Introduction

About 25% of the world’s population depends on groundwater hosted in carbonate rock aquifers (Stevanović 2019; Goldscheider et al. 2020). In aquifers of this type, which in carbonate often have enhanced porosity and permeability due to karst processes, focused recharge through preferential flow paths of the unsaturated zone is a dominant driver of groundwater pollution (Daly et al. 2002; Andreo et al. 2006; Worthington et al. 2016). This effect entails fast flow through fractures and macropores capable of mobilizing contaminants, even those considered nonleachable due to their absorption into rock surfaces or colloids, or due to their short degradation time. Large volumes of water and substances mobilized during major storms flow through the subsoil to reach the aquifer in a short time (hours, days, or weeks), thereby causing substantial groundwater contamination. Inadequate understanding of these fast flow processes often leads to serious underestimation of recharge rates and contamination risks. Hartmann et al. (2021) demonstrated that the dangers of such underestimation are especially severe in Mediterranean areas of Europe and North Africa. Studies that deal with the detection of preferential flow processes are still rare and methods that allow for visualizing preferential pathways are very few because being different hydraulic processes categorized as preferential flow, the complexity of identifying transport mechanisms that occur specifically through fast and focused recharge increases. Three main preferential flow modes are: (1) funneled flow of the traditional unsaturated flow theory that is due to the heterogeneity of the medium’s hydraulic properties; (2) fingered flow that is caused by wetting-front instabilities in the unsaturated zone; (3) macropore flow that occurs through large, elongated pores of the medium (Nimmo 2021).

The funneled flow process occurs for example at the interface between two parts of the medium that have different hydraulic properties, such as two layers of contrasting materials. Water flowing into the more conductive material increases its water content. The resulting increase of hydraulic conductivity draws more water to flow in that material, increasing this effect to generate preferential flow. As well as layered systems, this effect is common in such features as sand-filled karst sinkholes (Truss et al. 2007), sediment-filled fractures (Wealthall et al. 2001) as in the case of fractured limestones (Cherubini et al. 2008), and cement and or matrix material in which the lithoclasts constituting porous carbonate rock are very wet and conductive (Turturro et al. 2020). These mechanisms are crucial to lateral flow at large scales and are a major cause of long-distance flow in the unsaturated zone (Nimmo et al. 2002).

Fingered flow is initiated at instabilities that occur at a wetting front, for example at interfaces between contrasting materials. The resulting spatial variation of water content at that interface initiates and sustains fingered flow below the interface, again because the wetter portions (fingers) are much more conductive than the rest of the medium and therefore continue to have more flow. Fingered flow may also occur when the water supply at the land surface is not uniformly distributed (Smith 1967) in homogeneous, hydrophilic media (Wang et al. 2003), or if there is air trapping below the wetting front (Wang et al. 1998).

Macropore flow occurs when the pores are elongated and continuous over a substantial distance when supplied by a copious source of water, whether they are completely or only partially water-filled (Anderson et al. 2009). They are prevalent in many types of natural porous materials, including porous carbonate rocks.

Preferential flow is commonly examined in infiltration studies using dye tracers to visually enhance the patterns of wetting after excavation (Flury et al. 1994; Kodešová et al. 2012). This approach indicates the paths of funneled and fingered flow and allows their qualitative evaluation or quantitative analysis (Bogner et al. 2013), but it is difficult to apply to deep unsaturated zone preferential flow. Visible dye patterns are sensitive to macropore flow only when the dyed water is absorbed at significant distance into the matrix surrounding the macropore. Similar considerations apply for use of geophysical techniques able to provide both dynamic and static imaging (De Carlo et al. 2021; Kukemilks and Wagner 2021; Martin et al. 2022); detection requires that a substantial amount of matrix material be wetted. Although small-volume sensors for what are essentially point measurements of water content are not suitable for macropore flow; even if they happen to intersect a preferential flow path because the amount of water flowing in a macropore is too small to be sensed. Lysimeters (Shipitalo et al. 1994), subsurface drains (Richard and Steenhuis 1988; Smith and Capel 2018), and compartmented collectors (Dahan et al. 1998; Bloem et al. 2010), however, are often used as extensive flow collectors for preferential flux measurements.

The limitations of each of the mentioned methods motivated the authors to apply several different methods in combination to physically recognize preferential flow during infiltration and percolation through an unsaturated zone composed primarily of porous carbonate rock. This approach integrates experimental measurements at the laboratory, plot, and mesoscale to detect and quantify preferential flow paths in carbonate and karstic unsaturated zones. During field experiments with prescribed artificial rainfall, standard hydrological measurements combined with geophysical techniques were conducted in order to identify preferential flow that may be responsible for substantial amounts of focused recharge. The field investigation was conducted in a coastal area strongly affected by karst phenomenon where recharge rates are commonly underestimated, diminishing the perceived risk of groundwater contamination.

Study area

The studied area is in the low-lying coastal zone close to the city of Bari, the main county seat of the Apulia Region, southern Italy (Fig. 1a). It is part of the Lamasinata catchment with elevations from a few tens of meters in the east and up to 80 m above sea level (asl) in the west. There are no perennial streams as the area is strongly affected by karstic processes. This environment is widespread in southern Italy where groundwater is the main water resource for human, agricultural, and industrial needs. This region hosts several industries and agricultural operations. The climate is typically Mediterranean, with hot and dry summers and wet winters. Average annual precipitation is 600 mm, mainly concentrated between November and February (Passarella et al. 2020). The experimental site is located within the area owned by the Water Research Institute (IRSA) of the National Research Council (CNR), Department of Bari (41°07′ N and 16°49′ E, 16 m asl) that falls into a flat area with an elevation of ~22–23 m asl gently sloping towards the coast, 2 km away. It is equipped with five wells—N, S, C, W, E—which are cross-oriented (NNE/SSW and WNW/ESE; Fig. 1) and have been used over time for implementing different types of field experiments (Masciale et al. 2015).

Fig. 1
figure 1

a Study area’s country; b indication of two hydrological catchments in the study area for mesoscale numerical model, and experimental site for plot scale measurements; c experimental site for plot scale measurements; d schematic hydrogeological section across the wells of the experimental site; e log of electrical conductivity (EC) and temperature (T) in the east (E) well

Full size image

Geological and hydrogeological setting

The geological formations in the study area span from Mesozoic through Pleistocene. Pleistocene deposits consisting of litho-bioclastic calcarenites, named Calcarenite di Gravina Formation, lie on top of the Mesozoic fractured and karstified succession, known as Calcare di Bari Formation. This formation, locally made up of limestone and grey dolostone, is frequently interbedded with dolomitic limestone (Pieri et al. 2011).

The Calcarenite di Gravina Formation that constitutes part of the unsaturated zone in the test site is the first rocky layer that water flows through after infiltrating the soil. It is a marine-origin carbonatic sedimentary rock, widespread in the Mediterranean basin, where it often constitutes the thick layer of the unsaturated zone.

This formation indicates the start of the Bradanic sedimentation cycle and consists of transgressive and deepening-upward deposits above the Mesozoic carbonate platform (Tropeano et al. 2002). It is formed by lithoclasts and bioclasts, which create macropores, bound by cement in a micritic matrix, which in turn forms micropores (Turturro et al. 2020). This fabric, typical of dual-porosity media, leads to preferential flow.

Specifically, the studied Calcarenite di Gravina is characterized by open porosity with all pores being interconnected (Andriani and Walsh 2003). On the basis of the porosity classification of carbonate rocks proposed by Choquette and Pray (1970), the greatest contribution to the total porosity is provided by the primary intergranular porosity.

Mineralogically, the calcarenite is constituted of normal calcite with a low magnesium content (CaCO3 ≥ 97%); minor constituents form the insoluble residue which is commonly represented by clay minerals (kaolinite, illite, chlorite, smectite and halloysite) with negligible quartz, feldspar and hydrous iron and aluminum oxides (gibbsite and goethite) (Andriani and Walsh 2003).

The Mesozoic carbonate rocks host the main aquifer structure of the area. Karst cavities and tectonic fractures affecting these rocks make the aquifer markedly anisotropic. Moreover, the presence of dolomitic layers at different depths, which have low or very low permeability, locally causes groundwater flow to be under slight pressure (confined) and fractionated into distinct aquifer levels separated by nonpermeable rock layers. The aquifer is recharged by rainfall that mainly infiltrates through fracture and epikarst formations such as vertical shafts, dolines, and little canyons (locally referred to as “lame”) formed by erosive activity of ephemeral streams.

Karstic phenomena and fractures also occur below sea level, causing seawater intrusion into the aquifer. Consequently, freshwater, having a lower density, floats over saltwater not only in the coastal zone but also inland.

At the location of the test site, the water table is about 2 m asl (Fig. 1d) and almost flat having a hydraulic gradient of about 0.1%. The groundwater flow occurs approximately from SSW towards NNE, perpendicularly to the coastline. The unsaturated zone has an average thickness of 13 m and consists of, from top to bottom, a mixture of bare and waste soil (1 m thick on average), calcarenite (4–6 m thick) and the upper part of the limestone formation (Fig. 1d). The freshwater thickness ranges around 10 m, tapering down to a few meters near the coast. Below it, in the freshwater–saltwater transition zone, the water salinity increases step-wise without reaching the typical seawater values (about 55 mS/cm) as highlighted by the log of electrical conductivity (EC) in Fig. 1e. This delicate hydrogeological equilibrium, mainly governed by the freshwater–seawater balance, is strongly affected by uncontrolled withdrawals and the progressive decrease in rainfall from the effects of climate change, leading to the progressive salinization of the aquifer. For this reason, understanding of recharge processes and dynamics is crucial for protecting groundwater resources and preventing their decline or degradation.

Observations and methods

This study used a combination of laboratory, plot, and mesoscale experiments and modelling to determine the scale at which preferential flow processes are evident. Specifically, each of the used methods allow to detect the evidence of preferential flow at specific scale, being applicable at sample, field and catchment scales, respectively, as they were lens with different resolution that allow for capturing the process with various enlargements.

Laboratory-scale measurements

To understand flow processes at the core sample scale, using standard procedures, the average values of bulk properties of the calcarenite that constitutes the shallow rocky layer of the unsaturated zone were determined (Table 1). Mercury intrusion porosimeter (MIP) measurements were conducted to provide insight into the pore-size distribution, its effect on hydraulic characteristics and the potential activation of preferential flow through larger pore diameters (Turturro et al. 2021). The MIP is based on the fact that mercury, being a nonwetting liquid, does not spontaneously penetrate the pores by capillary action. The method needs a certain external pressure, related to the mercury surface tension, the contact angle between the liquid and the minerals of the medium, and the size and shape of the pore, in order to push the mercury into the pores themselves (Washburn 1921). Evaporation, quasi-steady centrifuge (QSC) and dewpoint potentiometer methods were used to determine water retention, θ(ψ), and hydraulic conductivity, K(ψ), relations, where θ is volumetric water content, ψ is matric potential, and k is unsaturated hydraulic conductivity. Use of both of these methods allows coverage over a wide range of water content.

Table 1 Bulk properties of Calcarenite di Gravina Formation
Full size table

The evaporation method (Wind 1968) measured the drying θ(ψ) function using core samples, 7.8 cm in diameter and 12 cm in height, laterally sealed by epoxy resin and saturated under vacuum. Along the sample height three holes, 1.22 cm in diameter and 2.45 cm deep, were drilled, into each of which was inserted the ceramic cup of needle-tensiometer to measure ψ. Both the sample weights and the matric potential, which changed continuously due to evaporation, were measured during the test.

The QSC method, details of which are in Caputo and Nimmo (2005), using an experimental apparatus that fits into a swinging centrifuge bucket, measured both θ(ψ) and K(ψ). A reservoir rested on the top of the core sample, 7.8 cm in diameter and 6 cm in height, controlled the water flow by having on its bottom a layer of certain granular material, whose conductance determined the flow rate. A tensiometer, external to the centrifuge, was needed for measuring ψ value of the sample once the water content became constant.

The dewpoint potentiometer (Campbell et al. 2007) method used a temperature-controlled model WP4-T (DECAGON Devices Inc.) to determine wetting ψ(θ) curves in the very dry range. The method measured ψ at the equilibrium between the liquid-state water in the sample and gas-state water in a chamber surrounding the sample, on the basis of the chilled-mirror dewpoint technique. Starting from the dry sample, once the equilibrium for each water increment was reached, the corresponding ψ value was performed by inserting the sample in the WP4-T device.

Plot scale measurements

Experimental setup for the artificial rainfall

Two plot scale experiments were performed, in wet and dry periods respectively, to study the response of the groundwater to artificial rainfall events and to identify preferential flow processes. The tests were carried out from 5 to 13 December 2018 and from 9 to 18 July 2019.

Each test consisted of a series of individual rainfall events, simulated by placing a network of 13 sprinklers on the ground surface in order to provide a uniform distribution of water droplets over the whole irrigated area of 7 × 7.5 m (Fig. 2). Specifically, the zone subjected to artificial rainfall was located among the wells C, W and S as shown in Fig. 1.

Fig. 2
figure 2

Experimental setup for the artificial rainfall test and for ERT monitoring

Full size image

The tests used brackish water, characterized by electrical conductivity (EC) of 2.26 mS/cm, to be considered as a tracer for identifying preferential flow. Through a pipe system, the brackish water source was connected to the sprinklers and set to a known rate of sprinkling (Table 1). Starting from the first test of December 2018 up to the end of September 2019, EC and water levels were measured in the monitoring wells at the uppermost surface of the water table to identify the saline water that was applied at the land surface. In addition, natural rainfall data were collected from the three nearest weather stations, far from the study area 4.3 km (41°7′ N, 16°52′ E), 4.9 km (41°6′ N, 16°53′ E) and 5.2 km (41°7′N, 16°53′ E), respectively. The starting data were semihourly rain data then converted in daily rains (mm). The considered daily rainfall data were the average of the daily rain values relating to the three weather stations. EC and water level values were manually measured during the test in December, and continuously recorded using data loggers in July. Specifically, in December two conductivity measurements per day were carried out (one in the morning and the other late in the afternoon) by sampling water within the first meter of the well’s water column to monitor potential water input coming from the surface ascribable to the artificial rainfall. The measurements were carried out, from 5 to 11 December, in well C only, while starting from 13 December they were also made in all the other wells (E–S–N–W). In July both EC and water levels were monitored in real-time by means of CTD Diver probes with a frequency of 2 h (one acquisition every 2 h). The probes were installed at a depth of about 0.5 m from the water level in the wells. During both test periods, different geophysical surveys were carried out as described in the following section (Table 2).

Table 2 Scheme of the tests and performed measurements
Full size table

Geophysical measurements

Among the geophysical techniques used for studying preferential flow, the electrical method is the most suitable one because the electrical resistivity, or its inverse electrical conductivity, is an intrinsic rock property which mainly depends on water which flows through micro- and macropores and fractures. For this reason, in the last decade electrical resistivity measurements are increasingly used for hydrology or hydrogeology characterization, due to their strict correlation with water (Blanchy et al. 2020; Palacios et al. 2020; Leopold et al. 2021; Pleasants et al. 2022). The resistivity method is based on the spatial variation of the electrical resistivity by using four-electrode measurements. In general, two transmitter electrodes, conventionally indicated in literature by the letters A and B as specifically designated in Fig. 3a, are used to inject electrical current into the subsurface and the potential difference is recorded between two receiver electrodes (M and N shown in Fig. 3a). The apparent resistivity, i.e., the resistivity that a homogeneous material filling all of the subsurface would need to have to give the actual measured voltages, is defined through the equation:

$${\rho }_{\mathrm{a}}=\frac{G\times \Delta V}{I}$$
(1)

where ρa is the apparent resistivity [Ohm × m], G is a geometric factor depending on the arrangement of the electrodes, ΔV is the potential difference [Volt] and I is the electric current [Ampere] (Binley and Kemna 2005). The apparent resistivity gives an approximation of the true resistivity distribution in the subsurface.

Fig. 3
figure 3

Three measurement sets of: a apparent azimuthal resistivity for the square array; b azimuthal resistivity data collected through the square array with electrode position and the square rotation

Full size image

In this specific study case, unconventional azimuthal resistivity measurements were also performed that give qualitative information for detecting preferential pathways as well as for estimating secondary porosity, dip angle and fracture strike of the rock as showed in literature (Lane et al. 1995; Busby 2001; Boadu et al. 2005). For azimuthal measurements the electrodes are placed in the vertices of an ideal square. For each square position three sets of resistivity measurements are recorded (Habberjam and Watkins 1967; Habberjam 1972; Lane et al. 1995). The two orthogonal (α and β) are used to provide the azimuthal orientation and the one diagonal (ϒ) as a check on the accuracy of the α and β measurements (Fig. 3a).

The measurements points within the square are rotated around a central point with predetermined angular increments (Fig. 3b). The location of each measurement is assigned to the centre point of the square. The array size (d), that is the length of the square side, affects the depth of investigation, the greater the length the higher the investigated depth. When the resistivity variations are caused by fractures in the medium, the apparent resistivity can be used for determining fracture strike and for estimating secondary porosity through a graphical or analytical approach, as shown in the aforementioned references concerning azimuthal resistivity measurements.

The simplest way to identify the main fracture orientation is by plotting the azimuthal apparent resistivity against the azimuth on a Rosette diagram. The principal fracture strike direction is perpendicular to the direction of maximum resistivity—further details on the analytical approach are shown in Habberjam (1972).

Traditional apparent resistivity measurements are collected along profiles by means of electrical resistivity tomography (ERT) in order to image the geological features of the subsurface. The four-electrode measurements are performed using equally spaced electrodes. The point of measurement is conventionally assigned to the intersection of the two diagonals starting from the centre of the dipoles, as made explicit in Binley and Kemna (2005, p. 137).

The contoured apparent resistivity data provide a qualitative representation of the investigated subsurface due to the limitations of the selected homogeneous half-space. Inverse methods were adopted to determine a possible distribution of the actual electrical resistivity that gave rise to the apparent field data. The inversion procedure provides a 2D or 3D visualization of the resistivity distribution in the subsurface, which is then converted into hydrological state variables through empirical petrophysical or site-specific relationships (Singha and Gorelick 2005; Hinnell et al. 2010; Mboh et al. 2012; De Carlo et al. 2020).

Two geophysical time-lapse ERT campaigns were carried out during the artificial rainfall tests performed in December 2018 and July 2019 in order to detect resistivity changes associated to the water infiltration into the unsaturated zone. The geophysical monitoring performed in December 2018 was focused on resistivity changes in the soil layer, 1 m thick, while the test performed in July 2019 was to investigate almost all the rocky unsaturated zone, down to 10 m depth. Azimuthal resistivity measurements were carried out in June 2019 and September 2019 to identify the main preferential pathways in the unsaturated zone.

The ERT monitoring of the unsaturated soil is presented and discussed by De Carlo et al. (2021). In this paper, the ERT survey and azimuthal resistivity data performed in July 2019 were integrated to gain information on the preferential pathways triggered within the subsurface.

To investigate deeper subsurface flow, two ERT profiles were performed by placing 24 stainless steel electrodes with 1-m inter-electrode spacing along 23-m long transects for each profile to reach depths down to 4 m below the surface. The ERT profile P1 was 1 m far from the boundary of the infiltration area, while the two profiles (ERT profile P1 and ERT profile P2) were 2 m apart, parallel to each other and located downstream of the experimental setup, with the alignment between the centers of the two ERT profiles (Fig. 4) oriented along the expected preferential pattern derived from the surficial azimuthal resistivity data. The infiltration associated with the artificial rainfall event was monitored in time-lapse mode with ERT datasets collected every 2 h.

Fig. 4
figure 4

Location of the ERT profiles

Full size image

The time-lapse inversion of the ERT datasets was performed with RES2DINV code. In this study case, the easiest and most common approach for inverting time-lapse datasets (Loke 1999) was used. Each resistivity dataset was inverted independently and subsequent tomograms at different times were subtracted from the background one, i.e. the resistivity distribution collected before the start of the irrigation test. This approach allows for the unique association of variations in resistivity to hydrological changes related to the water infiltration dynamics.

Azimuthal resistivity data were collected along three square arrays with different size (d = 8, 14 and 24 m), with 24 electrodes for each square, in order to reach the depth of investigation of ~3, 5 and 9 m below ground surface, respectively. The four-electrode array was rotated with angular increments of 15°, as shown in Fig. 3b, in order to provide good data coverage.

Mesoscale recharge modelling

The methods used in this study are based on estimation of episodic recharge by the water-table fluctuation method, in which the temporary, often rapid, rise of the water table after a substantial infiltration event is ascribed to the fraction of the infiltrated water that reaches the water table (Healy and Cook 2002). Recharge is calculated by the simple formula:

$$R=\Delta H\times {S}_{\mathrm{y}}$$
(2)

where R is recharge, ΔH is the change in water level associated with the episodic recharge event, and Sy is the specific yield of the material within which the water level fluctuates. Rate-of-change-based analysis tools were applied, employing a master recession curve (MRC) and episodic master recession (EMR) methods applied to water level hydrographs (Nimmo and Perkins 2018) to estimate episodic recharge associated with natural rainfall events. The methods require input files with columns of time, water level, and precipitation. An MRC is a property of the subsurface that for a given well location gives the characteristic relation between a measured water-level elevation in the well, H and its rate of change with time, dH/dt, within a period of recession with no external input (i.e. precipitation) to the system. The MRC describes the behavior of the water table in the absence of recharge. It is used to correct the measured ΔH in order to estimate the value of ΔH that can be ascribed solely to recharge. The corrected ΔH then is what is appropriate to use in Eq. (2) to calculate R. With its parameterized MRC and standardized procedures for applying it, the EMR method provides means to estimate recharge and evaluate long-term hydrographs to discern and illuminate trends of recharge generation with storm characteristics, seasons, soil conditions, and other factors. This method uses a structured iterative process to determine the required parameters described in detail in Nimmo and Perkins 2018. Critical parameter values that characterize water behavior at the well locations examined in the study were established for each site so that subjective influences did not affect episode-to-episode comparisons. The methods were applied to data sets from two wells, with diverse recharge episodes to evaluate how storm characteristics affect recharge and stormflow. A Sy value of 0.0043 used for all calculations was determined from field data and modeling (Borgia et al. 2002). Specifically, the data sets used to apply the EMR method were from well S, which is one of the five monitoring wells, and well BA22-IS, 5 km away. These wells belong to different and adjacent hydrological catchments, named Lamasinata and Picone catchment, with a surface area of 427 and 266.4 km2, respectively (Fig. 1).

Results

Lab-scale experimental results

The graph in Fig. 5 is the main outcome of the MIP test. It represents the incremental specific pore volume (mL g−1) vs. mean pore diameter, and clearly depicts the bimodal characteristic of the pore-size distribution that characterizes the dual-porosity nature of the calcarenite (Turturro et al. 2021). In fact, Fig. 5 shows two peaks corresponding to 0.14 and 27.5-µm-diameter pores that identify two pore classes, above and below 15 µm, corresponding to textural and structural pores, respectively (Kutilek 2004). The bimodal behavior of pore-size distribution lies in the bioclasts responsible for the larger pores and the intergranular cement that forms the smallest ones.

Fig. 5
figure 5

Pore-size distributions of the Calcarenite di Gravina Formation measured by the mercury intrusion porosimetry

Full size image

Figure 6 shows the water retention and hydraulic conductivity results, which indicate a bimodality consistent with the pore size distribution indicated by the MIP results. They are better fitted with the bimodal Peters-Durner-Iden function computed using HYPROP-FIT software (Iden et al. 2019), as shown in the figure, than they would be by the van Genuchten model or other unimodal models (Caputo et al. 2022).

Fig. 6
figure 6

a Water retention and b hydraulic conductivity data fitted with the bimodal Peters-Durner-Iden functions

Full size image

Plot scale hydrologic monitoring

Although the groundwater level and EC were monitored over a period of ~1 year (from December 5 2018 until September 2019), the data presented here focus on two time windows roughly covering the periods during which the two tests were performed (Dec–Feb and Jul). Figure 7a shows the water level fluctuations and the natural rainfall considering the average of values recorded in the three weathering stations during the first time window. The graph shows evidence of the same water-level-fluctuation behavior in the five wells and the rapid response to the precipitation events, such as the one on 25 January when there was 64 mm of rain. The slight differences among the wells are due to the different elevations of reference points.

Fig. 7
figure 7

a Water level trend in the five wells during the first time window (December 2018–February 2019); b zoom of the water level trend 5–20 December 2018

Full size image

In Fig. 7b, which magnifies the lower left quadrant of Fig. 7a, the intensity (mm) and duration of the artificial rain are also reported. The artificial rainfalls have no observable effect on water level despite their comparable or even higher intensity compared to the main natural rain event occurred in the same time window, which is not surprising, considering the small irrigated area with respect to the size of the contributing catchment.

The graphs in Fig. 8 show EC values in each monitoring well and the natural rainfall collected during the same period. For the case of Fig. 8a, the red area is magnified in the graph shown in Fig. 8b, which highlights a clear effect attributable to artificial rainfall for well C. In fact, the EC values recorded in that well exhibit an increasing trend during the first test, especially on 12–13 December. No comparison of this trend for the other wells is possible because their monitoring started at the end of the artificial rainfall events.

Fig. 8
figure 8

a Electrical conductivity (EC) trend in well C, c well N, d well S, e well W, f well E during the first time window (December 2018–February 2019); b zoom of the electrical conductivity trend in the period 5–20 December 2018

Full size image

Nevertheless, observations about EC trends after the rain event at the end of January (Fig. 8c–f) lead to some interesting evidence. The greatest variations, almost immediately after the rain, are recorded in wells N and C, both decreasing in EC. However, whereas in well C the EC values were restored to pre-rain levels in a short time, in well N the observed increasing trend is weak. In wells W and E, the EC values increase after the rain event and the variations are slow, more so in E than in W. Finally, in well S, the EC values exhibit a strong stability regardless of rainfall (Fig. 8d). Assuming these different degrees of steadiness are related to local subsurface conditions and not to different degrees of vertical mixing of water within each well, which seems likely given that subsurface materials are probably far more heterogeneous than the vertical column of water within a well, this evidence suggests that there are preferential paths, probably with different flow magnitudes, that drain water in some portions of the unsaturated zone more than others.

The graphs in Fig. 9 show the water level fluctuations (Fig. 9a) and EC values (Fig. 9b) recorded during the second time window (July). The natural daily rainfall and the duration of the test are also indicated. In July, the artificial rainfall had no effect on the water-table fluctuations. Natural rains with comparable intensity have a different influence on the water level due to the spatial extent of natural rainfall. Specifically, the event on 10 July caused a 20-cm water level rise, characteristically different than that on 16 July. This can be partly due to the irregular distribution of the summer rainfall and partly to the different moisture condition of the unsaturated zone. Regarding EC values (Fig. 9b), only well N exhibits appreciable variations; however, it is difficult to establish whether they are linked to artificial or natural rains.

Fig. 9
figure 9

a Water level and b electric conductivity trends in the five wells (C, N, W, E, S) during the second time-window (from 8 to 30 July 2019), also showing rainfall for this period

Full size image

Plot scale geophysical results

The azimuthal resistivity models highlight the high degree of heterogeneity of the unsaturated subsurface (Fig. 10). At 3 m in depth below ground surface the Rosette diagram shows a clear NE–SW preferential pattern (continuous black line in Fig. 10a), as confirmed by the high value of apparent anisotropy (λ = 2.52), that is the ratio between apparent resistivity measured perpendicular and parallel to the strike angle. The main rock parameters have been derived according to Lane et al. (1995). The strike angle, i.e. the direction of the line formed by the intersection of the preferential pattern with the azimuth, has been estimated as θ = 15° and the dip angle, i.e. the angle that the fracture makes with a horizontal plane, has been estimated as α = 21°.

Fig. 10
figure 10

Azimuthal resistivity data at three different depths of investigation: a 3 m; b 5 m; c 9 m from ground surface

Full size image

As the depth of investigation increases, the anisotropy of the rock changes over space. Figure 10b shows the Rosette diagram of the azimuthal resistivity distribution at 5 m in depth. At this depth, the apparent anisotropy has been estimated to λ = 1.96. Compared to the upper unsaturated layer, the lower value of λ reveals a weaker anisotropy. Nevertheless, two main preferential patterns have been detected along directions W–E and NW–SE (dotted black lines in Fig. 10b) with fracture strike equal to θ = 90° and θ = 342°, respectively. Due to the presence of more than one fracture strike, the analytic solution of the dip angle is not representative of the dipping of the fractures.

Compared with the other two square arrays, the deep azimuthal resistivity data (Fig. 10c) show a lower apparent electrical resistivity and a higher apparent anisotropy value (λ = 4.23) with a clear preferential pathway along the NW–SE direction (continuous black line in Fig. 10c), fracture strike equal to θ = 285° and dip angle to α = 8°. The main rock parameters of the upper unsaturated rock have been summarized in Table 3.

Table 3 Square arrays parameters at three depths (z) for the investigated rock
Full size table

This information has been merged with the time-lapse ERT findings to better characterize the subsurface in terms of preferential patterns in the unsaturated zone. As stated, the ERT survey could detect resistivity variations within a 4-m region below ground surface due to the reduced length, and, for this reason, it can be only compared with the surficial azimuthal resistivity data, corresponding to z = 3 m.

Figure 11 shows the electrical resistivity changes expressed as the percentage of difference over time during the irrigation dynamics. Five elapsed times have been considered, each 1-h long. Time t0 refers to the resistivity changes 1 h from the start of irrigation, time t1 refers to resistivity changes 2 h from the start of irrigation and so on. Between time t3 and time t4, the irrigation was stopped and, finally, time t4 refers to resistivity changes after irrigation ceased. According to the results of the surficial azimuthal survey, the ERT monitoring highlights significant resistivity changes in the ERT profile P1, i.e., the profile located 1 m to the irrigation system (Fig. 4). The main variations have been observed in the central part of the profile (Fig. 11a), which lies on the NE–SW direction identified as a preferential pattern from the azimuthal data.

Fig. 11
figure 11

Time-lapse ERT results during the artificial rainfall event at five elapsed time points for the: a ERT profile P1; b ERT profile P2

Full size image

Conversely, no significant changes were recorded in ERT P2 profile (Fig. 11b), the farthest from the irrigation system (Fig. 4), located 3 m far from the boundaries of the irrigation system This is not surprising considering that the dip angle recorded from azimuthal measurements at 3 m in depth, supposed constant along the whole area, forces the infiltrated water below the maximum investigation depth in correspondence to the profile P2.

Mesoscale numerical model

The EMR analysis performed in the study provides an opportunity to understand the nature of episodic aquifer recharge on a larger scale than the other methods previously described. Figure 12 shows EMR results for a period just over 200 days, from 10 December 2009 to 30 June 2010, which illustrates that the water table rises very rapidly, indicative of preferential flow through the unsaturated zone (see Table 4 for a summary of the results). The recorded data indicates 68 episodic events at well S and 29 at BA22IS due to the shorter period of data collection, from 2008 to 2011 and 2010 to 2011, respectively. The recharge behavior itself is similar between the two wells indicating that the unsaturated zone processes are somewhat consistent over a wide area (Table 4). Slight differences are due to the fact that each well was analyzed with precipitation data from the nearest rain gauge. Overall, the recharge is proportional to the precipitation duration, average precipitation intensity, and the total amount of precipitation in each event for both wells as shown in Fig. 13. Specifically, the precipitation duration, express in days, means the consecutive days characterized from at least one precipitation event.

Fig. 12
figure 12

Results from the EMR analysis for a well S and b well BA22IS. The starting date of this part of the record, shown here as day 590, corresponds to 10 December 2009, while the end date is 30 June 2010

Full size image
Table 4 Well monitoring periods, number of events identified by the model, and statistics for recharge and recharge to precipitation ratio (RPR)
Full size table
Fig. 13
figure 13

a–f The relationships between recharge and precipitation duration, amount, and average intensity for different monitoring periods (well S on left and well BA22IS on right)

Full size image

Figure 14, shows the relationships between recharge and precipitation duration, amount, and average intensity for both wells and the same monitoring period ranging from January 2010 to June 2011 for both wells. Specifically, the graph of Fig. 14a shows, for both wells, a good correlation between recharge and precipitation duration: the longer the precipitation is, the greater the recharge.

Fig. 14
figure 14

The response of well S and well BA22IS to recharge for the same monitoring period (January 2010 to June 2011) in terms of precipitation a duration, b amount and c average intensity

Full size image

The same is true for Fig. 14b, which clearly exhibits a very good correlation of the recharge with the total amount of precipitation for both wells. Regarding BA22IS, Fig. 14c shows a strong correlation of the recharge with the average precipitation intensity below a specific threshold of ~4 mm/day. For higher average intensity values, the recharge is not so strongly correlated with the precipitation intensity, possibly due to a runoff effect initiated because preferential pathways are fully saturated and cannot accommodate the additional infiltration water.

Discussion

The laboratory-scale tests on the calcarenite, a carbonate rock that constitutes the shallowest unsaturated rocky layer below the thin soil layer, establish that the bimodal pore-size distribution is consistent with the bimodal nature of the best fit of the hydraulic functions. The dual-porosity nature of the calcarenite due to the presence of both the inter-granular cement, responsible for the smallest pores, and the bioclasts, responsible for the large pores, may be responsible for smaller-scale effects with respect to preferential flow through that geologic unit. The bimodality of the calcarenite suggests the possibility of preferential flow occurring in the class of larger diameter pores.

The analysis of the data collected during the two tests conducted in January and July allows a possible conceptual model to be outlined of the saturated–unsaturated system intercepted by the five wells. The five monitoring wells intercept fractures that differ both in direction and magnitude, producing different effects in terms of electrical conductivity values as observed after both natural and artificial precipitation. The different behaviour observed in the monitoring wells after the natural and artificial rainfall events help to evaluate whether wells N and C intersect important local fractures capable of transmitting water coming directly from the overlying unsaturated zone such as to produce evident effects in terms of observed electrical conductivity. Wells E and W, on the other hand, intersect only a few small fractures that transmit small quantities of infiltrated water that do not produce significant EC variations after rainfall. This consideration is corroborated by a video inspection performed in a well 500 m from the five monitoring wells. Well S, however, seems to be affected by fractures not linked to the local dynamic effects of infiltrated water but rather to those at the catchment scale capable of producing only slower seasonal variations. These considerations highlighted a complex anisotropy of the studied physical domain that is verified by the geophysical measurements that allowed characterization of the subsurface in terms of preferential flow patterns.

The geophysical information obtained from two different techniques used in this study has been merged to better image preferential pathways that occur in the unsaturated zone. In the upper unsaturated layer an anisotropic distribution of the azimuthal resistivity measurements with a NW–SE direction, a strike angle of 15° and a dip angle of 21° from ground surface, have been observed. The anisotropic pattern forces water to move along such dipping layers via leading preferential pathways. This evidence is confirmed by the resistivity distribution of the ERT tomograms along ERT profiles P1 and P2. In particular, the pronounced decrease electrical resistivity over time in the middle portion of the ERT profile P1, which is the expected wetting area from the azimuthal findings, demonstrates that the water flow is intercepted from the ERT profile closest to the water source.

In contrast, the water moves below the investigated depth along the more distant ERT profile (ERT profile P2 in Fig. 4) due to the dip angle of the fracture system. This is indirect evidence of the utility of the geophysical tool to clearly visualize fracture orientation and preferential pathways.

The deeper azimuthal dataset cannot be compared with data measured in the wells because its depth of investigation is about 3 m above the water table. Nevertheless, the resistivity distribution provides information about the rock texture, that can be compared with what is learned from the upper azimuthal datasets. The lower resistivity values and the higher anisotropy ratio indicate more fractured and more permeable rock layers than the shallow layers.

In addition, the different moisture conditions of the unsaturated zone during the two experimental time windows affect the water transmission velocity. In winter (wet period) the fractures have a higher saturation degree and therefore transmit water more quickly, possibly causing a decrease of EC such as occurred in wells N and C after rainfall on 25 January (Fig. 8a–c). In contrast, as opposed to the process that occurs during the winter test, during the summer (dry period), when the degree of saturation is lower, fractures transmit less water, which is only sufficient to dissolve the salts precipitated in them, thus the EC values slowly increase.

It was clear that the artificial rainfalls had no effect on the water level despite their comparable or even higher intensity with respect to the main natural rain event. This is due to the small size of the area involved in the artificial rainfall with respect to the size of the feeding basin, and the saturated zone is conductive enough that water added from the small area is transported laterally fast enough that it does not form an observable mound at the water table. Obviously, the data measured in the monitoring wells represent the response of what happens in the feeding basin and not in the small and nearby area subjected to artificial rainfall.

The results of recharge estimation using the EMR method also shed light on the rapid episodic flow in response to natural rainfall events. The lag time parameter in the model is set to a value that reflects the time it takes infiltrated rainfall to reach the aquifer. For both wells, the lag time was less than 1 day. This confirms the fact, also brought to light by smaller-scale experiments, that rapid flow through the unsaturated zone is a very important process in recharging the aquifer and potentially transporting contaminants.

The two wells used for the EMR calculation, though they belong to different hydrological catchments, show the same behavior, indicating that they fall in the same hydrogeological catchment. The agreement of the results obtained in analyzing the datasets of the two wells allows for generalization of their behavior on a larger scale. Like the study of Allocca et al. (2015), this study confirms the value of the EMR technique for quantifying groundwater recharge processes at basin and episodic scales in a heterogeneous karst aquifer typical of Mediterranean areas.

Conclusions

This paper presents an unusual combination of methods applied at different scales: measurements on core-sized samples, plot-scale experiments with artificial sprinkling, and mesoscale through EMR modeling of episodic recharge. This combination has enhanced understanding of preferential flow activation in an unsaturated zone composed of thinly mantled soil, porous carbonate, and karstic fractured rock.

At the plot scale, electrical conductivity, temperature, and water level measurements in five monitoring wells after several artificial and natural rainfall events, showed different behavior in the monitoring wells, implying a strong anisotropy of flow in the unsaturated zone. The data collected in the wells suggest a complex conceptual model of the saturated–unsaturated system characterized by a network of fractures that differ in both direction and conductance.

The geophysical data, obtained in several acquisitions with a combination of two different techniques, confirm the complexity of the conceptual model. They allow localization and quantification of the activation of preferential paths, leading to a better characterization of the subsurface in terms of preferential flow patterns. The time-lapse geophysical surveys allow tracking of preferential flow, enhancing understanding of flow dynamics over large areas and providing quantitative estimations of the water storage, velocity, and spread.

The recharge model based on the ERM technique confirms the presence of rapid flow through the unsaturated zone, a crucial process in recharging the studied heterogeneous and karst aquifer. In this study, the combination of the outcomes of the different techniques applied at differing scales allows a more complete knowledge of how these preferential flow processes act in combination to promote rapid flow to the aquifer. This unusual multiscale approach adds to research advancement in this field not achievable by traditional single-scale approaches.