| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
a Environmental Technology Division, Pacific Northwest National Lab., P.O. Box 999, MS K9-33, Richland, WA 99352
b Dep. of Agronomy and Soils, Auburn Univ., Auburn, AL 36849
c Environmental Molecular Sciences Lab., Pacific Northwest National Lab., P.O. Box 999, Richland, WA 99352
* Corresponding author (mart.oostrom{at}pnl.gov).
All rights reserved. No part of this periodical may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher.
Received 15 December 2006.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONFlow Behavior ExperimentsNAPL Saturation Imaging...NAPL Detection and...Conclusions and Research...REFERENCES |
|---|
Abbreviations: AOI, area of interest • CT, carbon tetrachloride • DBP, dibutyl phthalate • DCA, 1,2-dichloroethane • DCM, dichloromethane • DNAPL. dense nonaqueous phase liquid • GPR, ground-penetration radar • LNAPL, light nonaqueous phase liquid • LRM, light reflection method • LTM, light transmission method • MIAM, multispectral image analysis method • NAPL, nonaqueous phase liquid • PCE, tetrachloroethene • RGB, red, green, and blue • STOMP, Subsurface Transport Over Multiple Phases • TCA, trichloroethane • TCE, trichloroethene
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONFlow Behavior ExperimentsNAPL Saturation Imaging...NAPL Detection and...Conclusions and Research...REFERENCES |
|---|
This review describes two- and three-dimensional bench-scale aquifer model experiments involving NAPLs. The aquifer models are often referred to as flow cells (Oostrom et al., 2006a), while the term intermediate-scale is typically used to denote the experimental scale. Lenhard et al. (1995) defined the conditions needed before an experiment can be classified as "intermediate-scale": (i) The experimental configuration has to allow small-scale processes to manifest themselves at a larger scale so that their relative contributions to flow and transport phenomena can be studied and quantified, (ii) the size of the experiment has to be small enough for the environment to be controlled, and (iii) the experimental-cell dimensions have to be compatible with measurement and sampling techniques.
One of the main advantages of intermediate-scale experiments is that field-scale processes can be mimicked under controlled conditions since NAPL flow in porous media are studied under the same capillary, viscous, and buoyancy forces as full-scale systems. Another use of flow cell experiments is to provide data sets to test and verify numerical flow and transport models. The experiments allow the comparisons between model simulations and experiments to focus on flow processes since the controlled setting and laboratory instrumentation reduce parameter estimation uncertainty.
Chevalier and Petersen (1999) provided an earlier literature review of two-dimensional laboratory experiments involving NAPL flow, transport, and remediation. The authors reviewed about 20 papers, with the most recent contribution from 1997. Oostrom et al. (2006a) reviewed approximately 65 original contributions related to aqueous dissolution and enhanced remediation. The current paper, for which a total of 64 NAPL flow cell research papers were reviewed, can be viewed as a companion paper of Oostrom et al. (2006a) since intermediate-scale experimental papers will be discussed in the areas of NAPL flow behavior, saturation imaging, and detection and quantification with tracers. The flow behavior papers involve investigations of NAPL flow and transport phenomena. The NAPL saturation imaging category discusses experiments containing methods to produce NAPL saturation images. The final category covers papers investigating laboratory efforts to detect and quantify NAPL with tracer techniques.
The main characteristics of the experiments are listed in Table 1. For our two companion review papers, the considered flow cell experiments are all two- or three-dimensional. One-dimensional experiments are not included. Another experimental criterion that had to be met before an experiment was considered to be a multidimensional, multifluid, intermediate-scale experiment was that liquid NAPL had to be present in the porous media at some point during an experiment. For that reason, the flow experiment by Lenhard et al. (1995) discussing organic vapor behavior was not considered because the trichloroethene (TCE) source was located outside the porous medium. However, the vapor flow and transport experiments by Johnson et al. (1992) were included because dense nonaqueous phase liquid (DNAPL) was emplaced in the porous medium. In addition, if experiments described in a conference paper were also published in a journal paper, only the journal paper is listed in Table 1.
|
|
Flow Behavior Experiments |
|---|
|
TOPABSTRACTINTRODUCTIONFlow Behavior ExperimentsNAPL Saturation Imaging...NAPL Detection and...Conclusions and Research...REFERENCES |
|---|
Qualitative Infiltration and Redistribution Experiments
LNAPL Experiments
The flow cell experiments described by Schwille (1967) were the first to provide subsurface visualization of LNAPL behavior. Catalan and Dullien (1995) studied the recovery of previously entrapped LNAPL through gravity drainage and pumping. Qualitative experiments investigating LNAPL movement near a sloping coarse-grained sand layer in an unsaturated system were reported by Schroth et al. (1998a). McDowell and Powers (2003) conducted experiments to study infiltration and redistribution of ethanol-blended gasoline in homogeneous sands.
In several flow cell experiments, Schwille (1967) observed various aspects of kerosene and heating oil behavior in the unsaturated zone and capillary fringe. Infiltration and redistribution drawings are shown of LNAPL in homogenous and layered systems, with or without moving groundwater. Although most of the results are currently common knowledge, Schwille's (1967) visualization experiments have had a profound impact on the understanding of subsurface LNAPL behavior.
Recovery of previously entrapped LNAPL through water table lowering and active pumping was studied in two flow cell experiments by Catalan and Dullien (1995). In the experiments, a volume of the LNAPL Soltrol 100 was injected into a homogeneous and a heterogeneous flow cell packing with a water table close to the bottom of the cell. After a redistribution phase, the water table was raised to entrap all the injected LNAPL. The next step of the experiments was to lower the water table again to its original position and to subsequently remove free NAPL using a pump. The recovery aspects of this paper were described in Oostrom et al. (2006a). The infiltration and redistribution characteristics of the LNAPL spill in the homogeneous packing were similar to what has been observed by other researchers (e.g., Schwille, 1967). The entrapped LNAPL could be removed easily by water table lowering followed by pumping. The observations for the heterogeneous experiments were totally different. The spill in this experiment, with several fine- and coarse-grained lenses in otherwise medium-grained sand, resulted in considerable amounts of the LNAPL moving into fine-grained unsaturated lenses and bypassing coarse-grained lenses. Due to the capillary barrier effect at the lower interface of these fine-grained sand lenses, the LNAPL remained in these lenses during this stage of the experiment. During the water table rise, some LNAPL was released from these fine-grained lenses, whereas substantial accumulation occurred in coarse-grained lenses. For these lenses, LNAPL did not move into the overlying medium-grained sand because the entry pressure was not exceeded. During the subsequent gravity drainage of the flow cell, the LNAPL saturations in the lenses could not substantially be reduced, and large volume of the originally injected LNAPL remained in macro-entrapped form.
Schroth et al. (1998a) performed qualitative experiments investigating LNAPL movement near a sloping coarse-grained sand layer in a fine-grained sand matrix. The coarse-grained sand layer was 5 cm high, and the interfaces had a slope of 21.8°. A water table was located near the bottom of the flow cell, and the lens was unsaturated. A water manifold applied water to the flow cell at a constant rate. Due to the slope of the layer, the water saturations in each experiment were higher at the lower end of the layer. The LNAPL Soltrol 220 was injected at two surface locations after steady-state water flow was established. The first spill was introduced at a location above the lower end of the layer, whereas the second spill was injected 1 d later at a point above the upper end of the layer. The experimental results indicate that that LNAPL behavior near textural interfaces is a strong function of water saturation above the interface. A total of three distinct LNAPL flow patterns were identified for a range of water saturations. Full diversion was observed when the water saturation was low. For intermediate saturations, partial penetration into the layers was observed, whereas for very high water saturations full diversion of LNAPL was again being observed.
A contemporary issue was addressed by McDowell and Powers (2003) through experiments looking at the behavior of ethanol-blended gasoline in unsaturated porous media. The addition of ethanol to gasoline affects the infiltration, redistribution, and dissolution of gasoline in several ways (McDowell and Powers, 2003). For instance, the hydrophilic nature of ethanol will result in an almost complete partitioning into the aqueous phase. Furthermore, the solubility of benzene, toluene, xylene, and ethyl-benzene, and xylene (BTEX) components will increase through a phenomenon frequently indicated as cosolvency. In addition, considerable reductions in surface and interfacial tensions can be expected when ethanol is added to a gasoline. The flow cell experiments were designed to compare the infiltration behavior of standard 87 grade gasoline and a 10% ethanol– 90% gasoline mixture. The gasoline and the ethanol were dyed with oil red O and fluorescein, respectively, to allow for visualization of the expected fluid separation. The gasoline spill behavior in the unsaturated zone was according to expectation. The gasoline–alcohol mix migrated through the unsaturated zone differently. Ethanol quickly partitioned into the residual water in the vadose zone, creating regions of high ethanol concentrations. Ethanol-depleted gasoline continued to move downward to the capillary fringe, where a lens subsequently developed, similar to what was observed for the gasoline-only experiments. For larger spill sizes, water containing ethanol moved into the capillary fringe and resulted in obvious interfacial tension lowering. The authors commented that current mathematical models assume that the composition of gasoline at the capillary fringe has the same ethanol content as originally present in the spill. The experimental results shown by McDowell and Powers (2003) contradict that assumption since they show that ethanol transport to the saturated zone is spread out over a long time due to the retention and slow release of ethanol from the unsaturated zone.
DNAPL Experiments
Basic DNAPL behavior in variably saturated porous media was explained and demonstrated using a flow cell experiment in Schwille (1988). The experiment by Kueper et al. (1989) studied tetrachloroethene (PCE) infiltration into a heterogeneous, saturated system, and the results were explained based on the retention properties of the used porous media. Visualization experiments of TCE movement in a porous medium system mimicking a fluvial depositional environment were documented by Conrad et al. (2002). Fure et al. (2006) infiltrated TCE and 1,2-dichloroethane (DCA) into flow cells with discrete and continuous, nonrectangular, low permeability lenses in studies designed to link DNAPL architecture and flux response. Although no DNAPL saturations were obtained directly, photographs were analyzed to indicate zones with and without DNAPL presence.
Schwille's (1988) book on DNAPL experiments in micromodels, columns, flow cells, and lysimeters has had a large impact on the scientific community interested in subsurface multifluid flow science. The experiments documented by Schwille have contributed greatly to the understanding of subsurface DNAPL flow behavior. The book contains many drawings explaining conceptual models and several illustrative photographs. In the book, a total of four flow cell experiments in two different flow cells with variably saturated porous media are discussed. In a large flow cell, PCE had infiltrated from a point source and from a sheetlike source with a length of 1 m. The point source PCE infiltration into the large flow cell resulted in a narrow but stable liquid PCE body in the unsaturated zone. The capillary fringe was able to retard the infiltration of PCE somewhat, causing a moderate lateral spreading. After that, the PCE penetrated into the saturated zone quickly. The sheetlike spill into the same flow cell was conducted to simulate a spill resulting from a tank car accident (Schwille, 1988). This infiltration resulted in PCE movement in several separate fingers or channels. The capillary fringe was not able to slow down the vertical movement and PCE appeared at the bottom of the flow cell much faster than for the point source spill. The smaller flow cell was used to infiltrate equal volumes of PCE and the less dense dichloromethane (DCM). The PCE moved down in a narrow band, whereas the DCM showed more capillary spreading in the unsaturated zone. The migration in the saturated zone was similar for both fluids. Schwille (1988) warned against generalizing the observed rapid penetration in both unsaturated and saturated regions after a DNAPL spill and suggested that actual flow behavior will be a strong function of heterogeneity and porous media hydraulic properties.
A laboratory experiment was conducted by Kueper et al. (1989) to qualitatively study the effects of porous media heterogeneity on the migration of PCE in saturated systems. Although the actual flow cell experiment was qualitative, the authors compiled a comprehensive set of hydraulic property values in independent column experiments. With this additional information, Kueper et al. (1989) were the first to explain the PCE flow behavior in terms of nonwetting fluid entry pressures and permeability contrasts at textural interfaces. The infiltrating PCE was found to flow laterally and cascade off lenses too fine to be penetrated. Kueper et al. (1989) explained that the inability to penetrate certain fine-grained sand lenses was due to the fact that the PCE could not generate a capillary pressure larger than the PCE entry pressure of these sands. In other areas of the flow cell, initial PCE lateral flow and pooling was observed on finer-grained lenses until the saturations were large enough to facilitate infiltration into these lenses. Kueper et al. (1989) suggested that similar experiments should be conducted, but with quantitative measurements of fluid saturations and/or pressures. They noted that of particular interest would be the value of the capillary pressure needed for the entry of a NAPL into a water-saturated sand lens on top of which prior pooling and lateral flow had taken place. Kueper et al. (1989) also commented that their results may not be applicable to field conditions with more well-graded materials.
Experiments to visualize remediation of infiltrated and redistributed TCE in a porous medium system mimicking a fluvial depositional environment were documented by Conrad et al. (2002). The remediation component of the experiments was discussed in Oostrom et al. (2006a). The initial conditions for each remediation experiment were established by injection of TCE into the flow cell from a small source area. The authors noted that heterogeneities in the porous medium structure resulted in TCE occupying a series of pools connected by fingers. Typically, a gravity-driven finger moved down until a capillary barrier was encountered. Continued invasion then caused a region of high saturation to form on top of the barrier until a sufficient pressure had built up to overcome the capillary pressure resisting displacement (Conrad et al., 2002). At that time, new fingers developed that moved downward to the next capillary barrier. The description provided by Conrad et al. (2002) of the displacement processes is similar to the explanations put forward by Kueper et al. (1989).
Fure et al. (2006) conducted four experiments to investigate the relationship between DNAPL (TCE and DCA) mass reduction and contaminant mass flux. The flow cells were packed with 40/50 and 50/70 Accusand lenses in an otherwise 20/30 Accusand matrix. The DNAPL was injected from the top or directly into the low-permeability layers. When injected from the top, the DNAPL was not able to move into the lower-permeability lenses. After equilibrium conditions were obtained, the flow cells were water-flushed using pore water velocities of approximately 20 cm h–1. Photographs obtained during the dissolution experiments were analyzed using an image analysis method to produce binary DNAPL saturation maps with 0.1 cm2 grid cells. The resulting binary maps provide clear, although qualitative, descriptions of the dissolution process for the various packings. The authors concluded that for their experimental systems, the relationship between mass reduction and contaminant mass flux was primarily controlled by the DNAPL architecture. In their analysis, the DNAPL source zones were divided into a collection of streamtubes where the variability in DNAPL saturation was integrated and transformed into an effective DNAPL saturation for each tube (Fure et al., 2006).
Quantitative Infiltration and Redistribution Experiments without Numerical Modeling
LNAPL Experiments
Van Geel and Sykes (1994) conducted a simple heptane infiltration experiment to test LNAPL and water pressure transducers and a light reflection imaging technique to obtain NAPL saturations. LNAPL infiltration and redistribution in variably-saturated systems with moving ground water were reported by Illangasekare et al. (1995a). Kechavarzi et al. (2000) conducted LNAPL infiltration experiments in homogeneous and heterogeneous systems in support of developing a multispectral image analysis method. One of these experiments was revisited by Kechavarzi et al. (2005) using measured data that were not provided by Kechavarzi et al. (2000). Soltrol 220 infiltration behavior, observed when establishing initial conditions for solute tracer experiments, was discussed by Barth et al. (2003).
Van Geel and Sykes (1994) reported a heptane spill in a homogenous system. The heptane flow behavior was documented through an image analysis technique that is discussed in the section "NAPL Saturation Imaging Techniques" below. The flow cell was also equipped with several water and NAPL pressure transducers. During infiltration the LNAPL front remained stable, and no fingering was observed. The LNAPL continued migrating downward until an apparently stable lens configuration was established in the capillary fringe. The NAPL pressures at the locations with hydrophobic tensiometers showed rapid increases on arrival of the NAPL and these responses were consistent with visual observations of the heptane NAPL body.
The experiments described by Illangasekare et al. (1995a) involved LNAPL infiltration into homogenous and layered systems with moving groundwater due to an imposed hydraulic gradient. Since the experiments were conducted in fairly large flow cells, the spill volumes were substantial (8–9 L). A dual-energy gamma radiation system was used to measure LNAPL saturations at selected vertical transects. The experiments in homogeneous systems showed considerable depression of the water table during LNAPL infiltration. After infiltration ceased, a reduction in pressure moved mobile LNAPL upward, leaving behind entrapped fluid below the water table. Downstream movement primarily occurred in the capillary fringe. In the heterogeneous experiments, fine-grained and coarse-grained curvilinear-shaped lenses were placed in the upper part of the saturated regions. Downward moving LNAPL entered the coarse-grained lenses but moved around the fine-grained lenses. The results were consistent with multifluid displacement theory (Illangasekare et al., 1995a).
Kechavarzi et al. (2000) conducted three experiments in homogeneous and heterogeneous systems to provide data to test a multispectral image analysis method. The imaging method is discussed in more detail in the section "NAPL Saturation Imaging Techniques" below. The LNAPL body in the homogeneous sand behaved according to expectation, and its migration characteristics are consistent with what has been observed in other LNAPL infiltration studies (Schwille, 1967; Van Geel and Sykes, 1994). In the heterogeneous experiment with a silt layer in the vadose zone, the infiltrated LNAPL moved around this layer. This behavior appears to contradict the findings in Catalan and Dullien (1995), who observed increased infiltration into fine-grained materials in the unsaturated part of their flow cell. However, the difference in behavior is consistent with the water saturations in the fine-grained layers. In Kechavarzi et al. (2000), the silt layer was fully saturated, whereas in Catalan and Dullien (1995), the fine-grained layers were unsaturated. In the former experiments, the LNAPL had to exceed the entry pressure of the silt before it could move into this material. In the Catalan and Dullien (1995) experiment, LNAPL was able to move freely into the relatively dry fine-grained porous media.
In a later study, Kechavarzi et al. (2005) provided more experimental detail on an LNAPL infiltration experiment in homogeneous porous media, previously presented by Kechavarzi et al. (2000). The additional information includes miniature resistivity data for water saturation and hydrophobic and hydrophilic tensiometers to measure fluid pressures. The authors claimed that the combination of pressure data and saturation data from the image analysis provided quantitative data essential for testing the predictive capability of numerical models. The pressure data were used to examine nonhysteretic saturation–capillary pressure relationships that are commonly used in multifluid flow simulators. The authors concluded that the empirical relationships were not accurate enough because residual saturation formation in the vadose zone was not included. They also commented that the inclusion of residual saturation is important because of its potential to act as a persistent source of contamination.
The tracer experiments to detect entrapped LNAPL by Barth et al. (2003) are discussed in the section "NAPL Detection and Quantification with Tracer Methods" below. Since the authors used an LNAPL (Soltrol 220) to mimic DNAPL behavior during the establishment of the initial conditions for the tracer experiment, the LNAPL emplacement phenomena are mentioned here. In the highly heterogeneous system, the LNAPL "spills" were injected from the bottom of the flow cell to allow capillary and buoyancy forces to redistribute an NAPL in a similar manner as a DNAPL sinking below the water table (Barth et al., 2003). During the upward infiltration, the LNAPL moved from coarse-grained to coarse-grained layers through fingers in finer-grained materials. These visual observations were confirmed by dual-energy gamma radiation measurements. This behavior, pooling in coarse-grained materials and fingering in finer-grained sands, is similar to the observations by Conrad et al. (2002) for the downward infiltration of TCE. Barth et al. (2003) stated that the overall resulting distribution in coarse-grained layers was ultimately determined by instabilities induced by variability at the pore scale.
DNAPL Experiments
The experiments reported by Illangasekare et al. (1995b) focused on the development of saturation data sets of DNAPL infiltrating into various heterogeneous systems. Hofstee et al. (1998b) studied the behavior of PCE in a stratified porous medium to determine why a DNAPL does not move into a fine-grained porous medium. The work by Coumoulos et al. (2000) compared infiltration at 1 g versus 20 g using a centrifuge test. Jalbert et al. (2003) emplaced PCE in a layered system to establish initial conditions for a partitioning tracer test. Oostrom et al. (2005) investigated the behavior of a multicomponent DNAPL containing carbon tetrachloride (CT) in a heterogeneous packing with several sloped layers. In all of the experiments in this section, a dual-energy gamma radiation system was used to determine NAPL saturation.
Detailed experiments involving infiltration of DNAPL with a high (dibutyl phthalate, DBP) and low viscosity (trichloroethane, TCA) into layered unsaturated and saturated systems were reported by Illangasekare et al. (1995b). The major goal of this study was to develop data sets that could be used for numerical model testing and development. The two DBP experiments were completed in unsaturated medium-grained porous media with either a coarse-grained or a fine-grained layer. The experiment with the coarse-grained layer showed virtually no effect on DBP flow, with the majority of the fluid finally ending up in the saturated zone. In contrast, no DBP was able to move to the saturated zone in the experiment with the fine-grained sand layer since considerable DBP accumulation occurred in that lens. Two additional experiments with TCE were completed in unsaturated systems with alternating 5- or 20-cm layers of fine-grained and medium-grained materials. The results of these experiments show a limited penetration of a spill due to the capillary barriers in the systems. Less vertical infiltration was observed for the experiment with the larger amount of textural interfaces. Illangasekare et al. (1995b) also conducted two experiments with TCA in a saturated medium-grained sand with either a coarse- or a fine-grained single layer. The experiment with the coarse-grained sand layer resulted in rapid infiltration into the coarse-sand layer and some modest pooling on the lower medium-coarse sand textural interface. After saturations increased to sufficient levels, that pool drained into the coarse-grained sand through several fingers. For the experiment with the fine-grained sand layer, no visual penetration of the TCA was observed, although the dual-energy system detected small saturations in the upper part of the layer. Since the entry pressure of the fine sand was too high for TCA penetration to occur, the explanation provided by the authors for the measured TCA saturations in the fine-grained material was related to potential errors in the gamma system.
The flow cell experiment by Hofstee et al. (1998b) may be viewed as the saturated version of the experiment outlined in Hofstee et al. (1998a). The flow cell contained a rather large fine-grained sand layer surrounded by coarse-grained sand. A 2.44 L PCE spill was allowed to infiltrate under constant head conditions from a small source on top of the flow cell. The initial infiltration of PCE was rather unstable, although the observed finger width was much larger than what had been seen in column experiments using the same materials (Hofstee et al., 1998b). The PCE pooled on the fine-grained sand, and later in time, a small volume moved around the lens to the bottom of the cell. The authors explained that the lack of infiltration into the fine-grained sand was the result of the combined effects of its lower permeability and larger entry pressure than for the coarse-grained sand. Although no simulation results were included by Hofstee et al. (1998b), they stated that a nonhysteretic model could not be used for this experiment because of the observed unstable infiltration and PCE entrapment.
Coumoulos et al. (2000) discussed DNAPL behavior in experiments at 1 g in a regular laboratory and at 20 g in a centrifuge. The 1-g experiments with TCA showed that small-scale heterogeneities caused an enhanced lateral spreading in the vadose zone. The authors stressed that better packing procedures could have prevented these artifacts. Fluid saturations in the 20-g experiments were not reported, although fluid pressure transducers were able to respond rapidly to the gravitational acceleration changes. The described 20-g experiments were clearly of exploratory nature and were not discussed in great detail.
The partitioning tracer component of the experiment described by Jalbert et al. (2003) is discussed in the section "NAPL Detection and Quantification with Tracer Methods" below. The initial conditions for the tracer experiment were obtained by injecting 1 L of PCE into saturated medium-grained sand with very fine and fine sand layers. The PCE moved around the very fine sand lens but was able to penetrate into the fine-grained sand layer. These results are consistent with the findings of Kueper et al. (1989) and demonstrate that smaller-grained materials do not always prevent downward movement of a DNAPL. Clearly, the entry pressure of the fine-grained sand was exceeded during the experiment, while the necessary PCE pressure to penetrate into the very find sand lens could not be obtained.
Oostrom et al. (2005) investigated flow behavior of a DNAPL in a heterogeneous, variably saturated porous medium and the removal of the CT component of the DNAPL mixture from a layered porous medium through soil vapor extraction. The remediation part of this paper was discussed in Oostrom et al. (2006b). A dual-energy gamma radiation system was used at various times to nonintrusively determine fluid saturations. The mixture, which contained the volatile organic CT, mimics the DNAPL disposed at the Hanford Site in Washington State. The flow cell was packed with two sloped coarse sand and two sloped silt layers in an otherwise uniform matrix of medium-grained sand. The water table was located 2 cm from the bottom, creating variably saturated conditions. A 0.5-L spill was introduced at the top of the flow cell from a small source area. It was observed that the DNAPL largely bypassed the silt layers but easily moved into the coarse sand layers. Residual DNAPL was formed in the medium-grained sand matrix. The DNAPL caused a distinct reduction of the capillary fringe. Most of the DNAPL ended up in a pool on top of the V-shaped fine sand at the bottom of the flow cell.
Infiltration and Redistribution Experiments with Numerical Modeling
LNAPL Experiments
Hochmuth and Sunada (1985) were the first to compare numerical modeling results to LNAPL behavior in a flow cell. Host-Madsen and Jensen (1992), Oostrom et al. (1997), Simmons et al. (1992), and Van Geel and Sykes (1997) tested numerical simulators against measured LNAPL saturations in homogeneous three-phase systems. Observed LNAPL behavior in systems with sloped layers was simulated by Schroth et al. (1998b) and Wipfler et al. (2004). Oostrom et al. (2006b) investigated the response of a viscous LNAPL to water table fluctuations.
Hochmuth and Sunada (1985) developed a simple numerical model to predict the final disposition of LNAPL lenses. The model was only applicably to completely uniform, coarse-grained porous media. Pores in the numerical domain were assumed to be either fully saturated or be at residual saturation. A comparison with a Soltrol C spill, spiked with transmission oil to enhance color difference with water, showed that the model was able to predict the geometry of the final lens reasonably well. However, the model could not be applied favorably to a LNAPL spill in a finer-grained porous medium since the increased capillary action was not incorporated into the code.
Numerical modeling of a quantitative NAPL flow cell experiment was first described by Host-Madsen and Jensen (1992). They reported two experiments where 4 L of the LNAPL Bayoil 82 was injected into partly saturated medium- and fine-grained sands. A dual-energy gamma radiation system was used to obtain data in a relatively coarse grid of 10 x 10 cm. For the simulations, the numerical model Eclipse (Exploration Consultants, 1984) was used, which was originally designed for petroleum reservoir simulations. The authors modeled the experiments using saturation–capillary pressure (S–P) relations based on the van Genuchten (1980) air–water retention relation, and relative permeability–saturation (k–S) relations based on Mualem's (1976) permeability model. This modeling approach is still currently used in several multifluid flow simulators. Parker et al. (1987) derived the equations for three-phase systems assuming that fluid wettability follows the sequence water > NAPL > air and that the total liquid saturation is a function of the air–NAPL capillary pressure and the water saturation a function of the NAPL–water capillary pressure (Leverett, 1941). The S–P van Genuchten-type relations (van Genuchten, 1980) for water (subscript l) and total liquid (subscript t) effective saturations in three-phase systems are (Parker et al., 1987)
 | [1a] |
 | [1b] |
(m–1 T2 L) and k are curve-shape parameters, m = 1 – 1/k, Pij (M T–2 L–1) is the capillary pressure Pi – Pj for a fluid pair ij, ß is an interfacial-tension based scaling factor (ßij = 
ij/gl; is the interfacial tension, M T–2), and the subscripts g and n denote air and NAPL, respectively. The effective water and total liquid saturations are given by
 | [2a] |
 | [2b] |
 | [3] |
A Soltrol 220 infiltration and redistribution experiment in a homogeneous, variable saturated flow cell was conducted by Oostrom et al. (1997) and compared with results of a simulation with the STOMP (Subsurface Transport Over Multiple Phases) code (White and Oostrom, 2006). As in the experiments described by Host-Madsen and Jensen (1992), a dual-energy gamma radiation system was used to generate NAPL saturation data. The data were used for comparison with simulation results applying both van Genuchten (1980) and Brooks–Corey (Brooks and Corey, 1964) nonhysteretic retention relations in combination with relative permeability equations based on Mualem's (1976) and Burdine's (1953) relations. For three-phase systems, the extended Brooks–Corey nonhysteretic relations for water and total liquid saturation are (Oostrom et al., 1997)
 | [4a] |
 | [4b] |
 | [5] |
The experiments described by Host-Madsen and Jensen (1992) and by Oostrom et al. (1997) were relatively simple since they only involved infiltration and redistribution of a LNAPL spill. As a result, nonhysteretic k–S–P relations could be satisfactorily used to simulate the migration processes. The flow cell experiment conducted by Van Geel and Sykes (1997) was designed to include hysteretic processes like nonwetting fluid entrapment and pore-geometry hysteresis. The experiment consisted of a LNAPL (heptane) infiltration and redistribution component, followed by water table fluctuations. The infiltration and redistribution component was described earlier by Van Geel and Sykes (1994) to demonstrate a light reflection method for LNAPL saturation measurements. The water table fluctuations caused LNAPL entrapment during water table rises and forced the fluids to be on alternating drainage and imbibition S–P paths. The flow cell was equipped with several hydrophilic and hydrophobic porous cups connected to pressure transducers and a data acquisition system. The experiment was modeled with a simulator including pore-geometry and nonwetting fluid entrapment hysteretic S–P relations according to Lenhard (1992), based on the van Genuchten (1980) retention equation, and k–S relations following Lenhard and Parker (1987), based on the Mualem (1976) permeability model. The model parameter values were obtained in independent experiments, including maximum entrapped LNAPL and air saturations and hysteretic pore-shape parameters. A comparison of nonhysteretic and hysteretic model results with experimental data illustrates the importance of fluid entrapment and pore-geometry hysteresis. For instance, the nonhysteretic model significantly overestimated the LNAPL pressures in response to water table elevations, whereas the inclusion of LNAPL entrapment limited the vertical rise of the LNAPL body, resulting in smaller volumes of mobile LNAPL that were able to migrate upward. Based on visual information and fluid pressure data, the authors commented that residual saturation formation in the unsaturated zone might have been an important process not incorporated in the hysteretic model.
In a series of infiltration experiments in unsaturated coarse-grained sands, Simmons et al. (1992) investigated infiltration and redistribution of three LNAPLs with various viscosity and water–NAPL interfacial tensions. The viscosity relative to water of Soltrol 220, mineral oil, and transmission oil were 4.7, 77, and 71, respectively. The water–NAPL interfacial tensions for the water–LNAPL fluid pairs were 0.043, 0.056, and 0.016 Nm–1, respectively. The fluid density and air–LNAPL surface tensions (
30 Nm–1) were similar for the three fluids. Assuming zero contact angles, a NAPL's tendency to spread on a water–air interface can be measured by the spreading coefficient Csp (M T–2)
 | [6] |
lg, ln, and, gn (M T–2) are interfacial tensions at the interfaces between water–air, water–NAPL, and NAPL–air, respectively (Adamson, 1982). According to Eq. [6], both Soltrol 220 and transmission oil are spreading liquids (Csp > 0). However, mineral oil is clearly a nonspreading liquid (Csp < 0) and tends to form discontinuous lenses on top of water–air interfaces. In the experiment, 0.5 L of LNAPL was allowed to infiltrate from a point source into a three-dimensional flow cell. Immediately after infiltration ceased, the flow cell was excavated and samples were obtained. Organic liquid contents were determined by a method that uses a porous polyethylene strip to extract LNAPL from a sample. The results show that the fluids with a positive spreading coefficient (transmission oil and Soltrol 220) spread evenly throughout the pore space. The infiltration time ratio of these fluids was roughly proportional to their viscosities. Although transmission and mineral oils have nearly the same viscosity, their migration behavior was rather different. The mineral oil displayed channeling (fingering) behavior, caused by its negative spreading coefficient (Simmons et al., 1992). The MOFYS code (Kaluarachchi and Parker, 1989) was used to simulate the experiments. When including air entrapment, the infiltration of both spreading LNAPLs was simulated well. The continuum-based model could not describe the infiltration of the mineral oil, indicating a definite limitation of such models. Simmons et al. (1992) stated that the fingering of mineral oil is likely driven by an unexplained interfacial phenomenon that requires further study. In a continuation of the research by Schroth et al. (1998a), Schroth et al. (1998b) investigated LNAPL movement near a sloping coarse-grained sand layer in a fine-grained sand matrix. The multifluid flow system was simulated using the STOMP simulator (White and Oostrom, 2006) with nonhysteretic van Genuchten (1980) and Brooks–Corey (Brooks and Corey, 1964) S–P relations (Eq. [1–5]) in combination with either the Burdine (1953) or the Mualem (1976) permeability model. The experimental results confirmed the findings of Schroth et al. (1998a) that LNAPL behavior near textural interfaces is a strong function of water saturation above the interface. Similar to Schroth et al. (1998a), three distinct LNAPL flow patterns were again identified for a range of water saturations. However, on the basis of data obtained with a light transmission method, water saturations could now be assigned to the flow patterns. Full diversion was observed when the water saturation just above the interface was smaller than 0.45. For saturations between 0.45 and 0.9, partial penetration into the layers was observed. For very high water saturations above 0.9, again full diversion of LNAPL was observed. Model simulations with all nonhysteretic S–P models resulted in reasonable agreement with the water saturations before the LNAPL injections. In general, the Brooks–Corey S-P relations (Brooks and Corey, 1964) in combination with the Burdine (1953) k–S model provided the best results for the three-phase part of the experiments. The authors felt that better comparisons could have been obtained if a fully hysteretic version of the Brooks–Corey S-P model (Brooks and Corey, 1964) would have been available.
Wipfler et al. (2004) also performed LNAPL infiltration experiments in unsaturated systems with sloped layers. The STOMP model (White and Oostrom, 2006) with the van Genuchten (1980) S–P model (Eq. [1–3]) was used to simulate the experiments. They conducted two experiments. In the first experiment, a coarse-grained sand layer was located in a fine-grained matrix, whereas in the second experiment the location of the sands was reversed. In contrast with the experiments by Schroth et al. (1998a,b), no water infiltration was considered from the top. The experiment with the coarse-grained sand layer showed LNAPL spreading above this layer with downward movement along the interface. After a certain travel distance, the NAPL pressure became high enough for infiltration into the layer. No NAPL moved below the lower layer interface with the fine-grained sand. The experiment with the fine-grained sand layer showed immediate penetration into this layer. After a certain distance, the LNAPL moved into the coarse-grained sand below the interface and accumulated in the upper part of the capillary fringe. The behavior of the LNAPL in both experiments was typical for multifluid flow near textural interfaces. The nonhysteretic STOMP simulation was able to capture the main features of the LNAPL migration quite well for both experiments.
Oostrom et al. (2006b) conducted an experiment to investigate the behavior of a viscous LNAPL under variable water table conditions. Two viscous LNAPL volumes (0.4 L) were released, 1 wk apart, from a small source zone on top of the flow cell into a variably saturated, homogeneously packed medium-grained sand. Following a redistribution period of 30 d after the second LNAPL release, the water table was increased 0.5 m in 50 min. After the water table rise, viscous LNAPL behavior was monitored for an additional 45 d. Fluid saturation scans were obtained periodically with a fully automated dual-energy gamma radiation system. Results showed that both spills followed similar paths downward. Within 2 h after the first LNAPL arrival, the capillary fringe was reduced across the cell by approximately 0.04 m (22%). This reduction was directly related to the decrease in the air-water surface tension from 0.072 to 0.057 N m–1. LNAPL drainage from the unsaturated zone was relatively slow, and a considerable residual LNAPL saturation was observed in this zone after 30 d of drainage. Most of the mobile LNAPL moved into the capillary fringe during this period. After the 0.5 m water table rise, the LNAPL moved up in a delayed fashion. The LNAPL used the same path upward as it used coming down during the infiltration phase. After 45 d, the LNAPL had moved up only approximately 0.2 m, and fluid entrapment was limited. The experiment was simulated using the STOMP simulator (White and Oostrom, 2006), including entrapped and residual LNAPL saturation formation. A comparison indicated that the simulator was able to predict the observed phenomena well, including residual saturation formation in the vadose zone, and limited upward LNAPL movement after the water table rise. The results of this experiment show that viscous mobile LNAPL, subject to variable water table conditions, does not necessarily float on the water table and may not appear in observation wells.
DNAPL Experiments
Numerical and experimental comparisons of DNAPL flow in saturated, heterogeneous systems have been reported by Oostrom et al. (1999a), Kamon et al. (2004), and Rathfelder et al. (2003). Experimental and numerical studies of residual NAPL saturation formation have been reported by Hofstee et al. (1998b), Oostrom et al. (2003), and Oostrom and Lenhard (2003). Stephens et al. (1998) conducted simulations to predict DNAPL behavior in a two-aquifer system connected by a fracture, and O'Carroll et al. (2004) focused on the simulation of DNAPL flow behavior in flow cells containing porous media with different wettability.
DNAPL infiltration experiments in combination with numerical simulations were reported by Oostrom et al. (1999a), Rathfelder et al. (2003), and Kamon et al. (2004). Oostrom et al. (1999a) described a flow cell experiment in which 0.5 L of TCE was allowed to infiltrate into a saturated, heterogeneous porous medium system. The flow cell was packed with five lower-permeability lenses in an otherwise coarse-grained sand matrix. A fine-grained sand layer was located at the bottom to prevent the TCE from leaving the flow cell. A dual-energy gamma system was used to determine fluid saturations. The main objectives of the work were to test the STOMP simulator (White and Oostrom, 2006) against observed TCE flow phenomena and to establish initial conditions for remediation using a surfactant solution (Oostrom et al., 1999b). The remediation component of the experiment was discussed in Oostrom et al. (2006a). The injection of the denser and less viscous TCE, compared with water, caused unstable displacement, resulting in fingers with observed widths ranging from 0.3 to 0.8 cm. The TCE pooled on top of a fine-grained lens and was able to penetrate and move through medium-grained lenses. Most of the injected TCE finally collected on top of the fine-grained sand at the bottom of the flow cell. In general terms, the results observed by Oostrom et al. (1999a) were similar to those shown by Kueper et al. (1989), who also noticed pooling on fine-grained materials and infiltration into medium-grained sands. Numerical simulations were conducted to investigate whether the behavior of the TCE near the interfaces could be predicted using a mode allowing for TCE entrapment according to procedures outlined by Kaluarachchi and Parker (1992) based on entrapment principles forwarded by Land (1968). No attempts were made to simulate the observed fingers. Using independently obtained parameter values for the Brooks-Corey saturation-capillary pressure relations (Brooks and Corey, 1964; Eq. [4–5]), the code was able to qualitatively predict the observed behavior at the interfaces. Simulation results suggest that most of the liquid TCE at the lowest interface was in free form, while most of the other TCE was entrapped.
In a smaller flow cell with just one horizontal fine-grained sand layer in a medium-grained sand matrix, Rathfelder et al. (2003) performed PCE infiltration experiments in systems saturated with clean water or surfactant solutions. In the flow cell with clean water, the injected PCE wrapped around the low permeability sand and collected at the bottom of the flow cell. When a surfactant solution was used for the aqueous phase, most of the PCE penetrated and moved through the lens instead of moving around the lens. In this experiment the surfactant-facilitated interfacial tension reductions substantially lowered the capillary resistance to vertical downward migration of the PCE (Rathfelder et al., 2003). The immiscible flow simulator M-Valor (Abriola et al., 1992) was used to simulate the experiments. In general, good agreement was found for the clean water experiments but the model predictions for the surfactant experiments were rather poor. A comparison of nonhysteretic k–S–P models showed that the Brooks–Corey S–P model (Brooks and Corey, 1964) in combination with the Burdine (1953) k–S model provided the best results for both experiments. Additional simulations showed that both grid refinement and the introduction of small-scale packing variability improved the predictions. It should be noted that although the experiments with the surfactant solution might provide data to test a multifluid flow simulator, it is unlikely that a DNAPL spill occurs in a system already containing such a solution.
DNAPL (hydrofluoroether) infiltration and redistribution in a saturated flow cell with two rectangular, horizontal fine-grained sand layers in a medium-grained sand matrix were investigated by Kamon et al. (2004). Both layers had a 10-cm-wide hole downstream of the DNAPL injection well. Two experiments were completed: one with stagnant groundwater and another one with moving groundwater (hydraulic head gradient = 0.033). Several hydrophilic and hydrophobic tensiometers, as well as electric conductivity probes, were installed. The experiments showed that DNAPL moved laterally on the fine-grained sand layers and subsequently downward through the holes in the lenses. In the experiment with the flowing groundwater, more DNAPL moved through the holes, and less spreading was observed on top of the lenses. This result is consistent with the direction of aqueous phase flow, forcing DNAPL to migrate toward the holes. The code NAPL (Guarnaccia et al., 1998), employing the van Genuchten (1980) S–P relations (Eq. [1–3]) was used to simulate both experiments. Although the van Genuchten (1980) model does not recognize a nonwetting fluid entry pressure, no DNAPL was predicted to enter the lenses. This result may be related to low DNAPL fluid pressures during redistribution, to the lower permeability of the lens material, or to the slight slope in the lenses toward the holes, as can be seen in Kamon et al.'s (2004) Fig. 8. In general, the nonhysteretic simulations provided reasonable results, which is somewhat surprising because the experimental data suggest that a considerable fraction of the injected DNAPL became entrapped by the end of the experiment.
At many contaminated sites, NAPLs persist in the vadose zone for long periods of time. This occurs because the permeability of the NAPL becomes negligible at some saturation and downward movement ceases, resulting in residual NAPL. To obtain data that can be used to study residual NAPL saturation and to test corresponding models, transient flow cell experiments were conducted by Hofstee et al. (1998b), Oostrom et al. (2003), and Oostrom and Lenhard (2003). For these experiments, a dual-energy gamma radiation system was used to determine fluid saturations at numerous locations. Hofstee et al. (1998b) based their experimental design on column experiments by Hofstee et al. (1997) which showed residual saturation formation when the nonspreading DNAPL PCE moved through unsaturated porous media. Nonspreading liquids (Csp < 0; Eq. [6]), like PCE with an estimated Csp of –0.0047 N m–1, tend to form discontinuous lenses on top of water–air interfaces. Hofstee et al. (1997) found that these lenses formed below a certain critical PCE saturation, resulting in increased amounts of residual PCE retained in the vadose zone. In the flow cell experiment described in Hofstee et al. (1998b), a fine-grained sand layer was packed between two layers containing coarse-grained sand. A 0.5-L spill of PCE was allowed to infiltrate and redistribute. The final distribution showed considerable residual saturations, particularly in the fine-grained sand, attributed to the nonspreading nature of PCE. The STOMP simulator (White and Oostrom, 2006) was used to predict PCE flow behavior in the flow cell. The simulator, without an algorithm to simulate residual saturation formation, was able to predict the transient PCE saturations in the vadose zone well but was not capable of predicting the final distribution of PCE retained in the flow domain.
In the residual saturation formation study by Oostrom et al. (2003), a rectangular zone of fine-grained sand was packed in an otherwise medium-grained matrix. A limited amount of CT was injected from a small source and allowed to redistribute until a pseudo steady state situation had developed. The experiments clearly demonstrated the formation of residual CT saturations in both sands. The spreading coefficient of CT in contact with clean water is approximately 0.004 N m–1, but the liquid turns from spreading to nonspreading when the water becomes contaminated with dissolved CT. The experiment described by Oostrom and Lenhard (2003) reflects conditions at the Hanford Site in Washington State, where an estimated 363–580 m3 of CT was disposed to the subsurface. A key subsurface feature at the Hanford Site is a sloped Plio-Pleistocene caliche layer, which was reproduced in the experiment as a sloped lens in a medium-grained, uniform, sand matrix. The caliche likely contains considerable amounts of CT. A total of 0.8 L of CT was injected in the flow cell at a rate of 0.5 mL min–1 from a small source area located at the surface. After apparent static conditions were obtained with respect to CT redistribution, saturation measurements indicated that all of the DNAPL that had initially moved into the caliche remained in this layer. Water was subsequently applied to the surface at a constant rate over the full length of the caliche layer to study CT displacement as a result of changing water saturations. Water saturation in the caliche layer rose to as high as 0.91 during water infiltration. Results show that 25% of the DNAPL present in the caliche migrated from this layer as a consequence of water infiltration, while 75% remained in the caliche layer. Simulations with the STOMP simulator (White and Oostrom, 2006) showed the shortcomings of current constitutive relative permeability–saturation–capillary pressure models in both CT experiments. The authors stated that these results indicated that nonspreading behavior of NAPLs should be implemented in simulators to account for the formation of residual saturations. Recently, White et al. (2004) described the implementation of residual saturation formation theories (Van Geel and Roy, 2002; Lenhard et al., 2004) into the STOMP simulator and application to several column experiments. Comparisons between STOMP simulation results and flow cell experimental data from Hofstee et al. (1998b), Oostrom et al. (2003), or Oostrom and Lenhard (2003) have not been reported to date.
Stephens et al. (1998) showed migration of the DNAPL TCA in a system consisting of two unconfined aquifers separated by a silt stone perching layer containing a single fracture. A water table was maintained in each of the two aquifers. The fracture, constructed from two bricks, had an aperture of 0.51 µm. TCA was injected at a constant rate of 5 mL min–1. The TCE migrated rather uniformly through the upper unsaturated zone before entering the capillary fringe. In the saturated zone, extensive fingering was observed. Without any apparent pooling on the bricks, the TCA was observed in the lower aquifer only 30 s after arrival in the upper fracture. Eventually, TCE piled up on the bricks when the fracture was no longer able to keep up with the injection rate. A modified version of the STMVOC code (Falta and Pruess, 1991) was used to simulate the infiltration process. The simulator was able to predict the general behavior in the porous medium and fracture quite well, although the observed fingering behavior was not simulated. The authors concluded that the numerical simulator was capable of accurately predicting DNAPL migration in a complex system with variable saturated conditions.
O'Carroll et al. (2004) completed a two-dimensional flow cell experiment to quantify the effect of spatial wettability variations on PCE migration in a saturated system. According to Craig (1971), wettability refers to the tendency of one fluid to spread or adhere to a solid surface in the presence of another immiscible fluid. The contact angle between the fluid–fluid interface and the solid is typically used as a measure of wettability. O'Carroll et al. (2004) pointed out that wettability may change spatially and temporally. However, for most subsurface applications, the NAPL–water contact angle is assumed to approach 0°, and the surface is said to be strongly water wet. Strongly organic wet conditions occur when the contact angle approaches 180°. In the experiment described by O'Carroll et al. (2004), two organic-wet and three water-wet fine-grained sand lenses were packed in an otherwise medium-grained sand matrix. The organic-wet sands were created by treating the sands with a 5% v/v octadecyltrichlorosilane in ethanol solution. PCE was injected at a point location for 66 min. Although photographs of the PCE distributions after 10, 20, 30, and 60 min were shown, a light transmission technique was used to determine PCE saturations only at t = 30 min. This aspect of the experiment is described in the section "NAPL Saturation Imaging Techniques" below. The photographs show that the organic-wet layers effectively retained the PCE and inhibited downward movement (O'Carroll et al., 2004). However, PCE moved around the water-wet fine-grained sand because its entry pressure was never exceeded. The experiment was simulated using the M-Valor code (Abriola et al., 1992) with nonhysteretic water-wet and organic-wet constitutive k–S–P relations. Since the results were only recorded during the PCE injection stage during which the DNAPL displaced water, the choice of nonhysteretic relations was appropriate. Modified van Genuchten (1980) and Brooks–Corey (Brooks and Corey, 1964) S–P relationships were used, including the addition of a shifting parameter for negative capillary pressures. The parameter values for each model were obtained by curve-fitting experimental column data. The k–S relations, adapted from the water-wet Burdine (1953) and Mualem (1976) models, were obtained from Bradford et al. (1998). As expected, the water-wet simulations failed to predict both PCE migration paths and retention behavior. For the organic-wet simulations, the constitutive relations based on the Burdine (1953) permeability model provided the best results, although both infiltration and spreading behavior in the organic lenses was not well modeled by any of the models. The authors recommended further experiments to investigate the permeability models and capillary spreading in organic-wet materials.
Upscaling Experiments
It has been well documented that heterogeneities on a small scale can dominate multifluid flow (e.g., Illangasekare et al., 1995b). As was noted by Braun et al. (2005), it is impossible to collect all the necessary information about the geological heterogeneities and representing them discretely in numerical models. Instead, upscaling approaches are needed that provide effective parameters for fluid saturation, capillary pressure, and relative permeability. Although multifluid flow upscaling is receiving considerable attention from modelers and theoreticians, hardly any flow cell experiments have been conducted to test and verify procedures. Only Allan et al. (1998) and Imhoff et al. (2003) conducted experiments to support upscaling methods.
The flow cell experiments by Allan et al. (1998) were conducted in a matrix of medium-grained sand with 10% coarse-grained and 10% fine-grained sand lenses. Experiments were conducted in a homogeneous packing, a packing with 10-cm-long horizontal lenses, and a packing with 20-cm-long lenses. Although not specifically mentioned, the lenses had a vertical dimension of 1 cm. A total of 1.5 L of TCE was injected into each saturated packing at a constant rate of 34 mL/min. The experiments were simulated with a discrete model, explicitly representing the heterogeneities, and with two upscaled models. In the first upscaled model, only the permeability was upscaled using a large-scale anisotropic permeability tensor computed by simulating vertical and horizontal permeameter tests. The second model included both upscaled permeabilities and the effect of lateral spreading of a liquid on horizontally layered lenses through a concept called "phase dispersion," introduced by Pruess (1996). This concept is directly related to differences in nonwetting fluid entry pressures for the porous media used in the experiments. The experimental results showed that the TCE migrated downward in fingers that were only visible at the front site at a few locations. Otherwise, TCE collected in the coarse-grained layers and pooled on top of fine-grained layers. Although the discrete continuum-based model was not able the capture the fingering, the authors concluded that the extent of the TCE migration was very well predicted by the model. The model simulations using the upscaled permeabilities underestimated the degree of lateral spreading and overestimated the rate of downward migration. The results of the model using upscaled permeabilities and phase dispersion showed slightly better results in terms of lateral spreading and downward movement. The authors concluded, not surprisingly, that more developments were needed to create meaningful upscaled representations. The experimental data generated by Allan et al. (1998) were compared by Braun et al. (2005) against an upscaled model using the assumption of equilibrium of capillary forces in percolation theory. Braun et al. (2005) commented that their approach yielded results that, in general terms, reproduced the main characteristics of the infiltration and redistribution experiments.
Imhoff et al. (2003) attempted to scale results from a single laboratory experiment to predict migration at larger length and time scales at sites where large volumes of DNAPL were disposed. They used a modified inspectional analysis technique for developing scaling relationships through nondimensionalizing two-phase flow mass balance and constitutive equations. A mixed DNAPL was released in a relatively large flow cell (Table 1), packed with a 12/20 Accusand matrix and a few sloped lower-permeability materials. A total of 2 L was released, and it took only 18.4 min before the DNAPL reached the downstream flow cell boundary. A total of almost 1.5 L moved out of the system during the experiments, and the remaining DNAPL was left in the form of discrete ganglia distributed uniformly in the region where the DNAPL traversed (Imhoff et al., 2003). Interestingly, although a finger wavelength of about 2 cm was predicted, no DNAPL fingering was observed. The authors suspected that capillary action prevented the formation of fingers under the conditions of the experiments. The laboratory experiment was scaled to four field prototypes, representing four different porous media in the field. Results showed disposal volumes ranging from 4.3 x 103 L m–1 source width to 4.3 x 105 Lm–1, disposed over 8.8 d to 24 yr. A comparison with data from actual field sites showed that these ranges were reasonable and that DNAPL migration may continue for considerable time after cessation of disposal (Imhoff et al., 2003). Limitations of the analysis were the two-dimensional nature of the analysis, the requirement that the porous media systems were homogeneous, and the assumption of similarity (Miller and Miller, 1956) for the porous media in the experiments and field prototypes. Given these limitations, Imhoff et al. (2003) suggested use of the scaling technique to explore general characteristics and approximate time and length scales of DNAPL migration in field-scale systems.
Lens and Pool Geometry Experiments
The behavior of LNAPL lenses in general and the vertical lens thickness in particular were experimentally investigated by Pantazidou and Sitar (1993), Schroth et al. (1995), and Chevalier (1998). Miller et al. (2004) reviewed results from these papers and provided an experimentally verified conceptual model for the geometry of both DNAPL pools and LNAPL lenses. Corapcioglu et al. (1996) conducted LNAPL infiltration experiments to test an analytic solution for oil lens thickness based on several simplifying assumptions.
Pantazidou and Sitar (1993) observed kerosene infiltration and redistribution into unsaturated homogeneous and layered systems using medium- and fine-grained sands. In addition, water and NAPL pressures were obtained in some selected experiments. In all experiments, 0.35 L kerosene was injected, which equals about 10% of the pore space above the capillary fringe. Results show that in the upper part of the unsaturated zone, the NAPL moved downward in a regular, circular shape and that finger formation was evident in the region with higher water contents. In none of the experiments, the kerosene entered the capillary fringe. The authors stated that the observed behavior was similar to what was observed by Schwille (1988). After redistribution ceased, the LNAPL lens thickness was recorded and compared with an expression for lens thickness, derived by considering the capillary forces that render the NAPL immobile (Pantazidou and Sitar, 1993). The expression was derived based on the assumptions that NAPL is draining at the top of the lens and that water is draining at the bottom. The expression relating the lens thickness, T (L), to fluid-pair interfacial tensions, pore neck diameter, and position with respect to the water table is (Pantazidou and Sitar, 1993)
 | [7] |
is the fluid density (M L–3), dn the average pore neck diameter (L), g the gravitational acceleration (L T–2), and hl the distance of the bottom of the lens to the water table (L). All other symbols were previously defined. In Eq. [7], the choice of dn is somewhat arbitrary. Pantazidou and Sitar (1993) used the relation dn = 0.4dg, where dg (L) is the median grain size, which was derived from several literature values for similar porous media. With this relationship, predicted and observed T values were within 0.6 cm.
The lens thickness expression proposed by Pantazidou and Sitar (1993) was used by Schroth et al. (1995) to compare predicted against observed lens behavior following spills of Soltrol 220 and Duoprime 55 (a mineral oil). Schroth et al. (1995) noticed that the final position of all lenses was in the reduced capillary fringe and that the upper NAPL lens boundary coincided with the top of the original capillary fringe. This final position and upper lens geometry was different from the lens locations above the capillary fringe reported by Pantazidou and Sitar (1993). From the discussions in Pantazidou and Sitar (1993), it is not clear why their lenses did not penetrate as deeply as the ones reported by Schroth et al. (1995). Potential reasons might be that Pantazidou and Sitar (1993) applied a relatively smaller spill size and that kerosene left behind more residual saturation in the unsaturated form. When using the relation
 | [8] |
 | [9] |
 | [10] |
A more comprehensive theory, including the consideration of saturation–capillary pressure hysteresis, of both LNAPL lens and DNAPL pool behavior under equilibrium conditions was provided by Miller et al. (2004). The theory is based on hysteretic Brooks–Corey S–P relations (Brooks and Corey, 1964) with a drainage entry pressure below which the nonwetting fluid phase is not capable of displacing the wetting phase and a imbibition entry pressure, representing the capillary pressure on the imbibition curve at which the nonwetting phase becomes discontinuous (Miller et al., 2004). A series of two-phase water–NAPL and three-phase water–NAPL–air experiments were conducted in homogeneous sands to verify developed equations for the maximum pool and lens thicknesses. For instance, for a DNAPL pool in a coarse-grained porous medium, on top of a horizontal finer-grained material, the authors showed that the capillary pressure at equilibrium along the bottom of the pool is the entry capillary pressure on the water–DNAPL drainage curve, Pe,dnl, because the DNAPL initially invaded a water-saturated region at these edges. At the top of the pool, the capillary pressure is equal to the entry capillary pressure on the water–DNAPL imbibition branch, Pe,inl, and all capillary pressures at other location in the pool are between Pe,inl and Pe,dnl. Based on this, the maximum thickness of a pool can be computed as
 | [11] |
Corapcioglu et al. (1996) conducted several flow visualization experiments with the LNAPL Soltrol 130 to determine lens thickness values as a function of time to test an analytical model for lens (mound) thickness. The model neglects capillary pressure gradients and assumes sharp interface principles. Another assumption is that the initial lens is rectangular. The rather complex solution, consisting of several real and imaginary components, requires input of the average NAPL saturation within the lens and the residual saturation. The analytical model was in agreement with the experimentally observed mound thicknesses for all practical purposes (Corapcioglu et al., 1996).
Unstable Flow Experiments
Unstable flow occurring during the displacement of fluids in porous materials may give rise to the development of fingers. The instabilities are caused by differences in viscosity and density between the fluid phases (Smith and Zhang, 2001). Several flow cell experiments have been conducted where denser and less viscous NAPL was allowed to displace water in a vertical downward direction. According to the widely used linear stability theory of Chuoke et al. (1959), such a displacement is inherently unstable. Held and Illangasekare (1995a) described effects of several physical and chemical parameters on flow instability. The experiments by Fishman et al. (1998) were designed to determine a parameter called the "density of fingers" to be used to define NAPL surface area available for mass transfer. Glass et al. (2000) focused on experiments in heterogeneous systems to test a scale analysis describing finger diameter and pool formation. Flow cell experiments determining effective interfacial tensions and finger spacing during unstable displacements were reported by Smith and Zhang (2001). Zhang and Smith (2001) developed a mobile–immobile–zone model and tested it against results from similar experiments as described by Smith and Zhang (2001).
The experiments by Held and Illangasekare (1995a) involved a three-dimensional flow cell to avoid wall interference. Based on an analysis by Chuoke et al. (1959), Held and Illangasekare (1995a) determined that displacement fronts are expected to become unstable for perturbations of a wavelength (or spacing) within the order of a few centimeters. The distance between adjacent fingers is usually referred to as the critical wavelength, 
crit (L). To allow free finger development, a flow cell was constructed that was several times larger in all three dimensions. Realizing that three-dimensional flow cells complicate flow visualization and data collection, a segmented cell was constructed. The segmentation allowed an easy sectioning of the cell after an experiment to obtain images (Held and Illangasekare, 1995a). A total of 10 experiments were conducted where a 1-L spill of either the DNAPL TCE, TCA, or DBP was allowed to infiltrate and redistribute for at least 3 d in a coarse-, medium-, or fine-grained sand. Trichloroethene and TCA have a viscosity less than water, whereas DBP has a viscosity of 20 times the viscosity of water. After apparent steady-state conditions were obtained, the flow cell was dissected and photographed. Results show finger formation for TCE and TCA for all porous media and for DBP in the coarse- and medium-grained systems. Critical wavelengths ranged from 1 cm in the coarse-grained sands to about 4 cm in the fine-grained sands, which could be attributed to differences in displacement velocities. The infiltration of DBP in the fine-grained sand was stable. In general, the experiments confirmed the destabilizing effect of gravitational forces and the stabilizing effect of viscous forces when water is displaced by DNAPL in a downward direction (Held and Illangasekare, 1995a). In a companion paper, Held and Illangasekare (1995b) used fractal analysis to explain the results obtained in Held and Illangasekare (1995a).
The only unstable flow experiments in which NAPL had to move through a vadose zone first before entering a saturated region were described by Fishman et al. (1998). In the experiments, a finite volume of PCE was allowed to infiltrate under constant head conditions into a homogeneously packed, variably saturated medium-grained sand. As was observed in other experiments with DNAPL movement through an unsaturated zone (Hofstee et al., 1998a; Oostrom and Lenhard, 2003), fluid infiltration was stable without finger formation. When PCE reached the capillary fringe, both spreading and the development of small fingers along the leading edge were observed. Once formed, these fingers tended to grow due to an unfavorable viscosity ratio (Fishman et al., 1998). To quantify the density of fingering through a horizontal plane, the authors estimated the cumulative area associated with fingers, divided by the total area occupied with DNAPL. Using this definition, the density of fingering, obtained from a light reflection method, ranged from 1 at the source area to 0.4 to 0.6 at the water table.
Glass et al. (2000) performed TCE infiltration and redistribution experiments in heterogeneously packed flow cells, reminiscent of a fluvial channel architecture. Three hydrophilic silica sands were used with nonoverlapping grain-size distributions. The authors produced two packing structures: a primary heterostructure and a secondary heterostructure (Fig. 1 in Glass et al., 2000). The only difference between the packings was that for the primary heterostructure, one of the fine-grained capillary barriers in the lower half of the cell was broken. The experiments resulted in TCE distributions characterized by a number of pools on capillary barriers and fingers that connect the pools together along a pathway. In the experiment with the continuous capillary barrier (secondary heterostructure), a considerable pool was formed on this barrier. An image analysis showed nearly uniform TCE saturations in pools but more variable saturations within fingers. In general, gravity-destabilized finger growth occurred in all units on entering until a capillary barrier was encountered, followed by gravity-stabilized pool growth (Glass et al., 2000). Based on the observed fingering, the authors questioned whether two-phase, continuum-based modeling would be able to capture the unstable migration behavior in the laboratory and the field. The authors developed a simple macroscale structural growth model that assembles length scales to predict DNAPL infiltration from a source into a heterogeneous domain. Glass et al. (2000) proposed the use of their length-scale analysis as a more appropriate means to model infiltration under natural gradient conditions.
The experimental work by Smith and Zhang (2001) was conducted to support the development of an expression for the effective interfacial tension, a parameter needed to predict the wavelength of fingering, , defined as the distance between adjacent fingers. The difficulty in determining this effective value has affected the applicability of the linear stability theory developed by Chuoke et al. (1959). Several investigators have attempted to relate the effective interfacial tension to the bulk interfacial tension using a proportionality constant C. These efforts have typically yielded constant values that are only valid for the systems that were investigated. Smith and Zhang (2001) modified a fractal theory developed by Chang et al. (1994) to describe wetting front instability for water flow into two-layered soils. The modified theory resulted in an expression for C in terms of fractal properties that were obtained experimentally by Smith and Zhang (2001). The C value was used to determine effective interfacial tensions and, in turn, values, for the flow cell experiments conducted with several types of glass beads. The theoretical values were close to the experimentally obtained values, except for the smallest bead size. No satisfactory explanation could be provided by the authors for the observed discrepancies for these beads. The Smith and Zhang (2001) experiments provided detailed photographic information on fingering mechanisms like splitting (development of secondary fingers), coalescing (the process of two or more fingers combining into one), and shielding (larger fingers outgrow their smaller neighbors and inhibit their growth).
A companion paper using the same flow cell and similar experimental protocols to test the validity of a mobile–immobile zone model was presented by Zhang and Smith (2001). In the theory, a finger was conceptualized as a winding tube with a constant diameter with the same length and volume as the actual finger. Within this tube, an imagined coaxial tube of a smaller diameter exists that is considered to contain mobile DNAPL. Outside the mobile zone, the DNAPL is immobile, except for the DNAPL in the fingertip. Zhang and Smith (2001) called the mobile zone the "core" and the immobile part the "sheath." Based on this conceptual model, a gradient of hydraulic potential could be developed within a DNAPL finger and this gradient could be related to finger velocity. The authors noticed that when the DNAPL head was kept constant, the fingers in the experiments grew almost linearly in time, which is consistent with a gravity-dominated gradient, and that finger velocity decreased with increasing permeability of the porous media. The fingers typically had irregular surfaces with numerous protuberances and small terminal branches. In general, the authors concluded that the theory was applied successfully to the experiments, although it was not made clear how the results obtained for the glass beads could be applied to other, more realistic, systems.
Vapor Transport
Intermediate-scale flow cell experiments related to vapors originating from disposed volatile NAPLs are scarce. Only Johnson et al. (1992) and Johnson and Kraemer (1994) reported experiments involving vapor behavior in porous media. Johnson et al. (1992) conducted an experimental and numerical study directed at the relative importance of density-driven vapor transport of TCE in the gas phase of the unsaturated zone. Their three-dimensional flow cell was rather large (Table 1) and was filled with either medium-grained sand or pea gravel with permeability values of 5 x 10–11 and 5 x 10–9 m2, respectively. The water table was placed at about 3.3 m below ground surface, allowing a considerable unsaturated zone with variable water saturations. For each of the two TCE experiments, one liter was poured over a 0.3-m x 0.3-m excavated area over a period of 1 min. Saturated TCE vapor has a relative vapor density of 1.15 compared to ambient air at room temperature. Several sampling ports were used to obtain detailed spatial and temporal concentration distributions. Results showed very little density-driven TCE flow in the medium-grained sand as a result of the relatively low permeability and rapid concentration decreases due to gas diffusion and partitioning into the aqueous phase. The TCE experiment in pea gravel produced a strong density effect, including fast downward movement and lateral spreading on top of the capillary fringe. The primary difference in vapor transport between the two experiments was attributed to the permeability contrast (Johnson et al., 1992). A modeling exercise, using the equilibrium three-phase model by Mendoza and Frind (1990), showed a good comparison for the medium-grained sand experiment but an underestimation of density-driven transport for the pea gravel experiment. The authors reasoned that the differences were the result of experimental nonequilibrium partitioning not incorporated into the simulator and differences between actual and model parameter values. No sensitivity simulations were included to investigate variances in input parameter values.
The behavior of diesel fuel vapors emanating from injected liquid sources was investigated in flow cells filled with kiln-dried sand (Johnson and Kraemer, 1994). A total of three replicate experiments and one experiment with a slanted flow cell were conducted. Total hydrocarbon vapor contents were obtained instead of individual component concentrations. Using an analytic solution of a gaseous diffusion model analogous to the Theis (1937) equation, diffusion coefficients were obtained for various configurations of the liquid diesel full source. The diffusion coefficients were subsequently used to simulate diesel fuel vapor behavior in a hypothetical, three-dimensional diffusion model for field-scale applications.
|
NAPL Saturation Imaging Techniques |
|---|
|
TOPABSTRACTINTRODUCTIONFlow Behavior ExperimentsNAPL Saturation Imaging...NAPL Detection and...Conclusions and Research...REFERENCES |
|---|
For the photon-attenuation methods, fluid saturation values are obtained at "points," and counting times may be relatively long. To create detailed two-dimensional images of NAPL zones, total counting periods may be in the order of days (Oostrom et al., 2006b). Photographic techniques have the advantage that images of larger domains can be obtained almost instantaneously and that no radiation is involved. However, these methods are typically limited to silica sands, relatively thin flow cells, and sometimes have rather cumbersome calibration procedures.
Gamma Radiation Method
The gamma radiation technique is based on the attenuation of monochromatic gamma rays by natural sources. Most gamma radiation systems currently in use are dual-energy systems with americium-241 and cesium-137 sources. An extensive review of dual-energy gamma radiation systems including calibration procedures was provided by Stillwater and Klute (1988) and an error analysis for the various applications was developed by Oostrom et al. (1995). The equation of attenuation of gamma rays through a system with n components (i = 1, n) can be written as
 | [12] |
i the volumetric content, x the path length (L), and the subscript j denotes the source (e.g., j = a for 241Am; j = c for 137Cs). For applications in this paper, the components are solids (subscript s), water (subscript l), and NAPL (subscript n). The contribution of the gas phase to the attenuation of the gamma rays is typically neglected, allowing dual-energy systems to obtain up to three-phase saturations. For a rigid porous medium, the incident count rate in Eq. [12] can be replaced by the count rate through the column or cell when the porous medium is dry. That count rate can be measured directly if the column or cell is packed with a dry porous medium. Indirect methods for determining Ijo are necessary for columns and cells packed under saturated conditions and are described by Oostrom et al. (1998). A drawback of dry packing is the possibility of cracking, settling, or consolidation of the porous medium, which affects the bulk density values. Packing under saturated conditions usually results in relatively large bulk density values that remain stable in the course of an experiment. Unfortunately, packing under saturated conditions precludes direct determination of the count rates through the dry material. To measure water and NAPL contents or saturations with either a dual- or a single-source system, the porous medium has to be rigid. If during an experiment the dry bulk density at a measurement location changes as a result of settling or consolidation, significant errors can be expected in the obtained water and NAPL content values. Dual-energy methods have to be used in water–NAPL experiments when the porous medium contains three fluids (water, NAPL, air). A single-energy procedure can be used to determine the NAPL content in cases where the total liquid content equals the porosity:
 | [13] |
 | [14] |
 | [15] |
 | [16] |
The first flow cell experiments where gamma radiation was used to image NAPL saturations were described by Host-Madsen and Jensen (1992). They reported that with a 20 mCi cesium-137 and a 30 mCi americium-241 source, counting times of 2 to 3 min were needed for each measurement. The total measurement period was kept relatively short because a fairly coarse 10-cm x 10-cm grid was used, and measurements were only taken at locations were NAPL was present. As a result of the coarse measurement grid, the interpolated images did not show much detail. The dual-energy gamma system used by Illangasekare et al. (1995a,b) at the University of Colorado consisted of a 50mCi cesium-137 and a 200 mCi americium-241 source and was used to image both LNAPL and DNAPL movement in partially saturated silica sands. The authors primarily showed the obtained saturations at selected vertical profiles instead of images of two-dimensional areas. This system is currently located at the Colorado School of Mines. Waddill and Parker (1997), at the Virginia Institute of Technology, used a relatively coarse measurement grid with a total of just 88 measurement locations for a flow cell that was about 140 cm long and 104 cm high. Similar to Illangasekare et al. (1995a,b), the measured NAPL saturations were shown as vertical profiles and not as two-dimensional images. Waddill and Parker (1997) did not report the strength of their cesium-137 and americium-241 sources.
The dual-energy gamma system at Auburn University, with a 200 mCi cesium-137 and a 200 mCi americium-241 source, was used in Hofstee et al. (1998a,b), Jalbert et al. (2003), and Oostrom et al. (1997, 1999a,b). Another frequently used system, with a 100 mCi cesium-137 and a 200 mCi americium-241 source, is located at the Pacific Northwest National Laboratory. Nonaqueous phase liquid saturation images obtained with this system appeared in Brusseau et al. (2000), Oostrom et al. (2003, 2005, 2006b), and Oostrom and Lenhard (2003). In the Auburn University and Pacific Northwest National Laboratory contributions, the experimental conditions were such that extensive gamma scans could be obtained during periods with near steady-state conditions or static equilibrium. Nonaqueous phase liquid saturations images were typically constructed with data from more than 1000 locations. Schaerlaekens and Feyen (2004), at the University of Leuven (Belgium) used data at 3000 locations to create plots of the initial TCE saturations after injection and redistribution. A smaller grid of only 210 locations was used for the investigation of surfactant-enhanced solubilization. Schaerlaekens and Feyen (2004) did not provide a description of the used dual-energy gamma system.
X-Ray Attenuation Method
The basic attenuation principles regarding X-ray methods are similar to those for traditional gamma radiation techniques (e.g., Oostrom et al., 1995). The fluid content equations derived for this method are presented by DiCarlo et al. (1997) and are similar in form to Eq. [15] and [16] derived for dual-energy gamma measurements. Hill et al. (2002) commented that X-ray attenuation techniques have three potentially useful characteristics: (i) they provide a broad energy spectrum, allowing for the simultaneous determination of several unknowns, (ii) they may provide relatively high photon flux rates, and (iii) they are tunable, meaning that the photon spectrum can be varied depending on the measurement type. Compared to gamma radiation sources, X-ray generators generate photon fluxes that are usually orders of magnitude greater, and they produce broad polyenergetic photon spectra. Since the broader spectra carry more information, better measurement quality can be expected if used with proper data reduction procedures (Hill et al., 2002).
The two types of X-ray measurements that are used for flow cell experiments involving NAPLs are the X-ray tube and synchrotron method. The X-ray tube method typically generates dual energies from an X-ray tube source (McBride and Miller, 1994). The relatively long counting times of the X-ray tube method can be overcome by switching to synchrotron radiation sources (Liu et al., 1993). The greater than 10.000-fold intensity increase over X-ray tube sources and radioactive elements produces fast saturation measurements and allows preferential flow to be precisely monitored in real time (DiCarlo et al., 1997). The high photon flux and tunable energies available at synchrotron sources allow lowering of the measurement variances and hence a reduction of counting times to seconds compared to approximately 1 min for the more conventional gamma radiation procedures.
Imhoff et al. (1996) described flow cell experiments where the X-ray tube method was employed to determine residual TCE saturations during flushing with deionized water. The experiments were conducted to investigate the effects of Darcy flux, initial TCE saturation, and particle diameter on the formation of dissolution fingers. X-ray scans were obtained at several times during the dissolution process at 960 separate locations covering a 20-cm horizontal by 12-cm vertical area with a counting time of 5 to 6 s per measurement. Using this frequency and spatial discretization, detailed saturation plots were obtained as a function of time. Imhoff et al. (1996) estimated a standard error as a result of random and systematic errors of 0.005. The authors claimed that measurements of this accuracy had not been reported before in subsurface hydrology experiments. The same X-ray tube system was also used to document remediation of TCE using "density-motivated" mobilization (Hill et al., 2001; Miller et al., 2000). The flow cell remediation experiments were discussed in Oostrom et al. (2006a). Moreno-Barbero and Illangasekare (2006) used an X-ray system to measure PCE saturations in a 7.5-cm long and 3-cm high rectangular source zone at 250 locations. Saturation profiles representing the average saturation in the horizontal direction were presented, including expected errors of the measurements.
Flow cell experiments using synchrotron X-ray measurements were conducted by DiCarlo et al. (1997), Rimmer et al. (1998), and DiCarlo et al. (2000). The method was used for monitoring finger flow of a LNAPL in a dry and partially water saturated sandy soil by DiCarlo et al. (1997). Selection of the energy levels is especially important when two liquids are present so that one liquid preferentially attenuates relatively low-energy and the other relatively high-energy X-rays (DiCarlo et al., 1997). DiCarlo et al. (1997), using the Cornell High Energy Synchrotron Source (CHESS), were able to simultaneously measure the incident and the attenuated radiation intensity with scintillation detectors placed perpendicular to the beam. In this fashion, the detectors counted the X-ray photons that were vertically scattered from the beam by the air. To control the magnitude of the scattered flux into the detector a lead aperture and an aluminum attenuator were placed in front of each detector such that approximately 15,000 counts s–1 were obtained in each detector for both energy levels. White radiation generated by the synchrotron is converted to monochromatic X-rays by diffraction from a pair of perfect silicon crystals. The energy levels of the monochromatic beams used by DiCarlo et al. (1997) were 20 and 40 keV, respectively, with corresponding intensities of 1011 and 1010 photons s–1. Although Compton scattering by the high-level energy radiation must be accounted for in the low-energy window for the more conventional dual gamma systems, this would not account for more than 1% in the synchrotrons. In their experiment DiCarlo et al. (1997) used Soltrol 220 (Phillips Company) doped with iodoheptane and a red Sudan IV dye, and distilled water doped with 50 g L–1 NaBr and FD&C blue dye. The dopants were used to improve the contrast between the liquids' attenuation coefficients, similar to typical practices for gamma radiation methods. As the NAPL infiltrated into a partially saturated porous medium and traveled toward a water table, saturation measurements were first obtained as a function of time at a height of 45 cm and later at a height of 30 cm from the bottom of the flow container. Once steady-state conditions had been reached, a horizontal scan was obtained at the 30-cm height to capture the width and saturations of the infiltrating oil and the remaining water. The same synchrotron system was also used by Rimmer et al. (1998) and DiCarlo et al. (2000) for two-phase NAPL–water systems. In both cases the unstable flow (fingering) of water was studied when infiltrating into NAPL-saturated porous media. In DiCarlo et al. (2000), the effect of surfactant on unstable flow was included in the study.
Light Transmission Method
Flow cell experiments using the light transmission method (LTM) to determine NAPL saturations were described by Conrad et al. (2002), Darnault et al. (1998, 2001), and Glass et al. (2000). Smith and Zhang (2001) and Zhang and Smith (2001) used the LTM to obtain DNAPL finger characteristics like width and characteristic wavelength, but not for saturation determinations. Schroth et al. (1995) used the technique to obtain outlines of LNAPL lenses after infiltration and redistribution. With knowledge of the spill volume and estimates of the lens volume, they were able to compute an average saturation value for the lens. Since Schroth et al. (1995) did not use the LTM to measure NAPL saturation, their application is not discussed in this section. The flow cells in LTM experiments were placed between a controlled output diffuse light source and a digital camera that measures light intensity transmitted through the flow cell as a function of time and space (Glass et al., 2000). The LTM used for NAPL saturation determination can be viewed as extensions of the techniques described by Glass et al. (1989) and Tidwell and Glass (1994) to determine aqueous phase saturations.
Glass et al. (1989) visualized water content in an entire two-dimensional flow field by illuminating one side of a thin slab (1 cm thick) of a porous medium. They observed an increase in light intensity with increased water saturation and attributed this to the closer matching of the index of refraction of the solid matrix and water relative to the solid matrix and air. By subtracting digitized images of the dry porous medium from flow cell images of the system containing water, an image that only showed the water content was obtained. Because frames could be taken every 1/30 s, it was possible to record rapidly moving water fronts, such as those occurring during fingering. Improvements of the LTM described by Glass et al. (1989) were introduced by Tidwell and Glass (1994), mostly in the analysis technique by image adjustments and conversion of adjusted gray levels to water saturation values. Tidwell and Glass (1994) also derived equations relating aqueous phase saturation to light intensity and light transmission factors at the sand–water and sand–air interfaces. To complicate matters, the equation includes an empirically measured parameter representing the average number of pores across the sample. Schroth et al. (1998a) provide a laboratory method to estimate the porous medium dependent parameter values using light transmission data for completely dry and water-saturated conditions. Schroth et al. (1998a) were not successful in using the LTM for the determination of NAPL saturations.
The LTM of Glass et al. (1989) and Tidwell and Glass (1994), applicable to relatively thin air–water–solid systems, was expanded by Darnault et al. (1998) to simultaneously determine NAPL and water volumetric contents. Darnault et al. (1998) found that the main problem to allow for NAPL and water saturation measurements in two-fluid phase systems using the LTM was the near identical refraction indices of these fluids. The similarity of the numbers made it impossible to measure liquid contents from light intensity measurements. As a solution to this problem for two-phase systems they proposed coloring the water phase with 0.005% FD&C blue #1, resulting in a direct relation of the hue value and the water content. The NAPL content would then follow directly by subtracting the water content from the known porosity. The blue dye was used because it produces a wide color spectrum going from yellow for the oil-saturated sand via green to blue for water-saturated soil. The authors explained their choice of hue as the best variable to measure water content through a discussion of the characteristics of color vectors. The most common way of defining color is in terms of a vector with the three components of red, green, and blue (RGB). Another way to specify the color vector involves hue, saturation, and intensity. Hue is the attribute that describes the pure color and is what we typical refer to when we use the term color. Saturation is the attribute that describes the degree to which the color is diluted with white. Intensity is the attribute that corresponds to the gray level (black and white) of the color image. Interestingly, hue values can be calculated from the RGB vector values derived from color images. Darnault et al. (1998) established a linear calibration curve of water content versus hue, using RGB information, on a two-dimensional chamber consisting of fluid saturated compartments containing silica sand with known quantities of oil and water. As was pointed out by Darnault et al. (1998), a calibration curve is dependent on the porous medium and fluid properties. Although hue is calculated from the RGB vector, the R, G and B values as such were not related to the water content, nor were intensity or saturation. The LTM was evaluated with a mass water balance experiment and results from a synchrotron X-ray experiment (Rimmer et al., 1998). In one case, 20 mL of water was injected into an oil-saturated sand. Based on the LTM, they determined 19.85 mL to be present in the porous medium, which was within 1% of the amount of water applied. In the other experiment, water was injected into the tank at three different constant flow rates, resulting in fingered flow. The LTM and the synchrotron X-ray measurements by Rimmer et al. (1998) showed not only similar behavior of fingered flow but also that the water content values were very much the same. Because of the short duration to take an image with a video camera (<0.05 s), transient flow phenomena could easily be recorded and the fluid contents quantified.
In a later paper, Darnault et al. (2001) were able to determine liquid contents in a three-fluid water–NAPL–air system by using both hue of the blue-colored water (CuSO4) to determine water content and intensity to determine the total liquid content. The air content was then determined by subtraction of the water and oil contents from the porosity. To obtain a calibration curve between hue and the water content and between intensity and the total liquid content, a two-dimensional calibration chamber consisting of cells filled with a porous medium and known quantities and fluid ratios of NAPL–water, air–water, NAPL–air, or air–NAPL-water was constructed, making the calibration procedure rather cumbersome. Depending on the mixing ratios of the three fluids, different techniques had to be used to fill the individual cells with the fluids and solids. For each cell hue and intensity were measured as the mean values in a centered rectangle (16 x 0.8 cm). Different equations had to be used to relate hue to water contents above and below 0.076. Above this water content, hue and water content were uniquely determined, but below this value hue depended on the NAPL and air contents in the cell. In the latter case, an iterative process is required to find the NAPL and water contents. For the details of the complex procedure to relate intensity to the total liquid content, see Darnault et al. (2001). Results from LTM were compared with data obtained by synchrotron X-rays for both a static and a transient finger flow experiment. Comparing water and NAPL contents determined by the two methods for the static experiment yielded regression coefficients of 0.97 and 0.92 for the water and NAPL, respectively. For the transient experiment, the two methods compared favorably based on measured finger widths in the air-dried and NAPL saturated sand. Despite the extensive calibration procedure to relate color vector to fluid contents, and the limitation of using silica sand and relatively thin two-dimensional flow cells, the authors called their method an effective tool for investigating three-phase systems. They argued that the unique ability to map the complete flow cell within less than 0.1 s more than compensates for the disadvantages.
Glass et al. (2000) recorded TCE infiltration and redistribution with an LTM in a highly heterogeneous flow cell containing three different silica sands. A digital camera with 1317 x 1035 pixels with a resolution of 0.25 mm2 pixel–1 and 4096 gray levels was used. Instead of coloring the water blue, as Darnault et al. (1998, 2001) did, Glass et al. (2000) elected to add oil red O dye to the TCE with a concentration of 0.9 g L–1. To compute TCE saturations, modifications of techniques presented by Tidwell and Glass (1994) were used. Although no equations were presented, Glass et al. (2000) stated that TCE saturations were calculated using energy absorption by the dye, analogous to X-ray transmission equations shown by Tidwell and Glass (1994). The calibration procedure used for this LTM appears to be less cumbersome than the procedures described by Darnault et al. (1998) for two-phase systems. The LTM described by Glass et al. (2000) needed a scaling approach based on an expected maximum TCE saturation value to estimate TCE saturation in pools. Problems with the LTM were also apparent when the TCE saturation was not uniform from the front to the back wall of the chamber and when high TCE saturations were close to completely water-saturated regions. In the latter case, the scatter of light caused blurring of the sharp transitions. Glass et al. (2000) estimated the overall error in TCE volume for the flow cell to be approximately 12%. They argued that although additional calibration could reduce this number, some error would always remain and that for the purpose of tracking infiltration fronts, saturation errors were of secondary importance. Given the quality of the high-resolution TCE saturation plots obtained for their two TCE infiltrations experiments (Plate 1 in Glass et al., 2000), the LTM seems to have been successfully applied to a multifluid flow cell experiment. Conrad et al. (2002) attempted to use the technique described by Glass et al. (2000) to obtain TCE saturations in a heterogeneous flow cell containing silica sands during surfactant flushing and in situ oxidation remediation. For this LTM application, a camera with 1024 x 1024 pixels (0.36 mm2 pixel–1) and 4096 gray levels was used. It was reported that the application was not successful because of several confounding factors. First, the introduction of the surfactant and permanganate remedial solutions affected light transmission. For instance, the salt in the surfactant solutions caused an increase in transmission, while decreased transmission was caused by the purple potassium permanganate solution and the formation of reaction products. Second, the TCE was more rapidly solubilized than the dye. Lastly, considerable exposure problems were encountered in the finest sand and at textural interface due to the rather large disparity in light transmission. These three effects negatively influenced the ability to quantify TCE saturations, and the authors stated that the digital pictures could only be used to display qualitative spatial and temporal differences during the course of the experiments.
Light Reflection Method
Van Geel and Sykes (1994), Fishman et al. (1998), and O'Carroll et al. (2004) used a light reflection method (LRM) to obtain NAPL saturations. Van Geel and Sykes (1994) used a similar LRM as developed by Schincariol and Schwartz (1990) and Schincariol et al. (1993) to quantify salt concentrations to document an infiltrating and redistributing LNAPL dyed red with Sudan III in a well-sorted, variably saturated silica sand. They used a series of color slides (Kodak Ektachrome 64T) to record the spill and its subsequent movement. The slides were initially digitized as both RGB and black-and-white images. The RGB images did not provide any additional information beyond that contained in the black-and-white images, and hence, the black-and-white images were used for subsequent analysis analogous to Schincariol et al. (1993). An image taken before LNAPL infiltration started was subtracted from images taken during the multifluid flow experiment, creating images containing only the changes in the gray-scale value due to the presence of the LNAPL or the compression of the capillary fringe (Van Geel and Sykes, 1994). The normalized images were then calibrated to the known volume of LNAPL in the flow cell to determine the saturation distribution. Main assumptions in the analysis were that the LNAPL was distributed uniformly across the depth of the box and that the image of the front side properly represented the internal saturations away from the wall. Van Geel and Sykes (1994) pointed out that the container's glass plates avoided preferential migration by the LNAPL and that the LNAPL front moved uniformly across the thickness of the container. Evidence of the latter was provided by pressure transducer data. The authors also noted that differences in porosity near the glass plates and in the interior of the tank were accounted for in their calibration procedure. Using the assumption that the differential gray scale is a linear function of LNAPL saturation, a linear regression resulted in a relationship whereby LNAPL saturations could be simply computed as the difference in gray scale multiplied by 1.55. A drawback of this LRM was that LNAPL saturation changes could only be quantified when the water saturation remained constant over time. With this restriction, fluid displacement in the capillary fringe could not be quantified. The authors stated that this shortcoming was being addressed, but to our knowledge an improved LRM has not been published.
Fishman et al. (1998) used a similar LRM technique as Van Geel and Sykes (1994) to estimate dyed PCE saturations from video camera images. They assumed that the red organic Sudan IV dye would act as a filter and that its optical density would then be proportional to the degree of saturation. Fishman et al. (1998) produced a calibration curve plotting PCE saturation versus observed gray scale. Instead of using independent calibration data, this curve was made using data from the actual infiltration experiment. The authors did not report estimation difficulties in the capillary fringe or below, although it is questionable whether the obtained calibration curve is applicable to both two-phase water–NAPL and three-phase water–NAPL–air systems. O'Carroll et al. (2004) estimated saturations in a flow cell by correlating the hue of light reflected from the dyed PCE to known saturations in small glass cells. The assumption was made that the observed behavior near the wall was representative of the entire cell thickness. Since PCE clearly showed unstable fingering behavior, this assumption was not correct at several locations. Due to lighting issues and differences in glass thickness between the flow cell and the calibration cells, the sum of PCE hue in the flow cell had to be normalized to the known total volume. The scaling procedure resulted in considerable errors, causing the authors to only show one measured NAPL saturation plot.
Multispectral Image Analysis Method
Kechavarzi et al. (2000) presented a multispectral image analysis method (MIAM) to determine dynamic fluid saturation distributions of water, NAPL, and air in two-dimensional intermediate-scale laboratory experiments. As with the LTM and LRM, this technique provides a nondestructive and nonintrusive tool for multifluid flow imaging for systems with rapid changes (Kechavarzi et al., 2000). Within three 10-nm spectral bands (centered at 500, 760, and 970 nm) of a visible and near infrared spectrum, the authors developed relationships between the optical density of the reflectance and the NAPL plus water saturation of sand samples, containing two or three fluid phases. A digital near infrared camera, fitted with interference filters, was used to record the reflected light intensity within the three narrow spectral bands. Rather than defining the optical density in terms of transmittance, Kechavarzi et al. (2000) defined the optical density, D, in terms of the reflectance:
 | [17] |
 | [18] |
 | [19] |
 | [20] |
and ß are regression coefficients. Kechavarzi et al. (2000) pointed out that depends on porosity and that, consequently, the linear relationships only develop if the porosity is constant. They also showed experimentally that values for ß, representing the reflectance of dry porous medium, were hardly effected by variations in porosity.
For three-fluid phase systems, Kechavarzi et al. (2000) showed that for each spectral band, the average optical density was a linear function of LNAPL saturation for fixed water saturations. Based on these results, the optical density in three-fluid phase systems can be written as
 | [21] |
 | [22] |
coefficients need to be large as well as and should be as different as possible for the two chosen spectral bands. Based on data in Kechavarzi et al. (2000), it is obvious that either the combination i = 500 and k = 760 nm or i = 500 and k = 970 nm should be selected. To solve Eq. [22], all and ß values can be obtained from two-fluid phase systems, and no three-fluid phase systems are needed. The authors stressed that during the image processing, it is important to overcome any uncertainty in uneven lighting across the experimental domain.
The MIAM was applied to three infiltration and redistribution experiments with Soltrol 220 in partly saturated porous media. Fluid flow behavior in one of the experiments was described in detail by Kechavarzi et al. (2005). For each of the porous media, extensive calibration procedures yielded the coefficient values needed to compute fluid saturations according to Eq. [22]. During the experimental phase of their study (Kechavarzi et al., 2000), either 3 or 4 L of LNAPL was introduced in a partially saturated sand, homogeneous or layered, with the water table near the bottom and a capillary fringe height ranging from 20 to about 50 cm. Differences between applied and measured saturations varied from 2.8 to 10.5%, depending on the experiment. The authors suggested a number of possible explanations for the observed discrepancies, including flow cell bulging, variability in porosity, and wall effects. It was shown that the best results were obtained for the experiment with the largest contrast in values, where the aqueous phase was dyed blue and the LNAPL red. The question of having different fluid saturations near the wall compared with the interior of the porous medium, due to wettability and porosity effects, could not be answered by the authors.
Other Imaging Methods
Two geophysical methods have recently been applied to generate NAPL saturation images for laboratory experiments: ground-penetration radar (GPR) and time-lapse three-dimensional seismic imaging. Johnson and Poeter (2005) used GPR to detect and quantify the DNAPL hydrofluoroether (HFE-7100), dyed with oil red O. The DNAPL was chosen because of its high permittivity value (6.85) versus the permittivity of typical DNAPL used in laboratory studies. The flow cell had unusual dimensions to allow for the construction of angled clay walls and an evaluation of the effects of two- and three-dimensional structures on the GPR response that were obtained before and after the DNAPL was emplaced. The authors were able to obtain detailed images of the bulk DNAPL distribution both in time and space. The images could then be used to quantitatively determine the DNAPL volume at certain depths using GPR modeling. It was argued that the method can potentially be used in the field for relatively shallow depths, provided that geological knowledge of the site is available and that the site materials have low electrical conductivities. If insufficient geological information is available, the authors claim that the method can still be used as a detection tool. Another emerging method to determine NAPL saturations was demonstrated by McKenna et al. (2001). They showed that time-lapse three-dimensional (TL3-D) seismic imaging could be used for kerosene saturation estimates in saturated systems with sloped layers. The method could be applied because a sufficient acoustic impedance contrast existed between kerosene and water so that seismic reflections could be obtained from oil–water contacts. The method was used to measure saturations and thickness of the kerosene contamination. Unfortunately, the authors did not compare the integrated measured volumes and the actual injected kerosene volumes.
|
NAPL Detection and Quantification with Tracers |
|---|
|
TOPABSTRACTINTRODUCTIONFlow Behavior ExperimentsNAPL Saturation Imaging...NAPL Detection and...Conclusions and Research...REFERENCES |
|---|
 | [23] |
 | [24] |
 | [25] |
The partitioning tracer method has been applied successfully to quantify NAPL volumes in columns with uniformly entrapped NAPL (e.g., Jin et al., 1995). The promising results from column studies have led to field applications of the technique, primarily to assess remediation performance (Annable et al., 1998; Meinardus et al., 2002). The use of this technique in the field was supported by an error analysis (Dwarakanath et al., 1999) pointing toward relatively small random errors when tracers with a retardation coefficient of 1.2 are used. In a review paper of tracer techniques for NAPL source zone characterization, Rao et al. (2000) noted that field applications will typically result in an underestimation of the actual NAPL volume due to limitations in hydrodynamic NAPL accessibility by tracers, subsurface heterogeneity and variable NAPL distribution, and nonequilibrium mass transfer. The combined effects of these two factors were referred to as "mass transfer limitations" by Imhoff and Pirestani (2004). Nelson et al. (1999) added tracer mass loss and sampling method as factors that can constrain the performance of partitioning tracer tests.
Intermediate-scale flow cell experiments testing the partitioning tracer method were conducted by Davis et al. (2002), Istok et al. (2002), Jalbert et al. (2003), Jawitz et al. (2003), Moreno-Barbero and Illangasakere (2006), and Nelson et al. (1999). The contributions by Jalbert et al. (2003), Moreno-Barbero and Illangasakere (2006), and Nelson et al. (1999) describe experiments designed to test one or more performance-constraining aspects of the technique. Jawitz et al. (2003) used higher-order temporal moments to analyze several experiments with various premixed amounts or the insoluble n-decane. The experiments by Davis et al. (2002) and Istok et al. (2002) were conducted to evaluate the push–pull variant of the tracer test technique, applying only one well to inject and extract. Barth et al. (2003) studied the effect of macro-entrapped NAPL on the behavior of conservative tracers to evaluate the applicability of such tracers to detect high-saturation, immobile NAPL.
Nelson et al. (1999) described a flow cell experiment examining the effects of porous-media heterogeneity, sampling method, and DNAPL distribution on the performance of the partitioning tracer method. The flow cell contained two rectangular zones with approximately 10% entrapped TCE. The two zones contained a medium-grained (Accusand 20/30 mesh) and a fine-grained (Unimin 70 mesh) sand, respectively, and were located in an otherwise homogeneous Accusand 20/30 mesh sand without NAPL. Aqueous samples were collected from point samples, representing depth-specific sampling, and at the extraction well. Calcium bromide and 2,4-dimethyl-3-pentanol were used as the nonpartitioning and partitioning tracers. The results from the experiment show that preferential flow, caused by differences in intrinsic and phase relative permeability, and physical nonequilibrium had a large effect on the tracer test performance. In addition, the authors found that point sampling yielded considerably higher saturation than vertically integrated sampling ports and put that finding into the context of the inherent problems associated with integrated sampling to describe heterogeneous porous media. In this experiment, a large fraction of the flow received by the integrated samples was water that had not been in contact with the DNAPL phase. On the basis of this experiment, the authors emphasized the usefulness of the partitioning tracer technique as a detection or performance indication method but questioned the appropriateness for volume quantification.
The only flow cell experiment evaluating the partitioning tracer test following a spill was described by Jalbert et al. (2003). The main objective of this study was to investigate the influence of nonequilibrium and NAPL nonuniformity on tracer test performance. In the flow cell, 1 L of PCE was allowed to infiltrate into a porous medium with three rectangular lower-permeability zones. The heterogeneous system resulted in a final PCE distribution with residual and pooled TCE. A dual-energy gamma radiation system was used to document the PCE saturation distribution before injecting the tracer cocktail, which included one conservative tracer (2-propanol) and two nonconservative tracers (2,3-dimethyl-2-butanol and 1-hexanol), into the injection well. Tracer concentrations were obtained at 14 sampling ports and at the extraction well. Average saturation estimates were obtained with a standard moment analysis and an inverse analysis using a one-dimensional, nonequilibrium convection–dispersion equation. In an analogy with two-site sorption models, the equation used differentiated between a NAPL fraction F, where equilibrium conditions exist, and a fraction 1-F of rate-limited sites. NAPL saturations obtained with both the direct and inverse analyses for each of the 14 ports were compared with cumulative PCE saturation estimates obtained by averaging the measured values along the horizontal line upstream from each sampling port. Results showed that both analysis methods produced similar saturation estimates at most ports but that both methods generally result in an underestimation of the amounts. The authors showed that the inverse procedure yielded a great deal of variability in the fraction F and found that their findings were difficult to interpret, since even values obtained at the same port were often quite different. These difficulties were not unexpected because observed long-lasting decreases in concentration (tailing) in breakthrough curves are typically the result not only of nonequilibrium partitioning but also of nonuniform NAPL distribution, permeability contrasts, and dilution effects (e.g., Nelson et al., 1999). The breakthrough data at the extraction well were analyzed with the moment method only, yielding a total volume of 380 and 499 mL PCE using the 2,3-dimethyl-2-butanol and 1-hexanol data, respectively. Given the injected volume of 1 L, it is clear that both tracers severely underestimated the PCE volume contained in the flow cell. The underestimation was consistent with the results for the sampling ports. The authors demonstrated that the tracers likely missed the pooled PCE located at the bottom of the flow cell, which accounted for 30% of the total volume. Despite their inconclusive theoretical analysis, the authors provided evidence based on a Damkohler number analysis that tracer partitioning in the flow cell likely occurred under nonequilibrium conditions.
Partitioning tracer tests to quantify various PCE pool configurations in a rectangular source zone were reported by Moreno-Barbero and Illangasakere (2006). A pool was conceptualized as consisting of a zone of high saturation without a sharp upper boundary but, instead, with a transition zone where the saturation changes from high to entrapped values. The authors hypothesized that high saturations in the lower portions of the pool would result in bypassing of the tracers in combination with rate-limited partitioning, leading to underestimation of the PCE mass. The source zone, consisting of #16-mesh sand with injected PCE, was centrally located in a fine-grained #70-sand. A total of four different PCE saturation profiles were generated by extracting different volumes of liquid PCE after injection. Estimates of the PCE in each source zone were obtained with an X-ray scanner. Bromide was used as the conservative tracer, while 2,2-dimethyl-3-pentanol, n-hexanol, and 6-methyl-2-heptanol were selected as the partitioning tracers. Results clearly indicate that for the three sample ports located directly behind the PCE pools, PCE volume estimates were much lower than the actual emplaced quantities. The volume estimates improved with an increase in the length of the transition zone length in the PCE pool. This result was not unexpected since the transition zone has lower saturations with a higher aqueous phase relative permeability, allowing more tracers to flow through this region, resulting in reduced estimation errors. Numerical modeling of tracer behavior using equilibrium partitioning coefficients also resulted in a severe underestimation of the average PCE saturations in most pools. The authors tried to improve the results by implementing an effective partitioning coefficient, representing nonequilibrium behavior. The effective coefficients, obtained in specially designed batch studies and through inverse modeling exercises, improved PCE volume estimates. Unfortunately, this approach yielded effective coefficients that were not unique, suggesting that a single coefficient might not be applicable to an entire source zone. Moreno-Barbero and Illangasakere (2006) commented that their research demonstrates the general difficulty of the traditional partitioning application to determine DNAPL volume and argued to include the use of a series of laboratory-determined effective partitioning coefficients in the analysis of tracer tests at heterogeneous sites.
In the experiments conducted by Jawitz et al. (2003), various combinations of clean 20/30 Ottawa sand was packed on top of a sand layer mixed with n-decane. The applied packing procedure resulted in systems with various values of the contaminated fraction f, defined as the volume fraction of the porous medium containing NAPL. Partitioning tracer tests were conducted for each fraction with methanol as the nonpartitioning tracer and 2,2-dimethyl-3-pentanol and 6-methyl-2-heptanol as the partitioning tracers. The experiments were conducted with various pore water velocities and tracer concentrations to investigate the effects of nonequilibrium and nonlinear partitioning. Compared with the other experimental papers discussed in this section, Jawitz et al. (2003) extended the data analysis beyond the typical use of first temporal moments to second and third temporal moments to estimate additional statistical parameters characterizing the n-decane distribution. The analysis included the nontrivial development of homogeneous and disturbed binary models for NAPL distributions. Although the extended analysis resulted in good agreement between estimated and measured n-decane saturations, the authors emphasized that quality breakthrough-concentration data sets are a requirement for estimating higher moments.
Istok et al. (2002) conducted three experiments in a single sediment pack. They used a wedge-shaped flow cell, approximating a radial flow field near an injection–extraction well for a push–pull test application of the partitioning tracer test. The wedge was packed with natural aquifer sediment. The conservative tracer was 1-pentanol, and a mixture of three nonconservative tracers was used: 1-hexanol, 1-heptanol, and 2-ethyl-1-hexanol. The first experiment was conducted to obtain tracer retardation coefficients in a packing without TCE. In the other two experiments, 208 mL TCE was injected at four depths using 52 injection ports after the sediment was drained, in an attempt to obtain a uniform distribution. After the injection, the sediment was resaturated and flushed for approximately 24 h to fully entrap the TCE. The injected volume is equal to approximately 2% of the total pore space within the treated volume of the sediment pack. The tracer injection and extraction rate was four times as high for the second than for the third experiment. Results showed that the push–pull method could detect and quantify sorption in the absence of TCE and detect NAPL when present. The TCE saturation based on observed retardation coefficients was generally underestimated, with the exception of the 4.9% estimate using the 1-hexanol data for the third experiment. After excavation, pool formation was observed at the bottom of the flow cell, and the authors inferred from this information that underestimation of TCE saturation was partly caused by saturation heterogeneity and mass transfer limitations. On the basis of the results from the three experiments, the authors concluded that the single-well, push–pull partitioning tracer technique could be applied to detect NAPL. However, like Nelson et al. (1999), they were not comfortable endorsing this technique to quantify NAPL saturation.
Push–pull experiments to quantify NAPL using naturally occurring radon-222 as the nonconservative in situ tracer were reported by Davis et al. (2002). They used the same porous media and NAPL as Istok et al. (2002) and employed similar packing procedures. In the approach reported by Davis et al. (2002), radon-free water containing the conservative tracer bromide was first injected into a well, followed by extraction of the test solution–groundwater mixture. Before the push–pull experiments were conducted, a 3-wk rest period was imposed on the system to allow radon-222 to reach >95% of its equilibrium value. Data analysis occurred through a standard moment analysis and through approximate analytical solutions of the injection and extraction phases of the experiments. Both methods predicted average NAPL saturations smaller than 1%, considerably less than the 2% emplaced in the flow cells. Consistent with Istok et al. (2002), the authors contributed the underestimation to nonequilibrium radon partitioning and the heterogeneous distribution of TCE in the sediment. Observed pooling at the bottom of the flow cell resulted in lower retardation factors due to the reduced interfacial area between the TCE and the injected solution. Modeling using the STOMP code (White and Oostrom, 2006) resulted in average TCE saturations of up to 7.4%, which is larger than the volume-averaged TCE saturation of 2%. According to the authors, the reasons for the larger estimated retardation coefficients obtained with the simulator were unclear. A subsequent modeling paper by Davis et al. (2005) presented a reanalysis of the Davis et al. (2002) result, using the same simulator. The authors demonstrated that by adjusting the initial radon concentrations, which vary as a function of the TCE saturation, and the heterogeneity in TCE saturation distribution, the computed average saturation value could be reduced to 1.8%, which is much closer to the actual value of 2%.
Barth et al. (2003) used a 10-m-long flow cell to test the applicability of conservative tracers to detect macro-entrapped NAPL in heterogeneous systems. This contribution is the only paper in this section that did not use partitioning tracers. The approach by Barth et al. (2003) included a comparison of tracer behavior in a clean versus a contaminated porous medium. The flow cell was packed under saturated conditions with rectangular zones of five different sands to create a heterogeneous zone serving as a simple analog of a random field-site sedimentary structure (Barth et al., 2003). Before NAPL was injected into the flow cell, several tracer tests with bromide and tritium were conducted to delineate the flow field and to obtain optimized permeability values of the sands using an inverse approach. Two Soltrol 220 (Phillips Petroleum) spills, equivalent to 7 and 4% of the pore volume, respectively, were introduced from the bottom of the flow cell. The design of the LNAPL injections allowed capillary and buoyancy forces to redistribute the Soltrol 220 similar to DNAPL sinking below a water table. Conservative tracer tests were conducted after both injections. A dual-energy gamma system was used to image the Soltrol 220 saturation distribution. Color digital pictures and the gamma data indicate that the LNAPL only reached the coarse lenses in the lower half of the flow cell after the first spill. The second spill migrated throughout the vertical extent of the flow cell. Additional aspects of the Soltrol 220 flow behavior in this experiment are discussed in the "Flow Behavior Experiments" section above. Results indicated that the equivalent permeability after the first and second spill was reduced by 20 and 32%, respectively. However, a moment analysis of extraction well breakthrough data showed that after the first spill, the first moment had not changed substantially, while the second spill had a much more dramatic effect (Barth et al., 2003). The authors encouraged evaluation of conservative-tracer sensitivity to macro-entrapped NAPL for field site applications. The protocol of such studies would require tracer tests at pristine sites, followed by additional tests after NAPL injection and redistribution. It is not clear how this method can be used at a site already contaminated with NAPL.
|
Conclusions and Research Recommendations |
|---|
|
TOPABSTRACTINTRODUCTIONFlow Behavior ExperimentsNAPL Saturation Imaging...NAPL Detection and...Conclusions and Research...REFERENCES |
|---|
Most of the experiments, with the exception of the experiments with gasohols by McDowell and Powers (2003), used laboratory-grade NAPLs. In addition, most of these NAPLs consisted of a single component. However, most NAPLs encountered in the field are fluids with several components that may have undergone various degrees of weathering. As a result of changing composition, NAPL flow behavior may change over time. It is recommended that more multicomponent synthetic NAPLs, or even field NAPLs, be used in flow cell experiments to improve the understanding of flow behavior at actual field sites.
An improved understanding of the behavior of nonspreading DNAPLs in unsaturated zones was obtained through the experiments with PCE by Hofstee et al. (1998b), and CT by Oostrom and Lenhard (2003) and Oostrom et al. (2003). These experiments provided clear evidence of enhanced residual saturation formation in the vadose zone, and detailed NAPL saturation data were developed. In these experiments, only minor fingering behavior was observed in the unsaturated zone. The flow behavior seen in these experiments needs to be simulated using recently developed residual NAPL saturation formation theories (Lenhard et al., 2004) that have been implemented into multifluid flow simulators (White et al., 2004). The experiments conducted by Simmons et al. (1992), with a strongly nonspreading mineral oil, showed preferential downward flow patterns through fingering. A continuum-based model clearly should not be used to simulate the observed behavior. It is of interest to determine through experimentation and theoretical development why and how the behavior of certain nonspreading NAPLs (e.g., mineral oil) in the unsaturated zone differs from others (e.g., PCE, CT).
For most other experiments in which fingering was observed, instabilities occurred in the saturated zone during infiltration of DNAPL with a lesser viscosity than water (e.g., Glass et al., 2000; Zhang and Smith, 2001). For several of these experiments, researchers also noted that continuum-based modeling could not be used to predict finger formation and propagation. To predict average behavior of such unstable processes, upscaling procedures are necessary. For instance, the experimental visual results of unstable DNAPL behavior in heterogeneous porous media, obtained by Allan et al. (1998), were used by Braun et al. (2005) to evaluate upscaling methods. Although the results of these particular exercises were not impressive, upscaling procedures are needed to capture unstable finger behavior in continuum-based models. Compared to the experiments by Allan et al. (1998), more detailed and quantitative flow behavior experiments are needed to provide the basis for the testing and verification of upscaling methods.
For most experimental and numerical studies, the assumption was made that the porous media are water wet. However, several experiments (e.g., O'Carroll et al., 2004) have demonstrated that porous media wettability may be temporally and spatially variable. The potential effects of oil-wet porous media on NAPL movement have been demonstrated by O'Carroll et al. (2004). Additional quantitative flow and transport experiments are needed in which porous medium wettabilities range from water- to oil-wet. The obtained data need to be used to test and verify theoretical models, such as the k-S-P model outlined by Bradford et al. (1998).
Detailed data acquisition is needed to resolve differences in theories developed to explain the behavior of NAPL pools and lenses. For instance, the theories by Pantazidou and Sitar (1993) and Chevalier (1998) do not consider hysteretic effects in the formation of pools and lenses. However, according to Miller et al. (2004), the formation can only be explained when pore-geometry hysteresis is considered, especially drainage and imbibition NAPL pressures. Although the Miller et al. (2004) theory seems to be supported by the experiments, fluid pressure and saturation data are lacking to confirm the theory. Experiments are recommended for which both saturation and fluid pressures are determined to obtain a good spatial and temporal data resolution for testing and verification of the various pool formation theories. The experiments should be designed to allow consideration of residual and entrapped NAPL formation effects on the pool geometry. In addition, quantitative pool geometry experiments should be conducted at sloped interfaces to test expressions for pool movement on a sloping interface (e.g., Miller et al., 2004). Most of these expressions do not include the formation of entrapped NAPL that is left behind when, for instance, a DNAPL pool moves downward on top of a sloping textural interface.
Vapor transport experiments have been rather scarce despite the growing evidence that contaminant transport in the gas phase may be substantial. Johnson et al. (1992) and Lenhard et al. (1995) showed that dense vapors emanating from TCE may be able to migrate rapidly in a downward direction and spread laterally on top of the water table. Although the importance of gas diffusion and density-driven advection has been demonstrated in flow cell experiments, a detailed knowledge of the volatilization process is still lacking. As a result, most simulators use equilibrium-based volatilization to describe mass-transfer from the NAPL to the gas phase. This assumption leads to rapid volatilization and overprediction of removal in a soil vapor extraction system. Column and flow cell experiments are needed to better describe the conditions under which equilibrium volatilization is acceptable and when kinetic mass transfer is needed. For the latter case, relationships are needed that not only can be used to describe the results of the conducted experiments but that also have predictive capabilities.
NAPL Saturation Imaging
For most experiments reviewed here, photon-attenuation methods such as gamma and X-ray techniques were used. Although these methods are relatively slow, we expect that they will continue to be used because of their accuracy and precision. Photographic methods such as the LRM, LTM, and MIAM have the main advantage that images of flow cell domains can be obtained instantaneously. Unfortunately, the calibration methods of the published experiments are rather complicated, and there are limitations to what kind of porous media can be used. Research toward simplifying calibration procedures is needed to establish these photographic methods as reliable saturation-estimation tools. Geophysical methods such as ground-penetrating radar (Johnson and Poeter, 2005), and three-dimensional seismic imaging (McKenna et al., 2001), have shown promise for NAPL experiments in flow cells but should be developed further for laboratory applications.
NAPL Detection and Quantification with Tracers
The flow cell experiments conducted by Istok et al. (2002), Jalbert et al. (2003), Moreno-Barbero and Illangasakere (2006), and Nelson et al. (1999) all support one or more reasons why, according to Rao et al. (2000), actual NAPL volumes are typically underestimated: limitations in NAPL accessibility, subsurface heterogeneity, variable NAPL distribution, and nonequilibrium mass transfer. The experiments by Jawitz et al. (2003) demonstrated that a more extensive analysis involving higher-order temporal moments and binary distribution models may be needed to obtain good agreement between measured and estimated NAPL volumes and the fraction of tracer-swept volume containing NAPL. However, Jawitz et al. (2003) applied their methodology to relatively simple layered systems without the consideration of pooled NAPL.
In all the experiments reported by the various authors, partitioning tracers were used with a retardation coefficient of more than 1.2, a requirement recommended by Dwarakanath et al. (1999) to reduce estimation errors. The lack of success in estimating NAPL volumes in intermediate-scale laboratory experiments is disturbing and is in contrast with apparent results of field experiments. For instance, both Annable et al. (1998) and Meinardus et al. (2002) expressed confidence in their estimates of NAPL saturations and total volumes obtained with tracer tests. The discrepancy between the results in the laboratory and in the field needs to be resolved. As long as the limitations of the tracer method are not addressed in well-designed flow cell experiments, the scientific community will have reservations about using this technique for NAPL volume estimation. Experiments are needed, preferably designed by a team with laboratory and field experience, that address the limitations of the method in a systematic manner. These experiments should include single and multicomponent (field) NAPL and should also address NAPL distributions in heterogeneous porous media developing from spills.
|
ACKNOWLEDGMENTS |
|---|
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONFlow Behavior ExperimentsNAPL Saturation Imaging...NAPL Detection and...Conclusions and Research...REFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  K. G. Mumford, J. E. Smith, and S. E. Dickson New Observations of Gas-Phase Expansion above a Dense Nonaqueous Phase Liquid Pool Vadose Zone J., May 21, 2009; 8(2): 404 - 410. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. R. Goswami, B. Ambale, and T. P. Clement Estimating Errors in Concentration Measurements Obtained from Image Analysis Vadose Zone J., February 10, 2009; 8(1): 108 - 118. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| The SCI Journals | Agronomy Journal | Crop Science | |||
| Journal of Natural Resources and Life Sciences Education |
Soil Science Society of America Journal | ||||
| Journal of Plant Registrations | Journal of Environmental Quality |
The Plant Genome | |||
