Approccio metodologico per la stima dei volumi di LNAPL recuperabili in falda e per l’analisi del comportamento della fase residuale in laboratorio

NAPLs (non aqueous phase liquids), for their chemical-physical properties, are the most common and harmful contaminants in groundwater as they represent potential long-term sources of contamination and are carcinogenic or toxic for human health (Baciocchi et al., 2010). Based on their relative density with respect to water, NAPLs are classified as DNAPLs (denser than water) and LNAPLs (lighter than water). The former ones are the subject of this study. In the vadose zone, immediately following the release, LNAPLs typically migrate downwards under the influence of gravity. This vertical migration continues only if the capillary forces exceed the residual soil retention capacity (CL:AIRE, 2014). If this happens, LNAPLs will continue to migrate downwards until the water table is reached. Once the LNAPL is in contact with the capillary fringe, the contaminant starts spreading laterally (Baldi and Pacciani, 1997; Brost and DeVaull, 2000), unless sufficient LNAPL potential energy exists for it to displace water and penetrate the water table (CL:AIRE, 2014). LNAPL can enter into the pores and dislocate water only if it reaches and exceeds the entry pressure, which is inversely proportional to the pore throat radius. The entry pressure will be greater, for smaller radius values, due to the inverse relationship between this pressure and pore throat radius. Hence, LNAPL will enter more easily in the large pore. The entry pressure and capillary forces are two of the key factors that control the behaviour of LNAPL and its partitioning in the residual phase (adsorption to solids particles due to capillary forces) and free phase (a liquid separated phase immiscible in water). Conversely, the partitioning in vapour (in the vadose zone) and dissolved phase (which forms plumes of contaminants in the phreatic zone) is controlled by Henry’s constant and solubility (Dippenaar et al., 2005; Jeong and Charbeneau, 2014; Pankow and Cherry, 1996). The free and residual phase represent the 99% of LNAPL in the subsoil and are difficult to detect and quantify, so they are subject of this study in order also to understand the fate and impact of these contaminants on drinkable water quality. In order to reach this focus, a contaminated site, located in Sicily (Southern Italy), which is characterized by a diffused contamination by LNAPLs in free and dissolved phase, chlorinated solvents, agricultural fertilizers and heavy metals, was used as a tool to estimate the volume of the free phase. In particular, the volume of the free phase present in the site has been quantified using two different conceptual models: the Pancake Model and the Vertical Equilibrium Model. According to the Pancake Model, the migration of LNAPL to the water table and its lateral spreading through the capillary fringe creates a buoyant pool with uniform and constant saturation (Baldi and Pacciani, 1997; CL:AIRE, 2014). Since hydrocarbons are assumed immiscible in water, the free phase is suspended on the capillary fringe above the water table and so the thickness measured in the monitoring well is an apparent thickness (Dippenaar et al., 2005; Gruszczenski, 1987; Testa and Paczkowski, 1989). The difference between apparent and real thickness is due to the absence of the capillary fringe in the monitoring well. In fact, this absence conducts to a lower level of water table in the well and so free product flows more easily in the monitoring well creating an exaggerated thickness of supernatant (Hughes et al., 1988). In addition, the weight of free phase depresses above the water table in the monitoring well facilitating the flow of free product in the well. Since according to the Pancake Model there is a difference between the apparent and the real thickness, it is necessary to correct the thickness measured in the well. There are different ways to correct this measure such as empirical factors (e.g. four according to De Pastrovich et al., 1979), or factors derived by field test (recharge test and baildown test). In the studied contaminated site, baildown tests have been used to calculate the exaggeration factor in order to estimate the real thickness of free phase in the aquifer. The baildown test consists of pumping only the supernatant (Hughes et al, 1988) or supernatant and water (Gruszczenski, 1987) from the well. Initially, before the pumping begins, it is necessary to measure the apparent thickness of the product in the well. During the test, the level of air/LNAPL interface and LNAPL/water interface are measured and their difference provides the free product thickness. The real thickness is considered as the measured thickness corresponding to the inflection point in the graph LNAPL/water interface vs time. The volume of the free product existing in the site at June 2013, has been calculated using the Pancake Model and Thiessen polygons and grid at regular square mesh (100 m x 100 m and 200 m x 200 m). First, we have identified the monitoring wells with supernatant and the measured apparent thickness, then we have subdivided the monitoring wells based on the type of free product detected (diesel with more of 70% of C10-C30, gasoline with more of 70% of C6-C9 and mixtures of gasoline and diesel). Based on this first evaluation, different exaggeration factors, obtained by baildown tests, have been used to calculate the real thickness. Once obtained the real thickness for every monitoring well, the specific volume has been calculated as the product of the real thickness and the effective porosity (that is 0.25 for sands and 0.20 for silty sands). The specific volume has been then multiplied for the area associated at each monitoring point using the Thiessen polygons and the grid at regular mesh. The total volume of the free product obtained using the Thiessen polygons was about 9000 m3, while using the grid at 200x200 m mesh and 100x100 m mesh the volumes were about 8700 m3 and 4800 m3, respectively. The Vertical Equilibrium Model instead, can been used only when there is not persistence of release of contaminant (Lundegard and Mudford, 1998), i.e. the scenario considered in the study site. In such case, it is assumed that there is not a discrete layer of LNAPL floating on the water table, but that LNAPL can penetrate below the water table. In addition, the pore fraction occupied by product is less than 100% due to the presence of other fluids such as air and water, and the LNAPL saturation varies with the depth (ITRC, 2009; Lundegard and Mudford, 1998). The shape of LNAPL saturation curves can be regular (shark fin) if there is a homogeneous aquifer or irregular if the aquifer is heterogeneous. Hence, according to the Vertical Equilibrium Model, the relationship between the LNAPL thickness in the monitoring well and its specific volume in the aquifer is linked by the capillary properties of the soil and the LNAPL characteristics. Usually, given a certain thickness in the well, the specific volume will be lower if a finer material constitutes the aquifer (Lundegard and Mudford, 1998). Therefore, the Vertical Equilibrium Model requires the knowledge of the saturation profiles to calculate the specific volume. These curves can be obtained in different ways. For instance, they can estimated based on empirical approaches (measuring saturation in core samples), by analytical modelling or with software as LDRM (LNAPL Distribution and Recovery Model, distributed by American Petroleum Institute) which requires information about the characteristics of the LNAPL and the affected aquifer. For the calculation of the LNAPL volume existing in the site at June 2013, we used the Vertical Equilibrium Model by employing the LDRM approach which, based on the input information, provides the specific volume (Dn) and the recoverable specific volume (Rn). Then, also in this case we have used the Thiessen polygons and the grid at regular mesh to estimate the total volume. The results of the calculation are about 5700 m3 for the Thiessen polygons, and about 4300 m3 and 3000 m3 for 200x200 and 100x100 grid at regular mesh, respectively. The recoverable volume calculation shows that about 70% of the product present in the site can be removed by pumping. The comparisons between the results obtained using the two models and different methods to calculate areas, show that there is a difference of thousands of cubic meter between the estimated volumes. In particular, the Vertical Equilibrium Model estimates lower volumes than the Pancake Model; this is probably due to the LNAPL saturation considered in the two models (100% in Pancake Model and less of 100% in Vertical Equilibrium Model). In addition, it was found that for both models the estimated volume was higher for 200x200 regular mesh than for 100x100 regular mesh. The difference between volumes estimated with the Pancake Model and the Vertical Equilibrium Model was about 51% using 200x200 mesh and about 37% using 100x100 mesh. This result allows to suppose that the use of smaller mesh can lead at a reduction of the differences, but at this stage the available data do not permit to use smaller mesh and confirm this theory. The use of the Thiessen polygons, that show a difference of 37% between volumes estimated with the two different models, reveals another problem related to the construction method of these polygons; in fact, since they are created as a function of the distance and spatial distribution of the monitoring wells, they have different shape and dimension and this influences the volume calculation. Monitoring wells with similar specific volume can be associated to polygons with areas much different and so the volume calculated for one well can be very different from that of another well. The comparison of volumes obtained by the Pancake Model considering only the piezometers and those obtained considering the piezometers and the wells confirms the influence of the Thiessen polygons on the volume estimation. In fact, in the first calculation, the volume of supernatant was about 9000 m3, while in the second one instead it was about 6800 m3. In addition to the uncertainty due to the area delimitation, other critical points in the calculation of the LNAPL volume have been encountered, such as 1) measure of the product thickness in the wells, because it can be influenced by the removal of the product that can be done the days before the measurement; 2) presence of different types of product; 3) lack of some site-specific data as α and N parameters of Van Genuchten (API, 2001), the porosity, the irreducible saturation of water and the interfacial tensions. A sensitivity analysis has been done about the porosity and it has revealed that a reduction of this parameter (from 0.25 to 0.20 for sands and from 0.20 to 0.15 for silty sands) provides a reduction of about 20-22% of specific volume for sands and of about 25% for silty sands. The recovery of LNAPL volumes present in the site is ongoing through the existing and new recovery wells. In fact, to date, the realization of 8 new wells and 3 piezometers for the LNAPL recovery have been foreseen. These new wells will accelerate the recovery of free LNAPL until the residual saturation will be reached, i.e. when the remaining volume of LNAPL will not be recoverable and it will remain as residual phase. The residual phase has been the focus of the second part of this work. Indeed, in order to accurately plan and realize an effective remediation coupled with a good cost/benefits ratio, it is necessary to understand its behaviour and interaction with the impacted soil and groundwater. To this end, we have carried out lab-scale column tests using different porous media (glass spheres, sandy soil A with 9% of silt and 1% of clay, sandy soil B with 14% of silt and 2% of clay) and toluene as contaminant. The cylindrical column (12.7 cm x 2.9 cm) was packed, from bottom to top, with 2 cm of large glass spheres (φ=5 mm), 1 cm of small glass spheres (φ=3 mm), 2 glass-microfibre filters (0.7 μm), 6.7 cm of soil (dried in an oven at 110°C) corresponding to approximately 74 g of soil (glass spheres φ=6 mm in the control test), 2 glass-microfibre filters (0.7 μm), 1 cm of small glass spheres and 2 cm of large glass spheres. The column was first saturated, using a peristaltic pump, with demineralized water from the bottom to remove gas bubbles trapped in the porous media and to estimate (based on the water trapped in the column) the pore volume (PV) of the column. Then, toluene was fed into the column to simulate the movement of organic liquid into the saturated zone (U.S. EPA, 1990) until the water present in the column was completely displaced by toluene, indicating complete saturation with the organic liquid (Powers et al.,1992; US. EPA, 1990). The column was then flushed from the bottom with demineralized water at relatively flow rates (from 0.5 to 1 ml/min) for approximately 25 PV to displace the free phase of toluene. After this step, it was possible to calculate the residual saturation, i.e. the saturation at which the NAPL becomes discontinuous and it is immobilized by capillary forces under ambient groundwater flow conditions (Mercer and Cohen, 1990). The residual saturation has been calculated as the ratio between volume of toluene remained in the soil and volume of toluene injected in the soil. Successively, to assess the dissolution kinetics of toluene from the residual phase entrapped in the column, demineralized water was fed into the column and effluent samples were analysed for toluene by static headspace gas chromatography-mass spectrometry (HSS-GC-MS) using fluorobenzene as internal standard. The results of these experiments showed that the residual saturation is function of the grain size distribution; in fact, residual saturations for glass spheres, soil A and soil B were respectively 36%, 70% and 77%. These data showed that residual saturation is inversely proportional to grain size dimension, indeed it increases with decreasing grain size. Dissolution profiles showed that the solubility of toluene in water was never reached during the test, except in some cases in the first samples collected by glass sphere test and soil A test. Conversely, in the soil B test, the solubility was never reached. In addition, it has been observed that the concentration of toluene decreases more quickly with the increase of particle size. This is due to capillary forces that in the finer soil are higher and so hold more toluene conducting to higher residual saturation and lower dissolution of residual phase. Experimental results are also been modelled using a first order kinetic model according to which the cumulative mass released by the soil is: Mout (cum, measured) = Σ Cmeasured,iΔVi (1) where Cmeasured (mg/l) is the toluene concentration measured at the column outlet at the i-th interval and ΔV(l) is the volume of water fed into the column during the i-th interval. This mass was compared with the mass calculated following the traditional approach that considers the concentration of toluene in water as the solubility: Mout (cum, equilibrium) = Σ Csol * ΔV= Csol * npv * PV (2) where Csol (mg/l) is the solubility of the contaminant in the water, ΔV (l) is the volume of water fed into the column during the i-th interval, npv (-) is the number of pore volumes fed into the column and PV (l) is the pore volume of the column. By combining eq. (1) and (2), it is possible to estimate the ratio (CF) between theoretical cumulative mass assuming saturation condition (eq. 2) and the experimental cumulative mass (eq. 1), both calculated for a specific number of pore volumes delivered (npv*): CF (mpv*) =Mout (cum, equilibrium; npv*)/ Mout (cum; npv*) (3) The higher is the CF values obtained through eq. (3), the greater will be the overestimation of the mass release calculated assuming equilibrium conditions between the eluate and the NAPL. The results of modelling show that there is a good correlation between experimental results and the first order model. The assumption according to which the eluate from the residual LNAPL is saturated with toluene may lead to an overestimation of the true dissolved toluene concentration in water. Finally, the experimental results were used also to carry out a risk analysis. In particular, an effective hazard index is directly proportional to hazard index (HI) estimated with traditional approach (ASTM-Risk-based Corrective Action) and inversely proportional to the correction factor (CF). The results of risk analysis showed that the traditional ASTM-RBCA approach can lead to an overestimation of the hazard index for human health and this is particularly significant for sources characterized by lower lengths along the groundwater direction and in scenarios characterized by higher groundwater velocity. Based on these findings, it was also found that the approach proposed in some countries (Carlon, 2007) to estimate the risk to groundwater resources based on conservative concentration thresholds may provide an inaccurate perception of the effective impact on groundwater quality. Indeed, the use of steady-state transport models combined with linear equilibrium partitioning model results in a situation that is representative only of the very early stages after the contamination event. The results indicated that after a relatively short time, the contaminant concentrations in groundwater are expected to decrease by orders of magnitude. This behaviour cannot be properly evaluated by a simple comparison with threshold values, whereas it can be easily accounted for by adopting a risk assessment procedure (especially if coupled with the approach proposed in this work) that calculates the risk based on the effective cumulative dose over the entire exposure duration.