In Situ Enhanced Source Removal

4.0    Performance Evaluation

4.1.0    Test Cell Design

Each of the tests was performed in a hydraulically isolated test cell.  The nominal dimensions of the test cell were 3m by 5m.  The as built dimensions are presented when a specific technology is discussed.  The test cells were constructed using Waterloo Barrier® sheet piles or sheet piles that were modified with a special additional cavity at the joint to allow grouting of the joint making a water tight seal.  The sheet piles were driven through the contamination zone approximately 2m into the clay. Prior to placing the cells in the field, preliminary core samples were collected for chemical analysis and to verify the depth to clay.  This also provided the information used to construct the multilevel samplers such that the sampling points were distributed across the expected zone of contamination.  Each of the test cells was instrumented with a common set of injection and extraction wells and multilevel samplers.  Figure 4.1.0.1 shows a cartoon of the plan view of a cell and Figure 4.1.0.2 shows a cartoon of the cross section of a cell.  The well screens Figure 4.1.0.3 were 2" stainless steel wire wrapped with a 0.01" slot.  The well casing was schedule 80 PVC except for the steam cell which had stainless steel casing in addition to the stainless steel screen.  The wells in the test cells were screened across the contamination zone to permit adjusting the water table within the test cells and delivering the remedial fluid to the point of contamination.  Casing in contact with the contaminant and water was stainless steel.  Outside of the cell a minimum of two monitoring wells were installed approximately 2m down hydraulic gradient from the test cell to monitor for potential leakage from the cells.  These wells had a 1.2m screen.  The bottom of one well was placed at the top of the clay sand interface and the top of the other well was placed approximately 0.3m below the top of the clay sand interface.  The annulus of each of the wells was filled with 20-40 sand throughout the length of the screen and a 60 cm bentonite plug was placed above the sand pack.

.
Figure 4.1.0.1  Plan view cartoon of the cell installation
 
 


Figure 4.1.0.2. Sectional view cartoon of a typical cell installation
 


Figure 4.1.0.3. Construction details for well installation
 


Figure 4.1.0.4 Construction details for multilevel sampler installation

Prior to use, each of the test cells was tested for leakage by raising the hydraulic head within the cell and watching the rate of decline.  The criteria for satisfactory fluid containment was that not more than 0.01 pore volume would be lost per day at the hydraulic head the tests were to be conducted.  Not all of the test cells initially met the leakage requirements.  In some cases, it was possible to rearrange the tests such that the desired head in the test cell did not need to be raised, thereby reducing the driving head for fluid loss and meeting the leakage criteria.  In other cases, the points of leakage were determined by injecting a tracer in the test cell and monitoring its appearance outside the test cell with ten monitoring wells placed around the test cell's perimeter.  A bentonite clay curtain was injected around the portion of the  test cell where the leaks were observed.  Some of the clay curtains effectively stopped the leakage while others only reduced the leakage rates.  When the target release rate could not be met by the combined sheet pile clay curtain system, hydraulic controls were initiated within the test cell to insure that there was no loss of remedial fluid, other than water, to the surrounding environment.  During the experimentation, monitoring wells outside the test cells were routinely monitored by an independent laboratory for the compounds injected for remediation and tracers used to characterize the contaminant distribution to insure that no injected fluids were lost to the surrounding environment.  The external laboratory never reported evidence of chemical leakage other than the tracer used to detect the leak locations.

4.1.1 Evaluation Methods

4.1.1.1 Core analysis


During the installation of the wells and multilevel samplers shown in Figure 4.1.0.4, cores were collected to determine the NAPL constituents distribution.  Typically, cores were collected for each well and along the center  row of multilevel samplers. Post remediation cores were collected along two transects, typically between the rows of multilevel samplers.   Core samples were extracted in the field and shipped to the laboratory for chemical analysis (see SOPs or detailed procedures). The chemicals targeted for analysis are listed in Table 4.2.1.1 along with selected physical properties of the chemicals.

Table 4.2.1.1 Target chemicals for chemical analysis and selected  physical properties of the chemicals
Molecular Weight Boiling point (oC) Vapor Pressure (mm hg) Solubility (mg/l) log Kow
decane
142.2
173
2.7
0.009
5.01
1,2-dichlorobenzene
147.0
179
1
100
3.38
ethylbenzene
106.2
136.2
7
152
3.16
naphthalene
128.2
217.9
?1
25
3.3
1,1,1-trichloroethane
133.4
76
100
4400
2.77
1,2,4-trimethylbenzene
120.2
169
  
57
3.65
toluene
92.1
110.8
22
515
2.69
undecane
156.31
195.5
1
  
5.74
o-xylene
106.2
144
5
175
2.77
m-xylene
106.2
139
6
35
3.2

Core samples were collected in the field when there was visual evidence of contamination or contamination was detected using a portable PID instrument.  Recovery of core material was poor due to large cobbles in the formation.  This resulted in irregular samples spacing and nonuniform samples distribution.  Once analyses were performed, a target zone for remediation was confirmed.  To calculate the mass of contaminant present,  it is necessary to know the area of the cell, and the bulk density of the formation.  The actual shape of the test cells is quite irregular.  The individual corners of each of the sheet piles were surveyed.  A 50 by 50 grid was placed over the extents of the cell and the area was calculated based on the number of nodes that were within the boundary.  The bulk density of the formation was based on the porosity of the formation (method described later) and assuming a particle density of 2.65 g/cc.  The target zone for remediation was determined by the principal investigator applying a given technology based on the contaminant distribution and available resources to purchase remedial fluids.  Four different techniques were utilized to analyze the core data to estimate the mass of each tracked contaminant, as follows:

1.    Core samples for a given cell that were collected within the target zone for remediation were averaged without regard to location.  This average value in mass per unit weight of core material was multiplied by the volume of the target zone and the average soil bulk density to determine the amount of the contaminant within the test cell. The mean and standard deviation of the mass concentrations were calculated.

2.    Since deposits are generally laid down in layers, it was decided to use a boxcar averaging technique to evaluate the mass by layers.  The target zone was divided into five layers, one for each of the multilevel samplers.  The mean and standard deviation were then calculated for each layer.  This removes some of the biasing due to a non-uniform sample distribution, considering each layer reasonably represents a similar environment.

3.    Inverse distance squared approach. -   Even with the data density collected in this study, this data set would be considered sparse.  The data set has more data vertically than it does horizontally.  To minimize the anisotropic effects of the data,  the data was transformed prior to analysis by multiplying the vertical dimension by five then performing an inverse distance squared analysis, retransforming after analysis prior to calculating the mass present. The scaling (weighting) factor selected was based on the relative data density, vertically compared to the data density horizontally, to equalize the relative weighting.  To use the inverse distance squared approach, the data was grided in a 50 x 50 x ( (maximum elevation of the target zone (ft) -  minimum elevation of the target zone (ft))*5+1) elements. The concentration (weight basis) for each element within the target zone of the three-dimensional grid was calculated. Assuming a bulk density of the soil as 1.5 g/cc the total mass was calculated.

4.    Three-dimensional kriging is a technique used in the mining industry to estimate the volume of an oer body.  This was also used as the most technically correct analysis approach.

Performance can then be determined by comparing the pre-remediation cores with the post-remediation cores.  The different techniques give different estimates of the mass of a contaminant within a cell.  The variability in the estimation techniques is generally within a factor of two.
 

4.2.1.1 Three Dimensional Kriging Description

Semivariogram Estimation

The semivariogram estimator used is:

4.2.1.1.1
where u = the random variable (contaminant concentration) being estimated

x = spatial position of known contaminant values

h = separation vector (lag) between occurrences of known contaminant values

n = number of pairs of occurrences separated by h.

A directional semivariogram analysis revealed little evidence of significant differences between the horizontal and vertical directions when the horizontal direction was defined by samples separated by not more than 10? off horizontal. Therefore, the omnidirectional semivariogram is used. The sine-hole effect model used here is developed after Jian et.al.(1996):

4.2.1.1.2
The model is fitted using the Marquardt-Levenberg weighted least squares method (SAS Institute 1992) and the weights associated with each experimental semivariogram point are:

4.2.1.1.3

 

Kriging

The following assumptions are used in the application of ordinary kriging: Over the extent of the test cell, the mean is independent of location. Over the extent of the test cell, the variance is dependent only upon the magnitude and direction of separation of the samples in space.
 

The block kriging equation (Carr, 1995) is

4.2.1.1.4
where u* = the block estimate for a block centered at location xo

N = number of data locations used to produce the estimate

ki = kriging weight for sample point i

xi = location of sample point i.

The block kriging weights (ki) are solved for using the linear system:

4.2.1.1.5
where r = covariance operator

hij = lag between samples i and j

k = vector containing kriging weights

and

4.2.1.1.6
This is the average estimated covariance for an orthogonal block. A Gaussian quadrature approach is used (Carr and Palmer, 1993) to estimate its value. Each block is defined by using the spatial extents of sampling as the domain and building a 20 x 20 x 20 grid of estimate locations.

The kriging weights must sum to unity for unbiasedness. Written in matrix notation, the kriging system solved is:

4.2.1.1.7
where l is a Lagrange multiplier. Notice there is no constraint that the kriging weights be greater than zero, which is often perfectly acceptable. However, this can produce erratic estimates. Therefore, a correction is applied. It consists of setting the negative weights to zero and restandardizing the remaining weights (Deutsch, 1996). This is done after the weights are obtained from Equation 4.2.1.1.7.

4.2.1.1.8
Isaaks and Srivastava (1989) present the following expression of the global error estimate of the mass estimate from kriging:

4.2.1.1.9
where lG is a vector of global weights, obtained by accumulating the individual (corrected) kriging weights each sample location received during all kriging iterations for each estimation node, then dividing the resulting vector of summed weights by the total number of estimation nodes. The quantityij  is the average covariance over the entire cell. It was obtained by applying the fitted semivariance model to all the possible pairs of estimation nodes to estimate covariances using the relationship:

4.2.1.1.10
The sine-hole model of the semivariance does not monotonically approach a sill. Therefore, the maximum model-predicted sill within the domain of sample lags was used as the variance for each estimation block.

The matrix ij contains the covariances between each pair of sample locations as in the kriging system matrix. The vector iA contains average covariances between each sample point and the entire cell region.

The reader is referred to Isaaks and Srivastava (1989) for an excellent description of the physical meaning of each of the three terms that comprise the error estimate in Equation 4.2.1.1.9.

The confidence interval on an estimate (U*) is calculated from the estimated standard devation, based on a normal distribution (de Marsily, 1983), using:

4.2.1.1.11
where M is the number of estimated blocks and

4.2.1.1.12
This final estimate of the uncertainty does not include the error in the semivariogram model estimate. For a more detailed description of the methods used in this analysis, the reader is referred to Faulkner et al. (in press).

The data (pre-remediation and post-remediation) are plotted for each chemical in each cell only as depth versus concentration.  The aerial information is utilized only in the inverse distance squared analysis and the three dimensional kriging and not displayed.  In addition to the data the mean and 95% error bars are plotted for the boxcar averaged analysis.  Error bars are only displayed when both the upper and lower limits are within the plotting range.  When the error bars exceed the plotting range only a tic mark is displayed.  Immediately above the figures is the calculated area of the cell as well as the calculated mass, mean concentrations, and variance of the mean for the different methods used.

4.2.2    Tracer analysis

Tracers are chemicals that are added to water to determine the flow path of the water and hydrodynamic properties of the media in contact with the water.  Hydrogeologic properties that can be identified include the ionic exchange capacity, the amount of hydrophobic material present, and the surface area of the hydrophobic surface, to name a few.  Tracers need to be carefully selected so that they perform as anticipated.  At Hill AFB three types of tracers were used. Nonreactive tracers were used to determine the water flow path and effective pore volume within the flow domain. Partitioning tracers were used to estimate the mass fraction of the NAPL. And surface-active tracers (surfactant-like compounds below the cmc) were used to determine the interfacial area of the NAPL perfused by the tracer.  For tracer analysis, each cell had four injection wells and three extraction wells and nominally 60 multilevel sampling points.  The discussion presented is limited to analysis of the extraction wells.  The attached database includes the data on the multilevel samplers.  Four different chemicals were used to determine the flow path, bromide, fluorescein, pentaflurobenzoicacid (PFBA), and methanol.  Laboratory studies demonstrated that bromide and PFBA could be effectively used as a conservative tracer by comparing the performance to tritiated water.  Methanol could also be used with only a small partition coefficient but biodegradation occurred complicating interpretation of the data.  Corrections can be made to minimize the error associated with biodegradation.  Assumptions need to be made regarding the type of reaction.  The most common approach is to calculate a first-order transformation rate which conserves mass.  This is relatively easy to perform but introduces assumptions that are not necessarily valid.  Fluorescein was used as a qualitative screening tool since analysis could be performed rapidly in the field.

To estimate the amount of hydrophobic material present or NAPL saturation, partitioning tracers were used.  Samples of NAPL were obtained from several sources throughout the site and candidate tracers were evaluated to obtain partition coefficients to the NAPL.  Studies were performed at both room temperature and at temperatures anticipated within the zone of contamination.  Temperature significantly changes the observed partition coefficient.  The partition coefficients used in this analysis are presented in Table 4.2.2.1.  The optimum partitioning tracer is one that will partition uniformly to the NAPL present at the site, remain stable throughout the duration of the study and have a partition coefficient that will give sufficient separation in the two elution curves to make it easy to interpret the results without increasing the time to a point that the amount of time required to complete the field activities is reasonable.  Assuming the tracer stability is equal, the optimum tracer would have the retardation that can be characterized during the field sampling event.  For this study, the most effective tracer used was 2,2-dimethyl-3-pentanol.
 

Table 4.2.2.1 NAPL:water Partition coefficients for the tracers
Tracer Chemical Partition Coefficient * Partition Coefficient † Partition Coefficient ‡ Partition Coefficient **
2,2-dimethyl-3-pentanol
7.42
11.8
10.28
12.9
hexanol
2.07
3.18
3.12
4.6
6-methyl-2-hepatnol
18.8
30.2
29.27
  
3-methyl-hexanol
4.77
      
t-butylalcohol
0.14
      

* Estimated values at 10o C (Gierke et al. in press and other sources) values used in this report
† Estimated value at 20o-25o C (Gierke et al. in press)
‡ Estimated Value at 20o-25o C (Meyers et al. 1997)
** Estimated Value at 20o-25o C (Annable et al.1998)

The NAPL saturation (Sn ) is directly related to the ratio of the time of travel for the partitioning tracer (tp) divided by the time of travel of the non-partitioning tracer (tn ) or the retardation factor (R) by the
equation

 4.2.2.1
where:
Knw is the partition coefficient as listed in Table 4.2.2.1
Sn is the NAPL saturation (L3/L3)
The above equation can be rewritten in terms of NAPL saturation and evaluated directly from the experimental data
 4.2.2.2
ti  is the mean time of travel of the pulse which is equal to the first moment divided by the zero moment minus one half of the time of the injected pulse. The subscripts p and n refer to the partitioning and non-partitioning tracer. This equation does not take into consideration any degradation of the chemical. By evaluating the difference between the zero moment of each tracer injected and the zero moment extracted it is possible to qualitatively assess the error introduced by this assumption.  The integrals can be evaluated by the trapezoidal rule.  As time increases the moment arm (t) has a big influence on the calculation. Small errors in concentration when multiplied by long times create significant errors in the calculated mean travel time.  Generally, concentrations are not measured until they reach background levels and it becomes desirable to extrapolate the experimental data to some predetermined relative concentration end point.  This is quite often done by fitting the last few measurements to an exponential function and calculating the integrals on extrapolated data.  In the figures presented here, only data above the quantification limit (0.5 mg/l for bromide and 1.0 for the alcohols) were used to extrapolate the curves.  In the tracer curves presented later, the filled red points are the reported experimental data.  The open green circles are the data used in the analysis.  For the data presented here, the concentrations were extrapolated to 0.001 mg/l when the moments were calculated with extrapolated data. Both unextrapolated and extrapolated data moment analyses were calculated.  There is a considerable amount of art in the process of extrapolating the data.  Generally, the relative error in a chemical analysis increases as one approaches the quantification limit.  The error is usually a percentage of the range + some small amount.  Efforts are made to minimize the errors near the quantification limits by using multiple standards but inevitably the percentage error of the true value increases as the quantification limit is approached. To determine the slope and intercept of the extrapolated line, the last 25% of the samples were used and extrapolation only took place when there were eight or more samples analyzed above the quantification limit.  The research groups contributing to this effort have individually used approaches which they felt were most appropriate for their individual data.  In this presentation, all of the data is being analyzed in the same internally consistent manner.  The result is significant differences between the results presented here and the results presented elsewhere in the literature for the same data. The question arises as to the precision that should be placed on this type of analysis. As a rule-of-thumb, it is suggested that reliable moment analysis should be based on ten measurements on the rising side of the elution curve and 20 measurements on the falling side of the elution curve.   Ideally, one should know the time of the injection pulse and the concentration of the injection with a start time of the pulse injection.  The injected chemicals should be uniformly mixed and the fluid flux maintained constant both before, during and after injection for a sufficient time to perform the test.  The extraction well flux needs to also be maintained at a constant flux such that the flow path remains constant during the entire measurement period.  To establish a flow path using a rule of thumb, two or more pore volumes of water would be injected and extracted at a steady flux prior to the injection of the tracers.  As continuous monitors become available and tracer analytical detection and quantification limits decrease, the accuracy of the analysis will improve. When the tracer analyses are being used to evaluate the performance of a remedial system, it is critical that the same volume of the aquifer be swept both before and after remediation and have this swept volume be the same as the volume being remediated by the technology.  If fewer data are available, it might be preferable to perform an inversion of the data to a solution to the convective-dispersive transport equation.

Two inversion techniques were utilized to demonstrate the feasibility of inverting the data to determine the NAPL saturation and an estimate of the heterogeneity of the NAPL within the flow field.  Both approaches solve the conservative one-dimensional convective-dispersive equation

 4.2.2.3
where:
 R = retardation factor
 c = concentration of the contaminant (M/M)
 D = hydrodynamic dispersion coefficient (L2/T)
 t = time (T)
 x = distanced along the flow path (L) and
 v = interstitial fluid velocity (L/T)
Using a one-dimensional model to describe a three-dimensional problem is a significant simplification. One should not expect a good correlation between the model fit and the experimental data.  The only time one would anticipate a reasonable correlation is when the system is homogeneous and the variability in the path length of the flow lines is small compared to the total length of the flow line.
Equation 4.2.1.3 was solved with flux boundary conditions after van Genuchten and Alves (1982) as
 
 4.2.2.4
where:
to is the time for the injected pulse tpulsein equation 4.2.2.2
co is the concentration of the injected pulse
ci is the resident background concentration 0 for this application


The solution is written in terms of dispersion.  Freeze and Cherry (1979) suggest that dispersion is a function of velocity following the equation
 


where:

D* = molecular diffusion of the solute in the porous medium (L2/T); and
d = dynamic dispersivity
Bresler (1973) suggested that D* could be estimated by the equation:
 
where:
D0 = molecular diffusion of the solute in the aqueous phase (L2/T)
a = empirical constant with a value ranging from 0.001 to 0.005
b = empirical constant with a value of approximately 10 and
q  = aqueous filled porosity (L3/L3)
The approach was intended for use in water unsaturated soils but is believed to be appropriate under water saturated conditions to correct for tortuosity of the formation. The quality and completeness of the data are quite variable from cell to cell. When all of the data were available and appeared reasonable, the inversion to equation 4.2.2.4 of the non-partitioning tracer calculated the time of travel and an apparent dispersion coefficient.  Similarly, the time of travel and the apparent dispersion coefficient were calculated for the retarded tracer.  Based on the calculated time of travel and the partition coefficient for the tracer into the NAPL, the saturation is calculated.  If the NAPL is distributed uniformly, the dispersivity should be a constant for both the non-retarded tracer and the retarded tracer.  From the results obtained during this study, there was a considerable difference between the apparent dispersivities suggesting a non-uniform distribution of the NAPL.

In a second approach, we extend the conceptual model presented by Millington and Quirk (1961). Millington and Quirk hypothesized the hydraulic conductivity function for a porous media could be described by transfer function describing variability in the flow of water through a series of capillary tubes with selected diameters which would mimic the pore size distribution of the soil.  This approach accounts for both the imbibition and drainage hysteretic nature of the soil characteristic curve and hydraulic conductivity.

A similar approach to describe chemical transport using a Darcy scale rather than  a pore scale analogy is used here.  In the analogy, each of the N-Darcy scale units would have a characteristic hydraulic conductivity, characteristic dispersivity, and characteristic partition coefficient or NAPL saturation.  Each of the N units would represent an equal fraction of the area normal to the flow field.  The hydraulic conductivity distribution is used to weight the flux through each of the N layers.  Chemical transport is calculated using a one-dimensional, convective-dispersive solution and the contaminant flux is recombined using the principle of superposition at the point of interest to describe the contaminant flux for the system.  A significant limitation of this approach is that it does not allow for diffusion from one flow path to another flow path even though there is a concentration gradient. However, using the approach permits getting a feel of the heterogeneity and, if an adequate inversion of the breakthrough curves is obtained, there is also a knowledge of the relationship between the NAPL and the flow field.

In the field, the hydraulic conductivity is described as being log-normally distributed (e.g., Nielsen et al., 1973; Warrick et al., 1977, and Sisson and Wierenga, 1981).  As an initial estimate, the hydraulic conductivity was assumed log-normally distributed and a constant dispersivity of (d = 0.0001 L), the variance (the square of the standard deviation) in hydraulic conductivity or time of travel was estimated.  For the retarded chemical, it was assumed that the NAPL was uniformly distributed.
 

4.2.0 Cosolvent Solubilization (Cell f)

4.3.0 Cosolvent Mobilization (Cell 3)

4.4.0 Surfactant Solubilization (Cell 6)

4.5.0 Surfactant Mobilization (Cell 5)

4.6.0 Micro Emulsions

4.7.0 Macro Molecules (Cell 4)

4.8.0 Steam

4.9.0 Sparging and Venting (Cell 1)

4.10.0 In-well aeration (Cell 2)