Introduction

Nitrate contamination of groundwater has become one of the most serious environmental concerns in industrialized countries. The negative effect of fertilizers and pesticides leaching into aquifers from agricultural activities has been intensively investigated (Böhlke et al. 2007; Chen et al. 2007; Nolan et al. 2002; Postle et al. 2004; Puckett and Cowdery 2002). While the export of contaminants into surface waters (McMahon and Böhlke 1996) have immediate effects on aquatic ecosystems such as eutrophication (Iversen et al. 1998; Lamers et al. 1998; Lucassen et al. 2004), effects on groundwater quality are not as apparent and may be underestimated. These effects pose a serious risk of long-term contamination and health consequences for future generations (Kraft et al. 2008). For instance, NO3 − contamination of drinking water is known to cause methemoglobinemia in infants (Cynthia et al. 2002), while elevated concentrations of SO4 2− may cause diarrhea (World Health Organization 1996). In many countries, a significant portion of drinking water is exposed to these threats as evidenced by recent European law adjustments for water protection, from the Nitrate Directive (Official Journal of the European Communities 1991) to the Water Framework Directive (Official Journal of the European Communities 2000) which specifically address this issue.

Nitrogen is the most widely used fertilizer and is added to soil in different forms and oxidation states, depending on soil features, crop needs, and local convenience to use manure or synthetic compounds. In the topsoil, microbial activities rapidly transform nitrogen to nitrate (Fenchel et al. 1998). Once infiltrating waters enter the subsurface, oxygen (O2) may rapidly decrease and NO3 − can be used by denitrifying bacteria as electron acceptor (Appelo and Postma 2005) in organic matter and sulfide oxidation processes (Aravena and Robertson 1998). If these substrates are not abundant, denitrification occurs at low rate and, therefore, NO3 − can travel long distances within aquifers (Andersen and Kristiansen 1984; Shomar et al. 2008; Taylor et al. 2006), depending on aquifer materials and structure heterogeneities. Thus, to determine the direction and rate of groundwater flow and to predict NO3 − solute transport, hydraulic conductivity distribution has to be ascertained (Gelhar 1993; Webb and Davis 1998).

For this purpose, a large tank was filled with fertilizer-contaminated aquifer materials, whose grain-size distribution and permeability was intensively characterized to obtain an accurate estimate of sediment heterogeneities. Multilevel slug tests were used to estimate k values along piezometer depth (Rus et al. 2001), and not pumping tests, since usually the latter method gives a large scale value of k and not the point value (Rovey and Niemann 2001). As low-permeability lenses may decrease the nitrate removal process while high-permeability zones could accelerate it, to assess the role of heterogeneities on the remediation efficiency the sediment clean-up time was estimated by flushing the tank with purified tap water. Furthermore, the dissolved organic acids acetate and formate were monitored throughout the experiment, in order to evaluate the role of organic matter on denitrification rate. This is because low-molecular-weight organic acids are good indicators of denitrification in alluvial aquifers (Baker and Vervier 2004; Rivett et al. 2008). The results were modeled and discussed in comparison with those from previous column experiments (Mastrocicco et al. 2009) performed on the same aquifer material but flushed at a flow velocity that was 64 times higher, to compare denitrification rates obtained with different methods.

Materials and methods

Field site

The sediments were excavated from a sand pit (in the Po Plain, Italy) located along a palaeo-meander bend of the Po River (Mastrocicco et al. 2009). The principal Quaternary lithofacies of the area include coarse-grained facies associations (fluvial-channel and crevasse sands) and fine-grained ones (floodplain, prodelta and marsh deposits) (Amorosi et al. 2003). Associated with the coarse-grained facies is the most productive aquifer, consisting of a 20–25 m thick Holocene sandy succession extending all over the area (Stefani and Vincenzi 2005). This aquifer is usually confined by a thick clay layer, which is often absent near the major paleo-channel bodies; hence, pollutants may penetrate downwards into the unconfined aquifer (Fig. 1).

Fig. 1
figure 1

Quick-Bird satellite image of the sand pit in February 2006: dashed lines represent the paleo-channel extensions interpolated using the available core logs (green circles), the blue cross-circles indicate the monitoring wells, and the solid line indicate the hydrogeological cross section through the area

Full size image

The excavation was completed in 2 h on March 8th 2006, using a crane excavator; initially the top 0.5–0.6 m of weathered soil was dug, then saturated sediments from 0.5 to 0.6 m below ground level down to 1.9–2.0 m were collected and immediately displaced into the tank located at the Hydrogeology Laboratory of the University of Ferrara. During the excavation, a geological survey was carried out in order to recognize sedimentary structures (Online Resources 1). The heterogeneous nature of the sedimentary succession, with upward-coarsening silt and sand sediments, gradational lower boundaries and sharp tops, indicates a crevasse splay facies (Amorosi and Marchi 1999).

Tank set up

The experiment was performed in a large tank (4 m wide by 8 m long and 1.4 m deep) located in the Hydrogeology Laboratory at the Scientific and Technological Pole of the University of Ferrara (Fig. 2). The tank was assembled with an internal structure of armed PVC fastened on an external structure of natural wood; the non-metallic materials were selected for future electrical resistivity tomography applications. The tank was filled with 42 m3 of unconsolidated material (35 m3 of natural sediments and 7 m3 of gravel), by means of a bulldozer equipped with a 2-ton tilting shovel mounted on a 8-m-long telescopic crane.

Fig. 2
figure 2

Plan view of the tank with the constant head reservoir on the left-hand side, creating a flux towards the outflow pipes on the right; gravel walls are represented near the inflow and the outflow walls; 2.5 and 5 cm i.d. piezometers are plotted with circles and triangles, respectively; the origin of the x, y, z axes was located to the left bottom corner of the inflow wall

Full size image

Saturated natural sediments were poured by the tilting shovel into the tank starting from the inflow gravel wall towards the outflow wall. Once a layer of 0.2 ± 0.05 m was created, it was compacted, using the bottom of the shovel, and new layers were added. The filling procedure took 10 h. Following this, the natural compaction of sediments was monitored for 4 months; the average bulk compaction was found to be approximately 0.02 m.

An external reservoir (constant head) was connected to the tank via three inflow pipes (Fig. 2); the reservoir had a large surface area to minimize the introduction of trapped air bubbles. The constant head was introduced to create a steady state flux with a mean head gradient of 7% in the sediments; in order to maintain a certain uniformity of the potentiometric surface, two gravel walls were built one at the inflow and the other at the outflow of the tank; the mean head gradient in the gravel walls was calculated by the Darcy law to be 0.1‰.

Considering the Darcy’s law in the form of

$$ \overrightarrow {i} = {\frac{{J_{\text{w}} }}{k}} $$
(1)

i is the head gradient (L L−1) and J w (L T−1) the volumetric water flux density.

The experimental hydrodynamic conditions, in both the tank and column experiments were assessed using Reynolds number Re (Venkataraman and Rao 1998), which expresses the ratio of inertial to viscous forces. This is a criterion to discriminate between laminar flow, which occurs at low velocities and turbulent flows. In porous media, Re is defined as Re = J w d/γ where J w is the specific discharge, d is the pore length dimension (mean pore size), and γ is the kinematic viscosity.

The results showed Re values of 2.1e−5 and of 3.1e−4 for the tank and column experiments. Since laminar flow occurs if Re < 1, the observed values indicate that under the present experimental flow conditions laminar flow occurred at low constant flow rates.

Twenty-six piezometers were installed using a hand-driven auger, on the base of a semi-regular monitoring grid (Fig. 2). A detailed topographic survey was carried out using a Nikon DTM-450 total station, to accurately determine the well case position in x, y, z axis (Fig. 2); piezometric heads were monitored every 2 months.

Seventy-eight undisturbed 1-inch cores were collected every 0.3 m by a Shelby sampler for grain-size analyses. Particle size curves were obtained using a sedimentation balance for the coarse fraction and an X-ray diffraction sedigraph 5100 Micromeritics for the finer fraction; the two regions of the particle size curve were connected using the computer code SEDIMCOL (Brambati et al. 1973). Bulk density and total porosity were determined gravimetrically; the organic carbon content (f oc) of the sediments was measured by dry combustion.

To estimate the k variability, 130 multi-level slug tests were performed within the saturated zone (every 0.15 m) using small inflatable straddle packers; a pneumatic initiation system was used to instantaneously lower the static groundwater level of approximately 0.05 m. The tests were run after the complete elution of resident water in order to avoid interference with the monitoring program, due to the possibility that slug tests could modify the flow field, although transiently. The estimated k values are valid only in the vicinity of the borehole, because k values may be biased by numerous factors. Usually, the bias is towards lower k values that can reach even one order of magnitude (Butler et al. 1996; Hyder and Butler 1995; Zlotnik 1994). This because the bias is due to the presence of low-k zones around the well screen induced by drilling disturbance. To minimize this problem, piezometers were modestly developed by bailing approximately 5 l from 1″ piezometers and 10 l from 2″ piezometers, to virtually eliminate all skin effects (Rovey and Niemann 2001) during slug testing. Water level in every piezometer was instantaneously lowered with a pneumatic syringe. All the acquired slug test responses were analyzed using the Bouwer and Rice method (Bouwer and Rice 1976).

Once the tank elution started, groundwater sampling took place every 2 months in order to get a snap shot of the distribution of solute concentrations; samples were taken from every piezometer via low-flow sampling technique using Waterra inertial pumps, from the external reservoir, from the inflow gravel wall and from the outflow pipes.

Analytical methods

In-well parameters were determined by the HANNA Multi 340i instrument which includes a HIcell-31 pH combined electrode with a built-in temperature sensor for pH measurements, a CellOx 325 galvanic oxygen sensor for DO measurements, a combined AgCl-Pt electrode for Eh measurement, and a HIcell-21 electrode conductivity cell for EC measurements. Samples were filtered through 0.22 μm Dionex vial caps. The major cations anions and oxianions (acetate and formate) were determined by an isocratic dual pump ion chromatography ICS-1000 Dionex, equipped with an AS9-HC 4 × 250 mm high-capacity column and an ASRS-ULTRA 4 mm self-suppressor for anions and a CS12A 4 × 250 mm high-capacity column and a CSRS-ULTRA 4 mm self-suppressor for cations. An AS-40 Dionex auto-sampler was employed to run the analyses, Quality Control (QC) samples were run every ten samples. The standard deviation for all QC samples run was better than 4% relative. Charge balance errors in all analyses were <5% and predominantly less than 3%. Alkalinity content was determined using a Merk Aquaquant titration package.

Geostatistical analyses and three-dimensional visualization

To describe grain-size distribution analysis, the median of the average grain radius expressed in mm (M), the uniformity coefficient (U) which is the ratio between d 10 and d 90 (d 10 and d 90 being the mean particle diameter in mm expressing the 10th and 90th percentile of the cumulative curve) and the coefficient of skewness (S) were calculated. The coefficient of skewness is a measure of asymmetry in the distribution; a positive skew indicates a longer tail to the right, while a negative skew indicates a longer tail to the left, while a perfectly symmetric distribution, like the normal distribution, has a skew equal to 0. The sample skewness S is calculated as follows (King and Julstrom 1982):

$$ S = {\frac{1}{{{ns}^{3} }}}\sum\limits_{i = 1}^{n} {\left( {x_{i} - \overline{x} } \right)^{3}} $$
(2)

where n is the number of data values for a sample x i , \( \overline{x} \) is the sample mean and s is the sample standard deviation. A classical statistical approach was used to infer k distribution derived from slug tests within the tank, using the Kolmogorov–Smirnov (K–S) statistic (Sprent 1993). K–S statistic is the largest difference between an expected cumulative probability distribution and an observed frequency distribution. The expected distribution is the normal probability distribution with mean and variance equal to the mean and variance of the sample data.

The three-dimensional representation of the saturated k and the distribution of solute concentrations were generated through the application of a quadratic inverse distance algorithm without smoothing and with an anisotropy ratio between the y and x axis equal to 2, accounting for tank dimensions (Deutsch and Journel 1998). The ordinary Kriging interpolation method (Swan and Sandilands 1995) was used to represent piezometric contours; the experimental variogram was fitted by means of a nonlinear Least Square regression method to a linear function with slope of 1.5e−3 and anisotropy ratio equal to 2, accounting for tank dimensions.

Modeling

Numerical modeling was required to quantify and compare the relevant hydrogeological and biogeochemical parameters obtained by the tank and column experiments. Assuming a uniform water content and steady-state flow conditions, the one-dimensional transport non-equilibrium convection–dispersion equation (CDE), including first-order degradation reaction, can be written as (van Genuchten and Wierenga 1976)

$$ \theta_{\text{m}} {\frac{{\partial C_{\text{m}} }}{\partial t}} = \theta_{\text{m}} D_{\text{m}} {\frac{{\partial^{2} C_{\text{m}} }}{{\partial x^{2} }}} - J_{\text{w}} {\frac{{\partial C_{\text{m}} }}{\partial x}} - \alpha \left( {C_{\text{m}} - C_{\text{im}} } \right) - \left( {\theta_{\text{m}} \mu_{\text{m}} } \right)C_{\text{m}} $$
(3)
$$ \theta_{\text{im}} {\frac{{\partial C_{\text{im}} }}{\partial t}} = \alpha \left( {C_{\text{m}} - C_{\text{im}} } \right) - \left( {\theta_{\text{im}} \mu_{\text{im}} } \right)C_{\text{im}} $$
(4)

where subscripts m and im pertain to the mobile and immobile region, respectively. C (ML−3) denotes solute concentrations as a function of distance x (L) and time t (T). D m (L2T−1) is the dispersion coefficient for the mobile region, the volumes θ (L3L−3), θ m (L3L−3), and θ im (L3L−3) are the total, mobile, and immobile water content. For θ m = θ, Eq. 3 reduces to the single-domain CDE. The solute-mass transfer between mobile and immobile regions is limited by the first-order rate coefficient α (T−1). The first-order degradation rate is μ (T−1). The elution curve of Br− from the tank outflow was analyzed to determine the hydrodynamic dispersion coefficient D. The CDE was solved by the code CXTFIT 2.1 (Toride and Leji 1999) in estimation mode to fit observed concentrations. Subsequently, the degradation rate μ was estimated from the NO3 − elution curve. The column elution was simulated following the non-equilibrium CDE, in estimation mode to fit observed Br− concentrations to the three unknown parameters D, θ m , and α. Subsequently, degradation rates μ m and μ im were estimated from the NO3 − elution curve.

In addition to infer possible mineral precipitation and/or dissolution, the mineral saturation index (SI) of calcite (CaCO3) and dolomite [CaMg(CO3)2] was calculated using the geochemical computer program PHREEQC-2 (Parkhurst and Appelo 1999) using the standard PHREEQC database. The SI is employed when large differences from equilibrium are supposed and reflect thermodynamic equilibrium when SI is equal to 0. While if SI > 0 this reflects super saturation conditions and precipitation may occur, if SI < 0 this reflects under saturation conditions and dissolution may occur (Appelo and Postma 2005).

Results and discussion

Tank characterization

The grain-size analysis of the 78 samples (Fig. 3) showed that sediments came from the same depositional environment, identified as a crevasse splay. Crevasse splay is a depositional feature which forms when a river breaches one of the banks, spreading sediment load on the floodplain in a pattern similar to a Delta: coarser sediments deposits first close to the breach while finer one are carried further to the edge of the crevasse. A considerable variability in the grain-size distribution was registered, because this depositional environment is characterized by a broad range of textures, due to traction and traction plus fall-out processes. These processes are distinct mechanisms of sediment transport in a basin characterized by decreasing velocity within river beds (Amorosi and Marchi 1999). The median of the average grain radius varied from 0.043 to 0.107 mm, typical of very fine sands; U varied from 1.65 to 3.12 depicting different sorting of the samples and S remained within 0.40–0.84, indicating a quasi-normal distribution with small tailing towards the large grain size. Following the Wentworth classification these sediments can be defined as silty-sands (Table 1).

Fig. 3
figure 3

Cumulative grain-size distribution curves of all the 78 samples analyzed, on the left; Wentworth classification triangle, on the right

Full size image
Table 1 Average grain-size distribution, hydraulic conductivity, bulk density, total porosity and f oc measurements for natural sediments, and gravel walls
Full size table

The k values measured in the tank range from 3.2e−7 to 1.7e−5 m/s, spreading over two orders of magnitude, with an average k of 3.1e−6 m/s, which is in close agreement with the k value of 1.8e−6 m/s, determined by applying the Darcy law to the tank’s outflow rate. The Kolmogorov–Smirnov test revealed that the K–S statistic is larger than the critical value of the K–S statistic at 90, 95, or 99% significance level for the lognormal distribution (Table 2); thus, the hypothesis that the underlying population is normally distributed is acceptable (Sprent 1993). This is represented in the histogram plots of Fig. 4. The normal distribution exhibits a K–S statistic larger than its K–S critical values and an elevated S value that implies a tailing towards higher k values.

Table 2 Summary of statistics for slug test results
Full size table
Fig. 4
figure 4

Frequency k distribution plots of all the 130 slug tests with logarithmic and linear scale for the x axis (left and right, respectively); the lines represent normal distribution curves fitted to the data

Full size image

Despite the considerable number of k measurements within such a small domain, it is difficult to infer trends or deviations from the normal distribution (see Online Resources 2 for the complete data set of k values). Although a lognormal distribution should be assumed and the calculated sample mean and variance are representative of the population distribution, in order to understand if there was a preferential distribution in hydraulic conductivity, a three-dimensional plot depicting isosurfaces of k equal to 5e−6 m/s was created with the inverse distance interpolator. Isosurfaces (Fig. 5) demarcate a large zone extending from the center to the right side of the outflow wall where groundwater should migrate faster than in other zones: three more zones of high permeability are also visible in Fig. 5, but they are smaller and probably not connected with the main one.

Fig. 5
figure 5

Three dimensional representation of isosurfaces of k equal to 5e−6 m/s; the shaded volumes indicate the gravel walls

Full size image

Tank characterization was completed by monitoring piezometric heads every 2 months. Steady-state flow was inferred as heads variation was null or within the measurement error (±1 mm). As a standard, the heads’ contour recorded in April 2007 is shown in Fig. 6. It is clear that, although piezometric contour lines lie roughly perpendicular to the no-flow boundaries of the tank, the head gradient is not constant as a result of local k heterogeneities.

Fig. 6
figure 6

Contour plot of piezometric heads in m; the shaded areas indicate the gravel walls

Full size image

In fact, the head gradient flattens in correspondence to high k zones (Fig. 5): this is consistent with the Darcy law which states that for a given flux per unit area the head gradient is steeper for lower k values whereas it is flatter for higher k values (Fetter 1999).

Evolution of solute concentration distribution

The initial average composition of groundwater in the tank was quite different from the composition of the flushing water, as can be observed in Table 3, although in both waters oxic conditions prevail with neutral-basic pH. The tank monitoring revealed that chloride could not be used as a tracer for the experiment, because it was present at the same concentration in the inflow-purified tap water (28–32 mg/l) and in the resident pore water (30–33 mg/l). Instead, bromide was present at trace concentrations in the inflow water (0.03–0.06 mg/l), while in the resident pore water (0.4–0.26 mg/l) it was nearly one order of magnitude higher. Figure 7 shows a continuous decrease of Br− concentration over 600 days, when it balanced the concentration of inflow water. This was considered to be the time required to flush resident pore water from the tank. Excluding physical non-equilibrium or sorption processes, at the applied flow rate of 26 l/day, this time corresponds to approximately 2 tanks’ pore volumes. Considering the Darcy’s law in the form of

Table 3 Average measured initial pore water chemistry in the tank and purified tap water employed to flush the tank, concentrations in mg/l
Full size table
Fig. 7
figure 7

Time series plots of selected inorganic and organic species at the tank’s outflow; error bars are not shown for clarity

Full size image
$$ v = {\frac{{k \times \overrightarrow {i} }}{{n_{\text{e}} }}} $$
(5)

v is the average pore water velocity (L T−1), i is the head gradient (L L−1), and n e is the effective porosity (L3L−3).

Referring to the average parameters listed in Table 1 and to an average head gradient of 7% in the sediments, and 0.1‰ in the gravel walls, the mean pore water velocity is approximately 2.6 ± 0.1 cm/day in the sediments and 3.8 ± 0.1 cm/day in the gravel walls, which, when multiplied by the length of 7 and 1 m, respectively, gives a total travel time of 590 ± 20 days, confirming that Br− can be used as a conservative tracer.

The behavior of NO3 − was different from Br−, as a decrease in concentration was recorded from the beginning of the experiment and NO3 − reached the concentration below detection limit in approximately 500 days (Fig. 7). Additionally, NO3 − concentration in the inflow water fluctuated from 6 to 10 mg/l, but after 500 days NO3 − did not breakthrough again, confirming the capacity of sediments to naturally attenuate this inorganic contaminant. The decrease of NO3 − concentration, compared with the conservative behavior of Br−, indicates a partial consumption by denitrifying bacteria, since other inorganic processes (like pyrite oxidation) were unlikely to happen, as an increase of SO4 2− concentration was not recorded (Fig. 6). The low NO3 − degradation rate (see Table 3) may be attributed to a low denitrification activity which started about 150 days after the beginning of the experiment, when oxygen became limited, as evidenced both in the outflowing water and in the piezometers (data not shown). The organic matter content of these natural sediments is generally low because of their high depositional energy (Table 1) and is also explicable by the local agricultural practices, based on synthetic inorganic fertilizers during the past 40 years. Labile organic matter and its mineralization by-products acetate and formate remained low throughout the entire experiment (Fig. 7), about two orders of magnitude lower than the available electron acceptors, as nitrate for denitrification and sulfate for sulfate reduction (Christensen et al. 2000).

The initial increase of acetate and formate concentration (Fig. 7) is imputable to an increase of organic matter mineralization. This was probably due to sediment mixing and oxygenation during excavation, transport, and filling of the tank. Acetate and formate rapidly declined below detection limit, both for dilution with the inflowing purified tap water and for complete mineralization, with nitrate as the main electron acceptor. However, the limited availability of acetate, formate, and organic matter in general, prevented a shift towards more reducing conditions within the tank; this is confirmed by slightly negative Redox potentials, measured at the outflow and in the piezometers and by the conservative behavior of sulfate, proving the absence of sulfate reduction. Like acetate and formate, ammonium increased during the first 80 days, reaching a peak of 0.4 mg/l at the second sampling. Nitrite trend was very similar in amplitude and shape but starting with null concentration at the beginning of the experiment, when sediment mixing and oxygenation could have enhanced denitrification.

NO2 − showed a peak 100 days later than ammonium (Fig. 7). In soils, nitrite may accumulate during nitrification and denitrification if these reactions are limited by electron donor or acceptors (Burns et al. 1996). Oxygen was somewhat limiting in the tank, since it was present in the inflowing tap water (5 ± 0.5 mg/l), but it was never present at the outflow during the whole experiment. This could suggest that NO2 − accumulation may have taken course during nitrification. However, this was unlikely because the measured conditions of low ammonium, low acetate availability, and very high nitrate concentrations, suggested NO3 − as the most probable candidate for NO2 − accumulation (Kelso et al. 1999; Oh and Silverstein 1999).

The slow decrease of Ca2+ compared with the one of Br− was related to dissolution of calcite and dolomite, often involved as secondary reactions during organic matter degradation (Prommer et al. 2007) which were likely to occur during the whole experiment. Dissolution reactions were likely to occur, since SI were positive with calcite and dolomite mean values of 0.12 and 0.51, respectively, in initial tank pore water, indicating precipitation might occur. While, the flushing tap water exhibited negative SI with calcite and dolomite average values of −0.15 and −0.38, these values suggest that the displacement of the initial pore water with tap water may lead to dissolution of carbonates, regularly present in these sediments (Amorosi et al. 2002). This would also explain the maintenance of a constant concentration notwithstanding elution.

Within the tank, evidence of the existence of heterogeneities was clear from the first sampling round (Fig. 8), while the elution of dissolved species was relatively homogeneous at the tank’s outflow (Fig. 7). Indeed, the initial concentration of all major cations and anions was variable within the piezometers grid (Fig. 8), with NO3 − concentration’s range being the most variable (135–380 mg/l) followed by SO4 2− (310–460 mg/l).

Fig. 8
figure 8

Isoconcentration surface maps of selected inorganic species within the tank domain; in every graph, the initial concentration map is transparent and the concentration map after 500 days is opaque

Full size image

No correlation was found between silt and clay contents and the initial dissolved concentrations in each piezometer. Results in Fig. 8 show that Br− was not flushed homogeneously, since the central portion of the tank had the lowest concentrations in accordance with the k field representation (Fig. 5). Whereas the left side maintained high concentrations due to the low permeability of sediments, which slowed down the flushing of groundwater. NO3 − concentration recorded in piezometers after 500 days was also very low or below detection limits, except for the left portion of the tank near the outflow. NO3 − persisted in this low-k zone characterized by low groundwater flow, as confirmed by the Br− distribution after 500 days. It is also noticeable that although the NO3 − concentration at the tank outflow extinguished after 500 days, within the tank NO3 − was still present and it took more than 750 days to disappear.

SO4 2− and Ca2+ showed a similar spatial pattern (Fig. 8), both exhibiting the lowest concentrations in the central portion of the tank. However, they did not approach the inflowing water concentration, except for a narrow zone along the tank, located in the center of the Y axis (Fig. 8), confirming a slow release of these ions into the groundwater.

Comparison with column results

Column experiments were conducted using Perspex columns with an inner diameter of 0.1 m and a height of 1 m filled with sediments collected from the same site and flushed by means of a constant head, a complete description can be found in Mastrocicco et al. (2009). The comparison was performed against the column experiment which exhibited analogous k value and texture to the tank’s sediments.

In the column, initial NO3 − concentration was 199 mg/l, k value was 4.8e−6 m/s, porosity 0.42 and the calculated clean-up time at field conditions was 850 days (Mastrocicco et al. 2009). To compare clean-up time at field conditions estimated via column results with the ones gained on the basis of the tank experiment results, the CXTFIT model was applied. Model results (Table 3) indicate that the elution of Br− from the tank outflow can be approximated by the equilibrium CDE, while in the column it can only be approximated by non-equilibrium CDE, to avoid an unrealistic D value (not shown). Physical non-equilibrium was instilled by the elevated pore water velocity employed, which enhanced the preferential flow through macropores. The mobile water content within the column is considerably lower than the total water content and the low value of α limited the exchange between mobile and immobile regions, producing a Br− decline from the beginning of the elution (Fig. 9).

Fig. 9
figure 9

Simulated NO3 − (solid line) and Br− (dashed line) concentrations and observed NO3 − (filled circles) and Br− (open circles) concentrations for the tank and column outflow

Full size image

Calculated degradation rates for NO3 − in the tank and column were of the same order of magnitude, although within the column μ m was more than three times higher than in the tank. The calculated degradation rates in both tank and column experiments were very low; this was due to scarce organic substrate availability, since about 40 years ago organic fertilization ceased and labile organic load underwent a substantial reduction and disproportion with respect to the total oxidative capacity. The extremely small μ im values denote the limited role of immobile water in the total NO3 − degradation rate, which on the contrary happens in the mobile region. In addition, the limited availability of biodegradable organic matter within the tank inhibited further denitrification, as shown by very low NO3 − degradation rates. In all cases, a good agreement between calculated and observed concentrations was obtained, with R2 ranging from 0.969 to 0.996 (Table 3).

Keeping constant the parameters of the calibrated model, an elution scenario within the tank was implemented using the same initial NO3 − concentration of the column experiment (199 mg/l) and the regional head gradient of the local unconfined aquifer (1‰), where the sediment were collected for column and tank experiments. The average time required to clean up groundwater from fertilizers by-products was calculated in approximately 2,510 days for NO3 − (Table 4).

Table 4 Summary of experimental and modeled elution curves
Full size table

The comparison of column and tank experiments to estimate the time required to clean up from NO3 − the same aquifer material at field conditions leads to quite dissimilar results. Namely, the estimated NO3 − clean-up time from the tank experiment is approximately three times greater than that for the column. This was imputable to two main reasons: (1) the flow velocity was accelerated in the column experiment in comparison with the tank conditions; this probably enhanced the removal from zones of low k present in the column. And (2) the heterogeneities present in complex alluvial depositional environments are difficult or impossible to reproduce in column experiments. As described in Chap. 3.2, the role of low-k zones is to act as secondary sources of NO3 − that are slowly released into more permeable zones, where the groundwater flux is focused. This process allowed maintaining elevated concentrations in some zones within the tank, although in the tank outflow groundwater mixing led to decreased NO3 − concentration below detection limits.

Conclusions

A large tank experiment was implemented to evaluate the clean-up time of a shallow unconfined aquifer contaminated by fertilizer by-products. Via grain-size distribution analysis and geological survey, the depositional environment was identified as a crevasse splay, characterized by a large spectrum of textures and a heterogeneous k field. To overcome the common problems associated with permeability testing on recovered samples, multi-level slug tests were performed. This technique allowed a precise three-dimensional reconstruction of the k field and its local heterogeneities within the tank.

NO3 − degradation within the tank and column was limited as witnessed by their low degradation rates estimated via model calibration and by scarce organic substrate availability as shown by the very low concentration of acetate and formate. Both the tank and column experiments on nitrate fate, using the same sediments, confirmed the high vulnerability of unconfined aquifers to this common pollutant and the low natural attenuation potential. Since the heterogeneity linked with geological layering was lost within the tank and column experiments, a direct comparison with field results is not straightforward, but both the laboratory experiments highlight the elevated persistence of NO3 − in these sediments. A methodological implication of this study is that laboratory experiments can provide robust results only when the flow velocity approximates the field conditions and relatively large samples are used, like in tank experiments. This is because considerable heterogeneities present in sediments need to be taken into account when fertilizer’s fate and transport within the saturated zone are studied in complex alluvial depositional environments.