Time-lapse cross-hole electrical resistivity tomography (CHERT) for monitoring seawater intrusion dynamics in a Mediterranean aquifer

. Surface electrical resistivity tomography (ERT) is a widely used tool to study seawater intrusion (SWI). It is noninvasive and offers a high spatial coverage at a low cost, but it is strongly affected by decreasing resolution with depth. We conjecture that the use of CHERT (cross-hole ERT) can partly overcome these resolution limitations since the electrodes are placed at depth, which implies that the model resolution does not decrease in the zone of interest. The objective of this study is to evaluate the CHERT for imaging the SWI and monitoring its dynamics at the Argentona site, a well-instrumented ﬁeld 5 site of a coastal alluvial aquifer located 40 km NE of Barcelona. To do so, we installed permanent electrodes around boreholes attached to the PVC pipes to perform time-lapse monitoring of the SWI on a transect perpendicular to the coastline. After two years of monitoring, we observe variability of SWI at different time scales: (1) natural seasonal variations and aquifer salinization that we attribute to long-term drought and (2) short-term ﬂuctuations due to sea storms or ﬂooding in the nearby stream during heavy rain events. The spatial imaging of bulk electrical conductivity allows us to explain non-trivial salinity 10 proﬁles in open boreholes (step-wise proﬁles really reﬂect the presence of fresh water at depth). By comparing CHERT results with traditional in situ measurements such as electrical conductivity of water samples and bulk electrical conductivity from induction logs, we conclude that CHERT is a reliable and cost-effective imaging tool for monitoring SWI dynamics.


CHERT experimental setup
The objectives of the CHERT experiment are to image the SWI in order to improve the geological conceptual model, and to infer the SWI dynamics. Stainless-steel mesh electrodes were permanently attached to the outside of the seven deepest PVC piezometers ( Figure 2a). The corrosive nature of saline environments causes the limited lifetime of the installation to be a main concern when planning the monitoring experiments. 90 Metal corrosion due to exposure to saltwater is expected always, but electrolysis due to current injection accelerates the corrosion process. Current is injected using cables. The parts most sensitive to corrosion are the connection points between the cables and the mesh electrodes. The ERT instrument was tested in the laboratory before it was employed in the field to determine the best strategy to delay corrosion at the connection points. The final prototype has the mesh and the cable tied together, with the connection point covered by a double silicone layer to prevent contact with water. The electrodes showed 95 signs of corrosion after 500 hours of full contact with saline water (55 mS/cm), under a constant current injection of 1A at a frequency of 3 Hz. When conducting a CHERT, the injected current is less than 1 A and the time of injection is a fraction of a second. Based on these laboratory test results, it was suggested that the instrumentation would last for at least two years, which was the minimum desired duration of the experiment. Details on the set-up and installation are described by Folch et al. (2019). 100 When performing ERT, we measure an "apparent" resistivity that is dependent on the geometry of the acquisition. The apparent resistivity is related to measured voltage differences through: where ρ app is apparent resistivity, K is a geometric factor that depends on the electrode array and site characteristics, V is the voltage between two electrodes measured during current injection and I is the magnitude of the injected current. Any 105 electrode configuration or array can, in principle, be used to perform ERT at the surface or between boreholes. For surface ERT, there are several well-established array types, such as Wenner, Schlumberger, dipole-dipole, or pole-pole. For CHERT, several studies have sought to determine the most informative and cost-effective arrays for monitoring dynamic processes (Bellmunt et al., 2012;Zhou and Greenhalgh, 2000). We have followed the scheme used by Bellmunt et al. (2012) in their study of the capability of CHERT for monitoring rapid plume migration. Figure 2b shows the electrode configurations used at 110 the Argentona site: dipole-dipole, pole-tripole and Wenner. Note that these data are acquired sequentially by considering one pair of neighboring boreholes at the time.
We use an optimized survey design that allows more than 5800 data points to be acquired in less than 30 minutes. After the installation of the electrodes around the boreholes (36 at each one), the data acquisition process was straightforward, with no need for large additional costs in maintenance or human working time. The equipment used was a Syscal-Pro multi-channel 115 (10-channel) system from IRIS instruments with 72 electrodes. The current injection time was 250 ms, and stacking of up to 6 measurements was done to meet data quality requirements. It takes 2 hours to complete the 4 CHERT acquisitions needed to cover the whole 2D transect from boreholes N225 to PP20. The combination of four such sections are referred to as a complete CHERT.
Sixteen time-lapse datasets were collected during two years (five in 2015, eight in 2016, and three in 2017), corresponding roughly to a complete CHERT every 90 days.
Data pre-processing was needed to remove anomalous and erroneous data points prior to imaging. Comparison of normal and reciprocal measured resistances is a common technique for appraising data errors (LaBrecque et al., 1996;Slater et al., 2000;Koestel et al., 2008;Oberdörster et al., 2010;Flores-Orozco et al., 2012). We set a threshold of 10% as the maximum 125 acceptable difference between normal and reciprocal measurements. Pseudo-sections of the apparent resistivities are easily created for surface ERT surveys, but there is no corresponding visualization technique for CHERT surveys. Instead, we plot geometric factors, apparent resistivities, and data errors versus data number, to identify electrode configurations with anomalous values. Clearly, for time-lapse studies it is important to ensure that changes observed are due to subsurface processes, and not to changes in the survey setup. Consequently, the sixteen datasets were scanned and compared to keep only identical 130 electrode configurations. This resulted in a reduced set of 2677 identical measurements that were extracted from each complete CHERT before being used in the time-lapse inversion. For forward modeling and inversion, we make the common assumption that the bulk EC distribution is constant in the direction perpendicular to the complete CHERT transect. The corresponding 2.5D electrical forward and inverse problem is solved on an unstructured mesh with tetrahedral elements using BERT (Boundless Electrical Resistivity Tomography) (Rücker et al., 2006;Günther et al., 2006) and pyGIMLi (Generalized Inversion and

135
Modeling Library) (Rücker et al., 2017). The inversion algorithm inverts the log-transformed apparent resistivities, into a 2D log-transformed electrical resistivity distribution. The objective function to minimize is: where φ d is the data misfit term, ∆d = d − f (m) is the vector containing data residuals; with d a vector containing field data; f (m) the forward response of the geoelectrical problem using model m, and n is the order of the norm. In order to make 140 the inversion less sensitive to data outliers, we apply the L1 norm scheme to the data misfit term using iteratively reweighted least squares (ILRS) (Claerbout and Muir, 1973 (Chasseriau and Chouteau, 2003;Linde et al., 2006;Hermans et al., 2012), containing site-specific information about how the resistive bodies are expected to correlate in space. Hermans et al. (2016) provide an example of how the inclusion of covariance information in ERT inversion improves the imaging of the target in terms of shape and amplitude, creating more realistic images. For this purpose, we 150 use an exponential covariance model implemented in pyGIMLi by Jordi et al. (2018). The spatial support of the geostatistical operator helps to reduce the tendency of anomalies being clustered around the electrode region where sensitivities are high. The parameters used in the covariance model were chosen in agreement with the expected fluid dynamic processes. Pore water is expected to flow through the horizontal layers shown in the stratigraphic correlation, so the variations that we expect to observe will be more correlated in the horizontal direction than in the vertical direction. The integral scales in the horizontal and vertical 155 direction are 10 m and 2 m respectively, the anisotropy angle is 90º and the variance of the logarithm of the resistivities was set to 0.25. The detailed description of this type of covariance model is found in, for example, Kitanidis (1997).
The minimization of φ is performed iteratively using the Gauss-Newton scheme. We start the inversion with a homogeneous model corresponding to the average apparent resistivity. In Equation 2, λ is the regularization parameter. We apply an Occam type inversion, in which we seek the smallest φ m while fitting the data (Constable et al., 1987). We set λ to a high value at the 160 first iteration and decrease it by 0.8 in each subsequent iteration. The iterative process is stopped when the data are fitted to the noise level.
To study variations in time, the simplest approach consists of inverting independently each dataset to analyze the evolution of changes. This approach may work when changes are large, but it is no longer considered state of the art because inversion artifacts tend to be time independent (though not always, see discussion by Dietrich et al. (2018)) and may mask actual 165 changes. Singha et al. (2014) describe time-lapse inversion as a way to impose a transient solution constraint through the analysis of differences or ratios in the data (Daily et al., 1992;LaBrecque and Yang, 2001), through the differentiation of multiple individual inversions (Loke, 2008;Miller et al., 2008), or through temporal regularization (Karaoulis et al., 2011). Daily et al. (1992) introduced the ratio inversion, in which data are normalized with respect to a reference model represented by a homogeneous half space. The method allowed qualitative interpretation of resistivity changes, but made quantitative 170 interpretation difficult. This motivated "cascaded inversion" (Miller et al., 2008), which consists of selecting as reference model the result of an initial inversion or baseline dataset. This approach removes the effects of errors and yields more reliable sensitivity patterns (Doetsch et al., 2012). The difference inversion by LaBrecque and Yang (2001) assumes that the changes from one acquisition to another are small, but this is not the case throughout the two years of monitoring at the Argentona site. In the newest approaches, a 4D active time constrained inversion is applied simultaneously to all datasets (Karaoulis 175 et al., 2011) penalizing differences between models. Although this is the most novel procedure for time-lapse inversion, it is computationally demanding. We have decided to apply the "ratio inversion", solving for the updates of a reference model.
For data at time-lapse t, φ d is: where d t is the data vector at time t, f (m ref ) is the calculated forward response of the geoelectrical problem using a reference The reference model for time-lapse inversion was built by inverting data from a complete CHERT and surface ERT from September 8, 2015. The surface ERT dataset consists of 1600 data points acquired along the transect shown in Figure 1a. We used the Wenner-Schlumberger configuration with 72 electrodes spaced of 1.5 m.
Inversion results are displayed in the next section in terms of bulk electrical conductivities, σ b (the reciprocal of resistivities 185 ρ b ).

Reference Model
Results of data used for the reference model are shown in Figure 3.

Time-lapse results
Time-lapse results are displayed in Figure 5 as the ratio between each bulk EC model and the bulk EC of the reference model Below, the time-lapse changes described in the previous paragraph will be interpreted along with precipitation and wave activity data to understand the origins of long-term and short-term behaviors in the dataset. (2) the rainiest periods during the 2 years of monitoring consistently occurred in the fall and spring. The winter and summer of 2016 were the driest periods. Wave-related data (normal y-axis) is from a numerical model called SIMAR 44, and it was generated using HIPOCAS. The numerical model is calibrated using data from wave buoys distributed along the Catalan coast.
Wave numerical models have limitations, and tend to underestimate wave height near the coast, but they give general insights 240 about the wave activity (WAMDI Group, 1988). In Figure 6, we show the significant wave height from the numerical model.
Significant wave height (Hs) is defined as the average height of the highest one-third waves in a wave spectrum (Ainsworth, 2006), and it is the most commonly used parameter because it correlates well with the wave height that an observer would perceive in a wave spectrum (thousands of waves that produce a wavy water surface). The wave data show increased wave In order to assess the impact of a heavy rain event at the site, we have computed the ratio of the CHERT bulk EC models from September 30, 2016 and October 21, 2016, 11 days before and 9 days after the heavy 220 mm precipitation. Figure 7a displays the conductivity ratio image, which reveals a decrease in the conductivity throughout the saturated zone, both above and below the -10 m a.s.l. silt layer, and an increase in the unsaturated zone, above the 0 m a.s.l, between nest N3 and PP20.
No difference is observed below -15 m a.s.l. The decrease in conductivity observed along borehole PP15 is most likely related to water flowing along the borehole (the site was flooded). Heads measured in piezometers N115 (black) and N120 (blue) are shown in Figure 7b, showing that heads increased 60 cm in nest N1 during the rain. Rain was accompanied by an increase in the significant wave height. After 10 days, when the complete CHERT was acquired, groundwater level had already dropped by 30 cm. wind speeds were recorded in the study area corresponding to strong gale winds (more than 50 km/h). Gale winds are usually accompanied by an increase in wave height and precipitations. Even though strong precipitations were not recorded in the area, simulated wave data shows an increase in significant wave height. Figure 8c shows the recovery of the bulk EC in the shallow layers around PP20 in September 2017.
Measurements of water EC from water samples are displayed in Figure 9. Piezometers from nests are screened at different 270 depths, and we have grouped them in three categories: N115, N215, N315 and N415 are in the "upper" group (colored in blue), because the screening depth is above -10 m a.s.l; N220, N320 and N420 are in the "transition" group (colored in green), with the screen around -12.5 m a.s.l, thus, just above the saltwater intrusion; and, N120, N125, N225, N325 and N425 in the "lower" group (colored in red), with the screen below the transition zone, where saltwater is considered to be concentrated. Similar to the plots of average bulk EC from complete CHERT in Figure 6, the major changes occur in the "transition" group, with an 275 increase of water EC by 300%, from 1000 mS/m to 3000 mS/m in the two years of monitoring. Apart from the increase in water EC observed in N115 (screened interval at -9.9 m a.s.l), no clear variations are observed in the "upper" and "lower" groups.
Note that N120 has higher conductivity values than N125, which suggests that a freshwater source is present or a desalination process is occurring below -18 m a.s.l. causes that year to look wet but produces floods rather than proportional recharge. Average yearly precipitation since 2000 is 584,1 mm. 2015 was the driest year of the sequence with only 355 mm of precipitation (38% lower than average). Actually, rainfall was below the long-term average during the last three years of monitoring. The 2015 to 2017 drought is the likely cause for the overall increase in the aquifer bulk electrical conductivity, due to the decrease in freshwater recharge.

285
The reliability of bulk electrical conductivity models obtained with the CHERT experiment can be evaluatedref using other independent datasets. Induction logs (IL) acquired at the Argentona site also provide bulk EC models. Induction logs were done using the GEOVISTA EM-51 electromagnetic induction sound. Figure 11 displays a comparison of the bulk EC from IL along piezometers N2, N4, N3 and N1 (from left to right) and extractions from the complete CHERT conductivity models along the same piezometers. N4 is not on the complete CHERT transect, but as we neglect heterogeneity perpendicular to the 290 transect, we assume nest N4 is comparable to nest N3. IL logs were performed in the piezometers of 20 m length of each nest, because the stainless steel electrodes installed in the 25 m length piezometers severely corrupt the IL signal. IL from May 2015 (light blue), before the beginning of the CHERT experiment are available for all piezometers. They are compared with the CHERT conductivity model from July 2015 (dark blue). In Figure 11c, an IL from July 2016 in nest N3 is compared with CHERT conductivity model from the same month. In Figure 11b, an IL from October 2017 in nest N4, conducted two weeks 295 after the end of the CHERT experiment, is displayed with the CHERT conductivity model from September 2017 of nest N3.
CHERT conductivity model can be well correlated with the IL from all piezometers. There are differences in the magnitudes of the bulk EC, but both methods agree on the location of the transition zone, from -10 to -12 m a.s.l.

300
Surface ERT reflects quite accurately the thickness of the unsaturated zone and the location at which the water becomes more saline, but it is impossible to image the difference between the transition zone and the actual saltwater intrusion. Using only the surface ERT bulk conductivity model, one could argue that SWI in the Argentona site displays the paradigmatic saline wedge shape of Abarca et al. (2007) or Henry (1964). Instead, the CHERT data model suggests two conductive anomalies, one in the unconfined aquifer towards the sea, and one in the semi-confined aquifer below the -10 m a.s.l. silt layer.

305
An important magnitude difference is observed between surface ERT and complete CHERT bulk EC models. The surface ERT model shows much lower bulk EC in the saltwater zone than the complete CHERT model. Studies trying to link hydrological and geophysical models in coastal aquifers (Huizer et al., 2017;Beaujean et al., 2014;Nguyen et al., 2009) have encountered difficulties using surface ERT-based models due to insufficient resolution at the depth of interest. This lack of resolution causes the underestimation of water EC, and thus, of water salinity. The differences in the models shown in Figure 3a 310 suggest that surface ERT is not able to correctly capture the conductivity contrasts in the subsurface. This finding is confirmed by the validation of the CHERT bulk EC models with induction logs (Figure 11).

Reference model: link between bulk EC and geological conceptual model
The complete CHERT produces a quite clear picture of the link between the bulk EC model and the stratigraphic units. We can explain the presence of two saline bodies with the presence of a continuous semi-confining layer, and the existence of up 315 to three different aquifer layers. This is relevant by itself because it was unexpected. The only geologic feature is a relatively minor but apparently continuous silt layer, which we originally discarded as relevant. Bulk EC imaging suggests that this layer may play an important role. The transition zone is not located at the depth of the silt layer. This silt layer is the one separating the unconfined from the semi-confined aquifer. It is not, however, separating the freshwater from the saltwater. The saltwater intrusion zone begins 2 to 3 m below the silt layer, thus suggesting that a significant flux of fresh water occurs below this layer.

320
This result is consistent with sandbox experiments of Castro-Alcalá (2019) who found that relatively minor heterogeneities may cause the saltwater wedge to split.
Part of the reason why the silt layer was not considered relevant lies on the difficulty of visualizing non-monotonic salinity profiles with traditional hydrology monitoring methods. Specifically, salinity profiles in fully screened boreholes (such as PP20) are always monotonic (EC increases with depth) and rarely reach seawater salinity. Our imaging points out that actual 325 salinity is non-monotonic and leads to the suggestion that it is the flow of buoyant freshwater within the borehole what explains both the observed step-wise increase and the fact that salinity is below that of seawater. The process is described by Folch et al. Weathered granite was found in the cores at the bottom of N1, below -17 m a.s.l. At this depth, the magnitude of the CHERT bulk EC model decreases. We can, thus, infer that the decrease in bulk EC at the base of piezometers N325 and N225 is related 330 to the continuity of the crystalline formation. Loss of resolution below PP20 and PP15 does not allow us to infer anything about the presence of weathered granite towards the sea. From the available data, we conclude that the decrease in bulk EC observed in the images has two causes: first, an important change in lithology from gravel to weathered granite; and, second, a decrease in water EC observed in the water samples from N125, with respect to the water sample from N120 ( Figure 9). The water EC values from N125 samples suggest that pore water is a mixture of fresh and saltwater. The granite is, most likely, 335 not an impervious boundary for mixing processes, but merely another source of heterogeneity in the system. The existence of a freshwater from bottom layers of the model is yet to be explored, but it is consistent with the findings of Dewandel et al. (2006), who described frequent highly transmissive zones at the base of the weathered granite in numerous sites around the globe.
The conductive anomalies fade while moving away from the sea. Above -10 m a.s.l, the small conductive anomaly stops 340 before PP15, and is no longer present around N325. Below -10 m a.s.l, the conductive anomaly is present until N325, but is weaker around N225. This diminishing trend in the reference model coincides with water EC values from piezometer N320 being slightly higher than water EC from piezometer N220. We identify a vertical mixing zone, but also, a lateral mixing zone between nests N3 and N2.
In summary, by comparing CHERT bulk EC model, water EC measurements and the site stratigraphic columns, we are able 345 to highlight several features. (1) The resistive anomaly observed at the top is certainly related to partial water saturation.
(2) The seemingly continuous silt layer found at -9 m a.s.l in boreholes N225, N325 and N125 does not represent a freshwater-seawater boundary. The freshwater-seawater boundary appears 2 to 3 m below, which implies that the silt layer is a semi-confining layer and freshwater discharges below.
(3) There are not one but two saline bodies, one in each aquifer. The lower one is a traditional one, but the upper one is more complex and will be discussed in Section 6.4. (4) The conductivity value of the most conductive 350 anomaly below -10 m a.s.l, interpreted as seawater-bearing formations, decreases at the top of the weathered granite. This decrease in bulk EC is explained by the reduction of water EC, and by a reduction in bulk EC due to the larger electrical formation factor of the granite. (5) CHERT bulk EC models show the location of a vertical transition zone, and also the extent of a lateral transition zone.

Seasonality: the natural dynamics
The time evolution of the average bulk EC displayed in Figure 6 shows that there are months with a decrease in bulk EC conductivity and months with an increase in bulk EC conductivity. These months are correlated with rainy and dry periods, and also with the occurrence of storm surges. During summer and beginning of autumn, the conductivity increases slowly until the rain period starts; in autumn, during heavy rains, conductivity decreases; during winter months, conductivity increases due to 360 sea storms; in spring, conductivity decreases, and it reaches its lowest point before the dry summer period begins again. In the deeper areas where seawater is already in place, average bulk EC does not show important variations.

The drought: long-term salinization
The time-lapse ratio image from September 2017 (Figure 5l), the average bulk EC at -12.5 m a.s.l ( Figure 6) and the water EC measurements in the transition zone ( Figure 9) indicate a clear increase of bulk EC in the lower aquifer since the beginning of 365 the experiment.
We conjecture that this increase in water salinity is linked to the drought that started in 2015, and had not yet ended by November 2017. In recent years, drought occurs every 8 to 10 years, and last a few years. This is visible in Figure 10  A 220 mm -a third of the region's average annual precipitation -rainfall event lasting less than a day occurred on October 12, 2016. It was a catastrophic event that created human and material losses due to flooding. The Argentona stream is an ephemeral stream that carries water a few days each year during monsoon-like rains, typically between September and December. A rainfall of this magnitude floods the Argentona stream, and the entire experimental site.
Do the CHERT images capture the effect of the heavy rain in the coastal aquifer? Figure 7a displays the difference in 380 conductivity obtained by the tomography from 11 days before the rain and 9 days after the rain. The bulk EC ratio image reveals a decrease in the bulk EC in both upper and lower aquifers. In October 2016, according to Figure 6, the increase in bulk EC that was taking place was interrupted after this heavy rainfall.
To understand the change in bulk EC, we must think in terms of water masses. When an important precipitation event occurs, freshwater flows through rivers and streams towards the sea. Inland, some freshwater infiltrates into the subsurface, pushing 385 in-situ water masses down and to the sides. The displacement of "old water" creates space for the newly infiltrating fresh rainwater, and this movement enhances mixing processes. Offshore, surface and submarine groundwater discharge is occurring at the same time. The observed change in bulk EC is most likely the result of the mixture of old saltwater with rainwater in the aquifer, which creates a new water, that is still saline but less than before the rain event. However, despite the rainfall magnitude, EC changes were neither dramatic nor long lasting.

390
The effect of the heavy rain that lasted only a few hours supports what was said in the drought section about this rain not being representative of the region's precipitation. One sudden episode, even of this magnitude, is not enough to make a significant difference in the seawater intrusion pattern and in the aquifer's long-term salinization.

The storm: a saltwater event
From July 2015 to October 2016, CHERT experiments had conveyed that the most conductive anomaly was concentrated 395 below the silt layer, but another strong conductive body appeared between nest N3 and borehole PP20 early in 2017.
The traditional SWI paradigm (Abarca et al., 2007;Henry, 1964) suggests that it is the freshwater head what drives the seawater-freshwater interface movement. When heads rise, the interface moves down and seawards because freshwater pushes saltwater seaward. When the groundwater table falls, the opposite occurs, and the seawater interface moves up and inland. The work by Michael et al. (2005) explains how other mechanisms, besides seasonal exchanges, can promote seawater circulation 400 enhancing the seawater intrusion and mixing. According to Michael et al. (2005), some of these mechanisms are: tides, wave run-up on the beach, and dispersion of saline water into freshwater discharge. In the Mediterranean Sea, tidal forcing is not a cause of important change in heads because the tidal amplitude is small (< 20 cm). Wave action and wind could drive changes in the sea level and thus in groundwater heads, but these effects are not long lasting.
A recent study by Huizer et al. (2017) about monitoring salinity changes in response to tides and storms in coastal aquifers 405 showed, through surface ERT experiments, as well as flow and transport simulations, that storm surges can have a strong impact on groundwater salinity. In time-lapse images of the Argentona site, storms seem to be enhancing the conditions for seawater to move inland, through the most superficial layers ( Figure 8a); and further infiltrate the soil from the surface through piezometer PP15, which is fully screened, and between nests N1 and N3. However, salinity increases from the top, rather than from an interface. Therefore, we conclude that these changes in salinity are the result of storm surges, rather than from interface 410 dynamics. In fact, six months later (Figure 8c), the unconfined aquifer has recovered, which implies a more dynamic system in the superficial layers. The CHERT experiment seems to constitute a good tool for the monitoring of such phenomenon near the coast related to tides, wave run-up, and submarine groundwater discharge.

Model validation
Differences between bulk EC models obtained from induction logs and CHERT are attributed to the differences in location and 415 in time of acquisition, considering they were performed neither at the same time nor at the exact same location.
The comparison of the bulk EC model with other independent data sources was very important to prove the reliability of the CHERT experiment. The use of other types of data such as induction logs and water EC from water samples have helped we can observe the increase in water EC in time and in space, but we can't know the depth of the interface or the lateral variations between wells. Induction logs reproduce similar data than the CHERT experiment, but only along piezometers.
Interpolation techniques must be applied to IL data to obtain a 2D image. The CHERT experiment involves real interaction between boreholes, and the interaction is taken into account in the imaging procedure.
6.6 The CHERT experiment

425
The CHERT experiment, contrary to surface ERT, is an invasive procedure because it needs the installation of boreholes, which may affect local dynamics. For example, the vertical anomalies along piezometer PP15, better observed in Figure 7a and The use of an optimized protocol to acquire a complete dataset in the least amount of time is recommended to capture dynamic processes with changes happening in a smaller time-step. This was not the objective of the CHERT monitoring experiment in Argentona from 2015 to 2017, but it is feasible (taking into consideration that metal corrosion will be accelerated by the injection of electric current, implying that the life of the instrument will be certainly shorter). Although surface ERT does not have enough resolution for the depth of interest, the combination of CHERT with surface ERT is suggested to understand 435 the most superficial layers of the subsurface. Future work will include hydrological modeling of density-dependent flow and transport at the Argentona site in order to reproduce the observed bulk electrical conductivity changes observed with the CHERT experiment. It is anticipated that this model can be used to predict future changes in the system.

Conclusions
The monitoring experiment using CHERT at the Argentona site, from July 2015 to September 2017, was successful in several 440 aspects, regarding both geophysical imaging and SWI understanding: 1) The use of CHERT increased the model resolution compared with surface ERT. Comparison of CHERT inversion to salinity profiles from induction logs is excellent and validates the methodology.
2) The increase in resolution allowed us to image unexpected salinity changes both in the upper layers, and the lower layers with only limited loss of resolution with depth despite the high salinity of water. 3) Imaging of spatially fluctuating salinity 445 has led to explaining the paradoxical salinity profiles often recorded in fully screened wells (step-wise increase but without reaching seawater salinity) as due to deep freshwater flowing up inside the well and mixing.
4) Time-lapse CHERT has also been successful in capturing long-term and short-term conductivity changes. Long-term changes include (a) seasonal fluctuations of groundwater flux that cause the seawater-freshwater interface to move seawards during periods of high flux or landwards during periods of low flux; and (b) the long-term salinization of the lower aquifer due 450 to an intense drought in the study area during the monitoring period. Short-term changes include (a) a decrease in conductivity related to a heavy individual rain event of 220 mm of precipitation (a third of the annual average rainfall) in only one day; and (b) an increase in conductivity in the beach area, coinciding with storms that caused strong winds and enhanced wave activity.
In short, employing CHERT at the Argentona site experiment proved to be a cost-effective and efficient tool to shed light on seawater intrusion dynamics through the analysis of bulk formation conductivity.

455
Code and data availability. Datasets and instructions to reproduce the CHERT experiment results will be made available for the scientific community through the H+ database (http://hplus.ore.fr/en/). Data will include surface ERT and time-lapse CHERT files, plus the necessary input files to run the time-lapse inversion in BERT and PyGIMLi.
Video supplement. A supplementary video has been produced to dynamically show the time-lapse evolution of the CHERT experiment at the Argentona site.
installation and during acquisitions; for ensuring site maintenance, and for the fruitful discussions that led to the hydrological interpretation of the geophysical images.
This work was funded by the project CGL2016-77122-C2-1-R/2-R of the Spanish Government. We would like to thank SIMMAR (Serveis Integrals de Manteniment del Maresme) and the Consell Comarcal del Maresme in the construction of the research site. This project also received funding from the European Union's Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie Grant