, and high spatial- and temporal-resolution ERT monitoring of karst subsurface hydrodynamics. Our experiment covers a 3-year monitoring period of the Rochefort site, a karst area located in south Belgium. The ERT measurements focus on a 2-D profile and comprise two sub-periods: a first 3-month period of daily ERT measurements started in April 2014 and a second 2-year series of almost uninterrupted measurements from March 2015. Additional hydrological data such as moisture probes and in-cave percolating water discharge measurements support the experiment. The monitoring site focuses on a small part of the karst area, at the entrance of the Lorette cave. Such a local-scale approach supports the need to study karst hydrology on all scales (Hartmann, 2016) to build extensive data sets available for strengthening hydrological models.
The study area is located over the central part of the Lorette cave, next to
the city of Rochefort in southern Belgium. The Lorette cave is one of several
cavities that belong to the Wamme–Lomme karst system (Marion et al., 2011),
a 10
In the Lorette cave and on a larger scale, in the Rochefort area, limestone
layers are part of an overturned syncline (Fig. 1b) comprising the Charlemont
limestone strata striking N070 with a moderate to high dipping value
of 50
The study site is part of the Rochefort Cave Laboratory (RCL)
(Camelbeeck et al., 2011; Quinif et al., 1997), located in the central
part of the Lorette cave in an underground area that covers about
1
In terms of hydrogeology, in low water conditions, the water table
shows up in the Lorette cave at
At the surface of the RCL site, a small building, located at the border of the large sinkhole, hosts the instruments data loggers and the resistivity meter. The eastern part of the site is mostly asphalted, with a parking area and two minor roads, while the rest of the area is wooded, including the sinkhole where the ERT profile is installed (Fig. 1c). The underground part of the RCL site benefits from infrastructures, such as steps and paths originally built for a former touristic exploitation of the cave in the beginning of the 20th century. Some of these infrastructures have been secured against collapse for our study.
Several environmental sensors have been installed at the RCL site:
soil moisture probes, in-cave percolating water gauges, and
rain and percolating water conductivity probes. They are intended to
support the ERT measurements. First, a vertical profile of five water
content reflectometers (WCRs) from Campbell Instruments (CS616) is
installed 2
Additionally, the Lorette cave is equipped with percolation discharge
monitoring concentrated in three specific locations. Two drip discharge
gauges are installed in the Val d'Enfer room, one of which (PWD1,
percolation water discharge station 1) monitors flows
dripping through a subvertical open fracture oriented N160 in a clayey
limestone layer, with the other one (PWD2) being installed under a karstified
area where drips come out of a particularly porous limestone layer. The third
station (PWD3) monitors one stalactite built on a massive limestone layer
associated with very slow discharge in the northernmost passage at the
vertical of the ERT profile. This area is generally much drier than the Val
d'Enfer room. The thickness between the surface and the monitored inlet
flows is
Specific electrical conductivity (SpC) measurements are also performed
in-cave at the PWD1 monitoring station, as well as at the surface for
monitoring rainwater conductivity, next to the ERT profile. Both
measurements are performed using a Campbell CS547A probe (accuracy of
Rainfall was monitored for the whole period of ERT monitoring using
a Lufft tipping bucket type rain gauge with
a 1 min sample rate, located on the RCL site itself. The locations of the
WCR profile and rain gauge are shown in Fig. 1c. Additional potential
evapotranspiration (
Figure 2 shows the rainfall and
Overall, soil moisture data inform on the dynamics of the infiltration
at the location of the vertical profile, showing repeated rainfall
infiltration processes. Every significant precipitation event
progressively infiltrates the soil layer, producing a sharp increase in
VWC followed by an exponential recession curve. The delay between the
beginning of the rainfall event and the first arrival of infiltrating
water depends on the intensity of the rainfall event,
evapotranspiration conditions, the depth of the moisture probe and
hydraulic conductivity parameters defining the soil retention
curve. In winter, a delay of
In parallel, in-cave percolating water discharge data bring crucial information on the infiltration processes occurring in the vadose zone at the RCL site. The three stations show different discharge dynamics given their location and the type of inlet flow that they sample. Smart and Friederich (1987) developed a drip discharge classification based on the relationships between maximum discharges and coefficients of variation of the discharge; they can be described as vadose flows for PWD1 and PWD2 and seepage flow for PWD3 respectively. Vadose flows refer to high discharges, albeit lower than for shaft flows, with a high variability, especially regarding their rainfall events responses. Seepage flows exhibit significantly lower discharges with low coefficients of variation but noticeable seasonal changes. Despite being classified as vadose flow, the PWD2 data set exhibits a strong seasonal pattern. It actually samples more of a dripping zone rather than one single inlet flow associated with one stalagmite or fracture. This could lead to overestimating the maximum discharge regarding the approach of Smart and Friederich (1987). PWD2 could therefore be described as seasonal drip, which differs from seepage flows by its higher coefficient of variation, following the modification of the classification after Baker (1997). The spatial proximity of PWD1 and PWD2 exhibiting different discharge regimes testifies to the high heterogeneity of the Rochefort karst.
Seasonal cycles affecting percolating water are strongly related to effective rainfall and soil moisture data, as shown in Fig. 2. Similar observations have been described and analysed in multiple studies, highlighting the buffering role of the epikarst in water infiltration (e.g. Genty and Deflandre, 1998; Poulain et al., 2015b; Sheffer et al., 2011; Aquilina et al., 2003). Arbel et al. (2010) distinguish perennial drip discharge, which explains the bottom threshold visible in PWD2 and PWD3 in summer (Fig. 2c), and seasonal drips that stop during summer and are characterised by longer recession times. Additionally, post-storm drips directly follow rainfall events and decay after a few weeks, exhibiting a high discharge variability. The first two types can be related to diffuse flow that propagates through the matrix, while the last type refers to quick flows and conduit infiltrations (Hartmann et al., 2014; Lange et al., 2010; Perrin et al., 2003). Overall, these classifications highlight the duality of water infiltration and recharge in karst systems.
Unlike PWD2 and PWD3, PWD1 does not exhibit a clear seasonal trend, even though the baseflow threshold and post-storm drip decrease during the driest periods, while longer recession curves are observed. This is especially the case for the August and September 2016 drought. Poulain et al. (2018) provides a specific analysis of the diffuse flow and quick-flow components of PWD1, supported by a vadose dye tracing test. It confirms the two-flow regime as a mixing of matrix and conduit infiltration. PWD2 and PWD3 seem to depend more on diffuse flow through the matrix but a part of quick flow is still present in the signal. In conclusion, drip discharge data reflect well their station's location: PWD1 samples inlet flows from an open fracture cross-cutting a clayey limestone layer, which explains the great quick-flow component from post-storm drip type percolation through the fracture. PWD2 monitors drip discharge from a porous limestone layer – water coming out directly from the rock matrix, without the presence of stalactites. Finally, PWD3 is installed on a dry location and samples inlet flows from a stalactite built on a massive limestone layer. This explains its very low observed perennial drip discharge described as seepage flow.
The electrical conductivity of the percolating water (Fig. 2d)
displays some variations following rainfall events and related
recharge processes, but no seasonal trends are evidenced. The observed
values average 0.25
In summary, results of the environmental monitoring of the vadose zone already bring valuable information on the infiltration processes occurring at the RCL site that will be useful for guiding the interpretation of the geophysical monitoring. Overall, hydrological seasonal trends are already discernible from these data sets, while different infiltration dynamics attributed to rainfall events are observed, illustrating the heterogeneity of the karst subsurface.
A preliminary study was necessary to assess the feasibility of ERT
monitoring at the RCL site and more specifically to define the most
appropriate location for installing the electrodes permanently. Seven
ERT surveys around the RCL site were therefore conducted in 2013,
which constituted an important step for the design of the
experiment. They resulted in identifying the sinkhole giving access to
the cave as an area with heterogeneous electrical resistivity features
likely to be of interest for monitoring complex hydrological
processes. A profile of 48 electrodes, with a pronounced topography,
was therefore installed permanently. Twenty-eight electrodes from
this profile are set at the top of the limestone massif and 20 others
along the slope of the sinkhole (Fig. 1c). Most of the electrodes are
buried 20 to 30
Two acquisition systems were installed at the RCL site. A first testing period lasted from March 2014 until June 2014. An automated time-lapse electrical resistivity tomography (ALERT) acquisition system developed by the British Geological Survey (Kuras et al., 2009) collected daily dipole–dipole (DD) measurements. After this testing period, we installed a four-channel Iris Syscal Pro resistivity meter in March 2015, which is still presently measuring. Daily multiple gradient (GD) and DD data were collected, except for summer 2015 and winter 2016 where DD arrays, which require higher injection power, suffered from battery malfunction issues. Both acquisition systems are remotely controlled from the office. Data are automatically sent to a server and checked for measurement errors. The acquisition system is installed in a brick shelter, furnished with a wired internet connection and a 230 VAC power access, providing ideal infrastructure for an ERT monitoring site.
The measurement protocols involve dipole–dipole and multiple
gradient types. They were chosen because of their effectiveness
for multichannel data acquisition purposes as well as their good image
resolution capabilities (Dahlin and Zhou, 2004). On the one hand, DD
arrays are well suited to image lateral features and allow efficient
collection of reciprocal measurements. Exchanging current and
potential electrodes should ideally deliver the same results, as
stated by the reciprocity theorem (Parasnis, 1988). Comparing forward
and reciprocal measurements provides a robust method for estimating
the data error and quality (LaBrecque et al., 1996b; Wilkinson et al.,
2012). The DD type surveys chosen in this experiment use dipole
lengths (
We developed a semi-automated workflow involving routines for data acquisition, storage, filtering, inversion and visualisation. A first data filtering is applied on the repeatability error of each measurement. During the acquisition, the potential difference on the measurement dipole of each quadrupole is measured two to four times by the resistivity meter. Distributions of the repeatability error are shown in Fig. 3a. For DD arrays, data having repeatability with a SD (repeatability or stacking error) over 5 %, as well as measured potentials lower than 1 mV, are automatically filtered. Following this step, reciprocal errors are computed for the DD type data set.
Relative frequency of
Reciprocal error is the resistance difference between normal and reciprocal measurement, i.e. when current injection and potential dipoles are swapped. Figure 3b shows the distribution of the reciprocal errors for the whole DD type data set, after filtering for repeatability errors and low potentials. Data with relative reciprocal errors over 20 % were also removed. Reciprocal errors are used as a noise estimate for the inversion procedure, where the resistance of each measurement needs to be weighted. Overall, filtering on repeatability error and reciprocal error leads to 15 % of all the DD measurements being rejected, mainly due to too-low measured potentials. GD type surveys have no reciprocal measurement available for each daily data set. A punctual reciprocity test was performed on GD arrays and showed relative reciprocal errors slightly higher than those of DD arrays. This is attributed to the fact that GD surveys have a measured resistances range significantly broader than that of the DD surveys. Furthermore, the signal-to-noise ratio shows a slightly different order of magnitude between normal and reciprocal measurements in case of GD surveys, which is expected to increase the reciprocal error. Another possible explanation for this is that real changes may occur during the GD surveys. Since the GD reciprocals take longer to measure (single channel) than the DD reciprocals, there is more time for greater changes to occur, leading to greater differences between forward and reciprocal measurements. This also explains why the DD reciprocal errors are greater than the DD repeatability error: stacking measurements are measured close together in time, but forward and reciprocal pairs are separated by larger times. Given that GD arrays have no reciprocal measurements available for the entire monitoring period, the repeatability error filtering threshold was set down to 0.5 %, which also takes into account the lower mean of the repeatability error distribution compared to that of the DD arrays, as visible in Fig. 3a. Following this filtering, only 1.5 % of GD measurements were rejected.
Contact resistances along the ERT profile also showed high temporal
variations, following the moisture conditions at the site (Fig. 4a). The
high clay content of the soil at RCL ensures a very good
electrode–ground contact in humid conditions. However, it favours
shrinking during dry periods that can reduce the surface contact of
electrodes with surrounding soil materials. Such processes therefore
increase the contact resistances that, in turn, are a source of
increased measurement errors. Higher contact resistances are usually
noticed as they produce greater repeatability errors and reciprocal
errors. In dry periods, this leads to a higher number of rejected data
after filtering. In August 2016, electrode #12, placed in the
middle of the slope of the sinkhole, started to show significantly bad
contact resistances (
In December 2015, despite the fact that the monitoring site was equipped with AC power supply, the injection batteries started to fail because of the increased power demanded by the resistivity meter to deliver higher voltages, especially for DD surveys. The battery malfunction led to the rejection of almost all the DD surveys acquired during that period. GD surveys were less sensitive to the problem because the power and voltage requirements were lower on average. The greater percentage of rejected data in 2015 is also attributed to the batteries being slightly less efficient for repeated high-voltage injections. The original starting type batteries were replaced by deep-cycle gel batteries in July 2016, especially designed for deep discharge, similar to those frequently used with solar panels. The long drying period from August to October 2016 is responsible for a progressive increase in the amount of rejected data for both DD and GD surveys.
Temperature variations of the subsurface can have significant impacts on
resistivity data (Brunet et al., 2010). Because a marked temperature gradient
is expected in the sinkhole, we modelled the 2-D temperature field using the
framework of pyGIMLi (Rücker et al., 2017;
Data are inverted with Boundless Electrical Resistivity Tomography software
(BERT), which is based on a finite element modelling (Günther
et al., 2006; Rücker et al., 2006). Each of the DD and GD data
sets after filtering and correction for temperature effects
constitutes one data set for the inversion, whereas the inverted
results of the first data set of the DD and GD series constitute the
reference model for the whole time series. The inversion is carried
out using a
BERT provides a coverage parameter calculated from the sum of the absolute sensitivities of the measurements. Areas with high coverage are better constrained by the measurements and the modelling choices than low coverage areas. The coverage is usually incorporated in the data visualisation as a transparency mask that is used to weight the data depending on their associated coverage values.
The inverted resistivity images of DD and GD arrays show a great
dispersion. Images from DD arrays have a better coverage at depth
than those of GD arrays. Hence, they are able to identify deeper
structures, while GD arrays have a better resolution at shallow
depth. In such a karst context, massive limestone layers are expected
to be highly resistive
(
Image of the inverted resistivities for DD
Field observations corroborate the presence of the conductive layer on the
plateau, as this area is characterised by a
The high resistivity values may correspond to those of reasonably non-karstified limestone. Intermediate values may indicate karstified limestone with a low clay or moisture content. Such values could also result from clay-rich interbeds between limestone strata. Given the resolution of the ERT image, such thin features cannot be precisely delineated. In the case of a highly conductive thin shape surrounded by highly resistive materials, the inversion process could only build a larger conductive area with an intermediate resistivity value. The strongly dipping conductive feature that crosses the resistivity image in the middle also deserves discussion. It could represent a fractured zone either filled with clays or characterised by a higher moisture content than the surrounding materials. The time-lapse imaging will give some insights on the role of this part of the section in hydrogeological processes taking place in the near surface by highlighting changes in resistivity through the year.
There are several ways to visualise spatiotemporal changes in resistivity resulting from a time-lapse ERT inversion, such as computing resistivity ratios, log of resistivity ratios or percentage of change in resistivity, or log of resistivity. The basis on which the time-lapse comparison is made must also be appropriately assessed. It may involve visualisation of resistivity variations that occur between each data set, highlighting sharp resistivity changes. Lower frequency resistivity variations usually need to be visualised as changes with respect to a baseline. Such a baseline does not necessarily have to be the same as the reference model used in the time-lapse inversion process. In many studies, the baseline used to track resistivity changes consists of resistivity measurements performed before the process to be monitored begins. In some cases, this stage is easy to assess. For instance, when monitoring artificial water injection, the baseline commonly consists of an ERT survey made prior to the start of the injection (e.g. Robert et al., 2012). In other cases, including this one, a clear baseline is more difficult to assess because the beginning of the process of interest, i.e. recharge processes, is not clearly identifiable. Here, the seasonality of the vadose zone moisture conditions (Fig. 2) complicates the choice of a clear baseline to visualise the time-lapse resistivity images.
As this study focuses on groundwater recharge, a baseline in dry
conditions was chosen. The long drought of August to mid-October 2016
during which WCR moisture measurements dropped dramatically (Fig. 2)
corresponds to the driest conditions encountered during the monitoring
period. ERT surveys of 15 October 2016 corresponding to the last day
of this drought were therefore chosen as baselines (
Our preference was for change in log of resistivity because it gives a better overview of gradual spatial variation than changes in absolute resistivity.
Figure 6b shows changes in log resistivity for 10 of the 467 DD data
sets (see the Supplement for an animation of the resistivity
variations for the whole time series of the DD and GD data
sets). Although this kind of display highlights broad variations in
resistivity, it assumes that the nature of the underground materials
would give comparable resistivity changes if exposed to similar
moisture conditions. This is usually the case in fairly homogeneous
environments, whereas it is not necessarily valid in highly
heterogeneous contexts. Such visualisation could therefore overlook
subtle changes in regions associated with resistivity variations of
smaller amplitudes. To address that problem, our approach consists of
displaying images of normalised resistivity (
Overall, changes in log resistivity compared to the driest conditions
encountered at the surface during the whole monitoring period show
that the majority of the variation in resistivity is negative. This
means that most of the subsurface experienced a decrease in
resistivity, consistent with an increase in the moisture. The area
most affected by intense (
Below this surface layer, changes in log resistivity progressively
attenuate, in the area of the higher resistivities imaged in Fig. 5,
except for a circular shape discernible in the middle part of the
plateau (centred around 5
The changes in the surface layer of the sinkhole follow the same
dynamic as that of the surface of the plateau, except that it shows
less intense log resistivity decrease (
Figure 6c comprises images of the normalised resistivities for the
same time steps as those of Fig. 6b. In addition to providing images with an
enhanced contrast, it allows clarification of the temporal signature
of some regions of the image, especially the deeper ones. For example,
the strongly dipping feature has a relatively low resistivity in the
image of 24 July 2015 (Fig. 6b-3) that equals negative changes in
surface layers of both the sinkhole and the plateau. When comparing
that information with the normalised resistivity image (Fig. 6c-3), it
appears that this strongly dipping feature displays normalised
resistivity values (
In conclusion, this highlights the fact that several subsurface regions of the ERT profile experience pronounced changes in resistivity through the monitoring experiment period. These regions have their own resistivity signatures, as illustrated in Fig. 5, which testifies to the substantial heterogeneity inherent to karst systems. To simplify the observations, Fig. 7 clusters the recovered resistivities of the subsurface of the ERT profile in eight distinct regions that display different spatial and temporal resistivity signatures, based on their average resistivity values and arbitrary thresholds. It differentiates superficial layers (1, 2, 3 in Fig. 7) and deep regions (4, 5, 6, 7, 8 in Fig. 7). Interestingly, these regions display different dynamics in terms of the temporal resistivity evolution in Fig. 6. In particular, dynamics of the surface layers look dissociated from those of three deeper regions: the moderately resistive circular feature in the middle of the plateau (5), the deep layer in the slope of the sinkhole area (7) and the high dipping feature (8). The rest of the image does not show much noticeable change (6) except low variations in the rest of the deep parts of the plateau in the right side of the image (4). Table 1 summarises the temporal evolution of each region through time, highlighting their distinct characteristics.
Subdivision of the ERT image in eight distinct regions based on their average resistivity values and arbitrary thresholds. Table 1 presents characteristics of the temporal evolution associated with each region.
Statistics of the temporal evolution of the median of the absolute resistivity values for the eight regions defined in Fig. 7. For the values of each region, the table shows the temporal mean, median, standard deviation (SD) and SD normalised to the temporal mean resistivity. GD data sets are used for superficial regions (1, 2, 3 in Fig. 7) given the better resolution of that protocol at shallow depths. Resistivity values of the deeper regions (4, 5, 6, 7, 8) come from the DD surveys, because of the greater imaging resolution at depth.
As bedding planes, open joints, fractures and small conduits play an
important role in water infiltration in karst systems, we investigated
the local geological structures of the limestone massif, based on
previous works from Vandycke and Quinif (2001) and Camelbeeck
et al. (2011). These authors reported (Vandycke and Quinif, 2001) and
monitored (Camelbeeck et al., 2011) active faults striking N070
(subparallel to the bedding foliations in the Lorette cave) with a recent
normal slip combined with a small sinistral strike-slip component (see
Fig. 1b). The relatively small scale of the monitoring area makes it
relevant to study in detail the geological structure and lithology of
the site. New observations from multiple sources have been gathered to
build a lithological model of the monitoring site, comprising a field
survey and the acquisition of a 3-D model of the RCL's main chamber,
the Val d'Enfer room. This 3-D model results from a drone
photoscan of the cavity and allows automatic detection of the
orientations of planar structures and hence a statistical analysis of
the main geological structures (i.e. sedimentary layers,
fractures and joints, and faults) as well as a precise lithostratigraphical
log of inaccessible outcrops from the cave's roof (Triantafyllou
et al., 2016). The Val d'Enfer room, which starts
20
The RCL site was also drilled 2
In the absence of faults with significant offset, the sedimentary layers crossed by the borehole, added to those visible in the Val d'Enfer room, comprise all the geological strata directly sampled by the ERT permanent profile. Although the borehole and field surveys in the cave provide direct observations of the geological bodies, they cannot inform about possible lateral variations in terms of karstification processes and local porosity, which are likely to vary vertically and laterally in such a karst environment. However, field observations suggest that the lithological nature of each layer observed in the borehole or in the cave is expected to remain constant.
Figure 8 summarises the structural observations of the Val
d'Enfer room (a) and the imaged borehole (c). On the southern side
of the Val d'Enfer room, massive limestone layers (46 to 49
in Fig. 8) correspond to the strata on top of which the ERT monitoring
profile is installed in the slope of the sinkhole. A succession of
thin clayey limestone (layers 31 to 45) is visible just below. This
clay-rich layer is associated with higher percolation discharges than
in the rest of the room. This pile of clayey limestone can be
simplified as two main clayey layers separated by a more massive
limestone, as drawn in Fig. 8c. Underlying layers (29 to 16) show
a remarkable homogeneity in terms of lithology or weathering
rate. They correspond to the first layers cross-cut by the borehole,
where the first 3.4
Constraining the geometry of discontinuities (i.e. joints and
sedimentary beddings) is crucial to understand the dynamic of the
local water infiltration. Some of the encountered open joints are 2 to
3
In the 2-D geological model of Fig. 8c, two main open fractures with respective orientation N070–N60 and N300–NE80 identified in the porous layers (15 to 13) are extrapolated to the surface, as they may play a major role in the water percolation. Figure 8 also highlights the fact that the drip discharge gauges do not directly collect percolating water from the layers below which the electrodes of the ERT profile are installed.
As presented in Sect. 4.3, the highest recovered resistivities are
around 6000
The structural model presented in Fig. 8 is an important source of
information in order to further investigate the spatial resistivity
distribution. It can be summarised as a series of massive limestone
strata that includes a pile of clayey limestone layers next to the
slope of the sinkhole, porous limestone strata with a greater fracture
intensity in the middle and massive limestone layers interbedded with
three thin porous limestone strata in the northern part of the ERT
profile (Fig. 8). All these strata are dipping
Results of the forward modelling of the resistivity model
ERT forward modelling provides a useful tool to verify structural
hypotheses, to test their electrical resistivity response and hence to
guide the interpretation. In an attempt to account for both the
structural information (Fig. 8c) and the ERT results (Fig. 5), the
karst underground has been segmented in six resistivity regions
(Fig. 9a). The lithological pieces of information are converted into
resistivity values using assumptions chosen to best fit the observed
data. The fractured area (
Based on the resistivity model, the potential of each quadrupole of a given protocol is computed by forward modelling, resulting in a synthetic apparent resistivity data set. This synthetic data set can therefore be inverted and the resulting ERT model can be compared to that produced with observed data. Synthetic apparent resistivities are computed using BERT forward model for full DD and GD protocols (i.e. without removing electrode #12) together with a randomly distributed noise level of 5 %. The synthetic data sets are inverted afterwards with an error model defined as the mean error distribution used for the inversion of the whole DD and GD field data time series.
Figure 9b and c present the results of the DD and GD protocols
respectively. The first remark is that these reconstructed images look
very similar to the inverted field results presented in Fig. 5. The
sensitivity of the GD protocol does not allow proper recovery of the
fracture anomaly while the DD survey successfully images
it. Nonetheless, the thickness and the original resistivity value
(200
More interestingly, with a model value of 600
Overall, without taking into account an epikarst layer that is thought to explain the small irregular conductive anomalies in the northern side of the inverted field data results, the inversion of these synthetic data sets explains the resistivity distribution substantially. Pearson's correlation coefficients of 0.74 and 0.79, for the DD and GD surveys respectively, were also computed to compare these synthetic models with inverted models of March 2017 (Fig. 5), representative of average moisture conditions at the RCL site. Such correlation coefficients support the visual similarities between the synthetic and observed images. In the upper layer of the measured data sets, areas enlarging the thin conductive layer simulated in Fig. 9 are therefore interpreted as a signature of epikarst features.
Figure 10 summarises the interpretation of the spatial resistivity
distribution made in the light of structural observations and ERT
forward modelling. The eight regions of Fig. 7 can now be named as
follows: soil plateau (1), epikarst plateau (2), epikarst
sinkhole (3), matrix plateau (4), porous limestone (5), massive
limestone (6), clayey limestone (7) and fracture (8). Note that the
highly conductive anomaly near the bottom of the sinkhole is
interpreted as the beginning of a fractured area dipping 30
In a limestone context such as the RCL site, low resistivities indicate either clay-rich areas or porous areas in humid conditions. However, only the moisture is subject to change on an annual basis. With that in mind, the resistivity variations of the eight regions defined in Fig. 7 can be tracked. The fact that the resistivity values of these regions seem to evolve distinctly regarding climate conditions points to different hydrodynamic behaviours coexisting very close to one another, in agreement with the percolating water discharge data sets. Figure 11 shows the temporal evolution of the median of the absolute resistivity values in the eight regions. GD data sets were chosen for superficial regions (1, 2, 3 in Fig. 7) given the better resolution of that protocol at shallow depths. Resistivity values of the deeper regions (4, 5, 6, 7, 8) come from the DD surveys, because of the greater imaging resolution at depth and especially the inability of GD arrays to properly image the fracture anomaly. It results in fewer but better constrained data points. To compensate for the lack of DD surveys, especially during winter 2016, corresponding GD data sets are added with a transparent mask as a guide in Fig. 11. Note that resistivity values of the surface layers from DD surveys and their temporal variations compare well with those of the GD surveys.
Superposition of the geological model from Fig. 2 on ERT results from March 2017, highlighting the interpretation of the resistivity distribution and anomalies. Numbers of the resistivity clusters defined in Fig. 7 are added for legibility.
Resistivity time series for the eight regions detailed in Fig. 7, expressed as the median of the log resistivity in each region. The median was chosen, rather than the mean, to limit the contribution of extreme values not representative of the robust central tendency of the cells of each region, especially at their boundaries. The dry and wet periods described in Fig. 3 are also included.
The first point to stress is that, as already visible in Fig. 6, all
the regions experience a seasonal resistivity variation that seems
related to the effective rainfall distribution. The time series
displays two annual cycles: April 2015 to March 2017. Data from 2014
comprise one survey in February, another one in March and a daily
series from April to June, which confirm the relatively low
resistivity conditions attributed to the end of winter and the
beginning of spring. This correlates well with the positive monthly
effective rainfalls that are responsible for a general recharge of the
soil moisture and groundwater reservoirs, hence resulting in lower
resistivity values. A general increase in resistivity is noted when
the effective rainfalls become negative in spring, with slight delays
in deeper regions. The slope of this increase varies by several orders
of magnitude from one region to another – superficial regions showing
the sharpest increases, especially on the plateau side of the ERT
profile. The massive limestone region displays a seasonal variation
with the highest absolute amplitude (note that Fig. 11 displays the
log of resistivity), which is associated with the highest average
resistivity. The area that shows the lowest seasonal amplitude is the
clayey limestone region, with only 220
Lower seasonal amplitudes have two possible causes: higher minimum or
lower maximum groundwater contents reached in summer or winter
respectively. In similar lithological compositions, if some regions
experience lower groundwater deficits, this would result in lower
resistivity variations. However, these regions exhibiting such low
resistivity variations are also those characterised by low resistivity
distribution (
Converting resistivity variations to groundwater content changes is
one particular advantage of ERT monitoring. However, this requires
specific site characteristics in terms of subsurface homogeneity that
a complex karst system may not offer. Figure 12 presents the
co-evolution between data of the 10
Relationship between the resistivity (median) of the surface
limestone region and the saturation from the 10
As shown in Fig. 2d, the specific conductivity of the in-cave drip
discharge shows a quite stable behaviour
(
Therefore, Fig. 12 illustrates the complexity inherent to the estimation of water content from ERT data even if interesting interpretations may come out of case studies that assume constant pore-water conductivity (e.g. Beff et al., 2013; Brunet et al., 2010; Chambers et al., 2014; Garré et al., 2011; Michot et al., 2003). Such assumption can clearly not be the stated here. Likewise, Uhlemann et al. (2016b) point out similar limitations, attributing abnormal resistivity changes in wetlands to pore-water conductivity variations, which also results in the inapplicability of converting bulk electrical resistivity to moisture contents.
In such a heterogeneous karst context as the RCL site, other important limitations for estimating groundwater contents derived from ERT data concern the porosity, the clay content and the calibration of fitting parameters of petrophysical relationships (Archie, 1942; Revil et al., 1998; Rhoades et al., 1976; Waxman and Smits, 1968). Additionally, absolute electrical resistivity values imaged after inversion, especially for high resistivities, remain dependent on inversion parameters and resistivity contrasts, which mitigate the accuracy of the results. In the absence of precise calibration factors, determining groundwater contents from ERT measurements, even time-lapse, remains highly challenging in a karst environment. Nonetheless, the identification of variable dynamics attributed to groundwater content changes in different spatially limited areas of the subsurface may help in developing hydrological models applied to the vadose zone of complex karst systems.
To be able to qualitatively analyse the temporal resistivity behaviour
of each defined region as a characteristic of their hydrological
dynamic, Fig. 13c and d present the median of difference in log
resistivities with respect to dry conditions (15 October 2016), as
defined in Eq. (
Effective rainfall and data of the 10
Superficial regions on the plateau side of the profile, attributed to
the soil and epikarst, are very dynamic and hence are members of the D1
type. This is in accordance with Klimchouk (2004), defining the
epikarst as a dynamic system. The sharp decrease following rainfall
and progressive increases during every period without significant rain
testifies to the strong hydrological relationship with the atmosphere,
i.e. precipitation and evapotranspiration. The soil and epikarst
layers have their moisture contents directly monitored with the WCR
probes. A decrease in resistivity is noticed following rainfall
events, usually for effective rainfall greater than 2
The relationship between the WCR data and the fracture drip discharge (PWD1), already highlighted in Fig. 2, may therefore be extended to the entire soil layer of the RCL site. The subsurface layer must be characterised by a lower porosity and clay content given its higher average resistivity as evidenced by Fig. 11. However, it shows a dynamic very comparable to that of the surface layer. This highlights the great hydraulic connectivity between the soil layer and the epikarst and hence is typically seen as the principal reservoir that feeds vadose flows (Arbel et al., 2010), in this case the fracture drip discharge station (PWD1). In the presence of a thin soil layer, as it is the case at the RCL site, underlying materials, e.g. the epikarst, must act as a reservoir for a considerable amount of groundwater available for the vegetation (Williams, 2008). However, the fact that the average resistivity progressively increases from the surface to the subsurface layer must also be interpreted in the light of the smoothness constraint of the inversion.
The surface of the slope of the sinkhole also displays a D1-type dynamic, albeit more damped. Sharp decreases in resistivity are still observed following major rainfall events, but the seasonal dynamic is less marked, with a less clear correlation with the WCR probes. This is most likely due to (i) runoff processes occurring in this zone given the strong topography or (ii) to a very poor capacity in terms of water retention because of the progressively thinner soil layer. The fact that no specific resistivity variations are noticed in Fig. 6, next to the first five electrodes at the bottom of the sinkhole, is in agreement with the absence of soil layer in that area and the subvertical topography where surface runoff is likely to be predominant. This justified why this area was not included in the surface region tracked in the slope of the sinkhole as displayed in Figs. 7, 11 and 13.
D2 dynamics are attributed to the deeper regions comprising the clayey
limestone, porous limestone, massive limestone and the matrix to the
north of the plateau. The three latter regions exhibit a very similar
seasonal variability – the massive and porous limestone regions
reaching their maximum approximately at the same time,
A delay is also noticeable at the beginning of the resistivity
increase, especially during the 2016 dry period, in the porous
limestone and more strongly in the massive and clayey limestone
curves. This lag period reaches 15 days in August 2016. Such trends
attest that these deeper regions stay less affected by the surface
conditions in the beginning of drying processes. This indicates
delayed infiltration in deeper areas. Then the rise in resistivity is
significant only after the upper layers reach
The fractured region also exhibits a dynamic close to the D2 type, yet on which a greater variability in response to rainfall events is superimposed. Such resistivity decreases are different from the response of surface layers that typically shows a sharp decrease depending on rainfall intensity followed by a slower increase, which corresponds to soil drying processes evidenced by the WCR data. In the fractured region, the resistivity perturbation induced by rainfalls is more ephemeral, with the resistivity curves retrieving most of the time the resistivity value prior to precipitation 1 day after the rainfall event, which mimics the sharp recession curves following rainfall events displayed by PWD1 data.
Figure 14 particularly focuses on ERT data on the rainfall event
scale, showing cross-correlation functions of the eight resistivity
regions with the environmental variables on a lag window of
Cross-correlation of the ERT data sets of the eight regions defined in Fig. 7 and environmental data sets.
As a result, most of the correlation peaks occur at lag 0 or lag 1 day. No
significant peaks are noticed at negative lags, except for the clayey
limestone region that exhibits a behaviour totally different from the rest.
This will be discussed further in this section. For other regions, no
interpretation can therefore be drawn concerning variations of resistivity
that could precede variations of the environmental variables. D1-type regions
exhibit very similar patterns, being unsurprisingly most negatively
correlated either to the rainfall, the 10
In contrast to this, D2-type regions exhibit patterns different from
each other and from those of D1-type regions. The matrix plateau and
porous limestone region evolve quite similarly with
a cross-correlation peak at 1 day lag for the rainfall, 10
The porous limestone and rainfall cross-correlation coefficient decreases progressively, being still negatively correlated at positive lag 2 and 3 days. The resistivity of the porous limestone still decreasing 2 to 3 days after a rainfall event is a possible explanation, highlighting a likely diffusion of the rainwater infiltration within the porous limestone layer. Despite being associated with the D2-type regions on a seasonal basis, the massive limestone does not show any clear correlation with any of the environmental data on the rainfall event scale. This is likely due to the very low to zero rainwater contribution to the bulk resistivity of that area given its low porosity and fracture density.
Finally, the clayey limestone shows a quite surprising trend, being
positively correlated at lag 0 day with most of the environmental
data. In other words, increases in resistivity are concurrent with
rainfall events. This is also particularly visible in Fig. 6b-5 and
b-10. We propose two possible explanations of this observation:
(i) a time-lapse ERT inversion artefact and (ii) an influx of rainwater
more resistive than the actual pore water.
The first hypothesis pointing out time-lapse ERT inversion
artefacts was already addressed in the case of shallow infiltration
monitoring by Clément et al. (2009). They demonstrated using
synthetic models that infiltration processes at shallow depths
usually produce a decrease in resistivity in the upper layer, while
an increase in resistivity may be observed at intermediate depths,
where the resistivity is actually not changing. Descloitres
et al. (2008) identified such time-lapse artefacts when tracking
seasonal recharge processes by ERT, especially located in the
subsurface of slopes. These studies point out that infiltration
through a succession of layers with different resistivity signatures
may enhance such artefacts. Hence, the relatively complex
resistivity distribution in the subsurface of the slope of the
sinkhole may favour such artefacts, as the conductive clayey
limestone lies below a more resistive layer, which is itself below
the surface conductive layer. According to Clément
et al. (2009), a proposed solution to avoid such artefacts involves
adding decoupling constraints in the inversion along a shallow line
evolving together with the infiltration front, determined by
external data. This seems in our case quite unrealistic
given that the evolution of the infiltration front remains unknown
in the clayey limestone. The plateau is likely to provide a major
part of the water infiltrating the clayey limestone, given the
strong runoff processes expected in the slope of the
sinkhole. Hence, a minor part of infiltrating water is still
expected to come from the sinkhole itself, especially through open
fractures. The hydraulic conductivity is also constrained by the
lithological nature of each layer, which complicates the
problem. Therefore, the complex infiltration in the clayey limestone
justifies not adding such a decoupling line in the
inversion. Alternatively, artefacts could also result from the
underestimation of the noise (LaBrecque et al., 1996b) when
weighting the observed data of each data set for the inversion. The
use of reciprocal errors for the DD surveys (used for tracing
changes in the clayey limestone in Figs. 11, 13 and 14) must
mitigate this possibility. The latter proposed hypothesis regarding these increases
in resistivity following rainfall events in clayey limestone must
also be discussed. In some cases, an influx of a great quantity of
fresh, hence less conductive, water into a partially wet clayey
material can result in an increase in resistivity, especially given
the inverse power relationship between the bulk electrical
resistivity and the saturation (see Fig. 12), as evidenced in the
model proposed by Waxman and Smits (1968). The only option being
that the clayey limestone stays constantly at a high saturation
rate, with highly conductive pore water (
Overall, the observations concerning the resistivity dynamics of each region are determinant, as they provide an image of the sources of the drip discharge type measured in the cave system. In particular, the nature of each region and its variability in terms of resistivity must be interpreted in conjunction with its likely contribution to the karst hydrodynamic at the RCL site. Firstly, the role of the soil and epikarst is evidenced as being very dynamic (D1) with regard to the atmospheric inputs. This reactivity indicates a limited buffering role for the epikarst in terms of hydrological recharge, as evidenced by previous studies (Aquilina et al., 2003; Bakalowicz, 2005; Ford and Williams, 2008; Genty and Deflandre, 1998; Poulain et al., 2015b; Sheffer et al., 2011). Indeed, the likely very thin epikarst in the middle of the plateau as pointed out by the resistivity images and other field observations strongly limits water storage. Based on the resistivity variations, the epikarst, thicker to the north of the plateau, acts anyway as a buffer during spring, being still a likely source of diffuse infiltration to deeper areas while responding to the water demands of the vegetation. Storms cause ephemeral recharge in these surface layers but also increase rapid infiltration to deeper areas as revealed by the dynamics of the fractured area following rainfall events, even in summer. This second effect seems damped during the summer.
Diffuse flows propagating through the rock matrix are also evidenced with D2 dynamics, which are consistent with previous findings (e.g. Lange et al., 2010; Perrin et al., 2003). Carrière et al. (2016) particularly highlighted via ERT monitoring of rainfall events that considerable amounts of water can be temporarily stored in the vadose zone of karst systems and more specifically in the porous matrix. At the RCL site, the fact that all regions reach a minimum threshold in winter periods is also interesting, meaning that they approach their maximum water retention capacity. The relatively weaker effects of winter rainfalls on the resistivity variations compared to those of summer storms testifies to the high saturation present in the rock matrix during such periods. Given the inverse power relationship between saturation and bulk electrical resistivity (Archie, 1942; Revil et al., 1998; Rhoades et al., 1976; Waxman and Smits, 1968) as illustrated by Fig. 12, progressive increases in saturation are expected to result in lower and lower resistivity decreases, as for the drying period of August and September 2016. This means that in relatively porous rock, such as the porous limestone at the RCL site, an important recharge occurs in winter. Due to the aforementioned limitations in accurately estimating groundwater contents, the intensity of the deficit, when the maximum of resistivity changes is reached, remains unconstrained.
Schematic view of the hydrological processes occurring in the subsurface of RCL site.
The role of the conduit porosity is also clearly evidenced by the greater variability in the fractured region, compared to that of the massive limestone and porous limestone. D3 dynamics can be interpreted as an evidence of the direct hydraulic connectivity between surface layers and post-storm drip discharge of the percolating water (Arbel et al., 2010) as monitored in-cave. Furthermore, the seasonal variation revealed in the fractured zone is an indicator of the likely water exchange from the conduits to the matrix porosity in that area. Enhanced porosity near fracture walls as seen in core samples can temporarily increase stored groundwater. When the fractures are saturated with percolating water after rainfall events, a gradient towards these porous areas may occur, as already modelled in previous studies (Bailly-Comte et al., 2010; Hartmann et al., 2013). Similarly, an input of water from porous areas close to the fracture walls or from cross-cut porous layers is likely to occur in dry periods, participating to the vadose baseflow observed in PWD1, as Poulain et al. (2018) found at the RCL site. Alternatively, a slower infiltration can also occur in narrower fractures or in bottlenecks towards the main fractures. Figure 15 summarises these interpretations and the hydrological functions of each of the regions imaged by the ERT monitoring at RCL site.
This first long-term permanent ERT monitoring of a complex karst vadose zone has revealed seasonal recharge processes with variable dynamics. ERT allowed clustering of distinct areas showing contrasting evolutions through three hydrological cycles. Such different behaviours are attributed to distinct categories of vadose reservoirs responsible for specific percolation types. They could be associated with the sources of distinct percolation drip discharge measured in-cave. This study was able to differentiate three groups with distinct resistivity variations and to link them to sources of in-cave drip discharge:
Upper conductive layers, comprising the soil and the epikarst,
which are in direct contact with the atmosphere, hence showing the
highest variability (D1). Deeper regions characterised by a more diffuse and damped
seasonal variation, showing a delayed recharge (D2). The resistivity
values of these areas depend on the lithology, the porosity and the
clay content, which also determine their groundwater retention
capacity. A conductive fractured zone (D3) exhibiting a dynamic close to
that of D2 group, but with a greater variability in response to
rainfall events, revealing a preferential pathway for
rainwater. Water exchange between conduits and porous matrix is
likely to explain the seasonal variation.
These observations are consistent with previous knowledge of hydrological processes in the karst vadose zone (Ford and Williams, 2008; Goldscheider and Drew, 2007), while bringing a first detailed view of the sources responsible for the duality of flows typically observed in karst environments: quick flows and diffuse flows (White, 2002). Moreover, given the small resistivity variations measured in winter, the recharge processes in all areas of the monitored site are expected to be highly efficient. The main constraint on the amount of groundwater volumes stored in the vadose zone appears to be the matrix porosity.
Hydrogeophysical experiments in karst systems, especially targeting the vadose zone, are very challenging, as already raised by Chalikakis et al. (2011). This study proves that, combined with a detailed structural and lithological survey on a local scale, as well as with additional environmental measurements, ERT monitoring is able to image and track through time recharge processes within the vadose zone of a karst system.
Improved ERT inversion strategies for highly heterogeneous environments could provide more constrained results reducing the occurrence of artefacts similar to those experienced in the clayey limestone of our field site. Synthetic modelling approaches could also help in validating the assumptions made on the water infiltration processes. In parallel, 3-D imaging would improve spatial resolution, hence reducing uncertainties as compared to 2-D imaging. In such case, a real effort should be made for optimising more complex measuring protocols with regard to highly resistive karst environments. Clustering tools could also improve the detection of structures in the ERT images, as proposed by Xu et al. (2017). In such cases, fully taking advantage of the time-lapse information, i.e. distinct resistivity dynamics, seems the most important but challenging aspect.
Combining these techniques with other geophysical methods would also be definitely interesting, e.g. to image hydrological characteristics of the subsurface on larger scales. This is especially the case of passive seismic noise monitoring networks that have recently proved their applicability to track groundwater content variations at several depths (Lecocq et al., 2017; Voisin et al., 2016), even in karst (Fores, 2016).
Overall, these findings support existing models and bring new opportunities to understand the karst system, often modelled as black boxes. Imaging the sources of drip discharge signals conventionally monitored in numerous cave systems contributes to improve the understanding of karst subsurface hydrodynamic behaviours. In particular, the joined analysis between time-lapse ERT results and percolating water measurements is definitely a novel, promising approach to investigate the sources of distinct in-cave flow types and their lithological and structural constraints. More specifically, this study calls for similar geophysical monitoring to be tested in different karst environments, in combination with conventional hydrological monitoring or dense monitoring networks of cave drip water (Mahmud et al., 2016, 2018), to gather new types of hydrological data to be included in karst hydrological modelling, such as lumped karst modelling of vadose zone infiltration processes. Such novel approaches are required to face future challenges for the management of karst groundwater resources worldwide and the increasing risks of contamination issues raised by the increasing agricultural demands. Better constraining recharge processes of karst aquifers also brings grist to the mill of the study of speleothems, with implications on paleoclimatic researches.
Data used in this research paper (comprising ERT,
cave drip discharge, soil moisture, rainfall and water conductivity)
are available at
Photos of the ERT profile, with the trench at the top of the
plateau
The authors declare that they have no conflict of interest.
This work is part of the Karst Aquifer Research by Geophysics
(KARAG) project (