Assessing the Hydraulic Performance of a Bioremediation Curtain for Plume G 
at the Schoolcraft Site, Michigan Using Interactive Groundwater (IGW)

OBJECTIVES

We describe in this webpage a 3D flow and transport modeling effort for a contaminated aquifer at the Schoolcraft site in western Michigan (Figure 1) using Interactive Groundwater (IGW) (Li and Liu, in review). The ultimate goal is to design a broadly applicable ‘‘biocurtain’’ which be applied to remediate this site using a periodic recirculation well gallery installed normal to groundwater flow. The project takes advantage of IGW’s real-time and hierarchical modeling (Li et al., in press) and visualization capability for iterative conceptualizations, data sufficiency evaluations, hypothesis testing, and remediation design.

Figure 1 - The regional head pattern and the VOC plume (the plume G) in the vicinity of the village of Schoolcraft, MI.

The specific objectives of this modeling effort are:

• To design a cost effective, in-situ biocurtain that can intercept the entire plume G and to optimize key design parameters of the recirculation gallery system, including 1) the spacing between wells within a delivery unit, 2) the spacing between the delivery units, and 3) the operational schedule (e.g., the shut off time), and

• To evaluate the effects of aquifer heterogeneity and seasonal irrigation on the hydraulic performance of the remediation system.

Figure 2 shows a conceptual sketch of the delivery system involving two recirculation units. The well spacing that needs to be optimized are marked with a question mark.

Figure 2 - A conceptual sketch of the delivery system. The objective is to maximize the well spacing within the delivery unit and the spacing between the units and to achieve a tracer breakthrough of at least 40% at the extraction well and 20% at point W between the units in 24 hours.

SCALE DEPENDENT FLOW SYSTEMS

Characterization of flow and transport phenomena during the delivery and treatment process is crucial for designing the biocurtain. But this process can be complicated by potential interaction of flow dynamics across a number of spatial scales present at the site and in the vicinity. These different spatial scales, illustrated in Figure 3, include: the “delivery unit-scale” O(10m), the “biocurtain scale” –O(100m), the “site scale” –O(1000m), and the “regional scale” –O(10,000m) (see Figure 3). 

Figure 3 - A multi-scaled groundwater system: the “delivery unit-scale” O(10m), the “biocurtain scale” –O(100m), the “site scale” –O(1000m), and the “regional scale” –O(10,000m)

At the unit scale, the closely-spaced injection and extraction wells create rapid head variations and a fine grid (on the order of 10 centimeter) is required to resolve the local spatial dynamics. For the accurate modeling of transport and delivery process an even smaller grid size may be required to drive numerical dispersion to an acceptable level. Creating such a high resolution, 3D unsteady model for large numbers of evaluations on a desktop computer is only feasible for a relatively small area. 

On the other hand the presence of heterogeneity can potentially induce unit-to-unit interactions and evaluation of this interactions requires high resolution modeling of multiple delivery units or the entire biocurtain. 

Additionally, potential impact from seasonal irrigation can further complicate the site characterization since it may require expanding substantially the modeling domain from the site-scale to a regional scale (on the order 10,000 meters).

REAL TIME MODELING AND HIERARCHICAL PATCH DYNAMICS MODELING

Here we apply IGW real-time modeling (Li and Liu, in review) and IGW hierarchical patch dynamics paradigm (HPDP) (Li et al., in press) to the design of 3D delivery and transport at the Schoolcraft site and to investigate how the potential complex interplay of the flow dynamics across multiple scales in the vicinity may influence the hydraulic performance of the remediation design.

IGW is a 3D real-time groundwater modeling environment. The software functions as a numerical laboratory in which the modeler can freely explore: visually creating an aquifer of desired configurations and then immediately investigating and visualizing the groundwater system. IGW allows the modeler's thought processes to progress naturally and intuitively with the information visualized, overlaid, and compared at the instant it is required for analysis, providing a real sense of continuous active exploration and engaged problem solving [Li and Liu, in review; Li and Liu, 2003].

IGW’s hierarchical patch dynamics paradigm (HPDP) takes advantage of hierarchy theory, divides and conquers complexities, and decouples scale-dependent dynamics hierarchically. The HPDP provides dynamic model coupling, visual interactive steering, and freeing the modelers from the impractical task of having to interact offline and iteratively with potentially large numbers of modeling patches. The HPDP provides a valuable tool for understanding scale-dependent processes and allows solving complex problems without solving large (often-ill-posed) matrix systems [Li et al., in press] 

SITE CHARACTERIZATION

Our modeling effort was systematically integrated with field investigations, including 1) local drilling to characterize the site lithology, 2) soil sampling, lab column experiments, and aquifer tests to determine the hydraulic properties, and 3) tracer tests to understand the 3D delivery and recirculation processes.

Figure 4 shows a typical stratigraphic cross-section developed based on borehole data collected at the pilot testing site.

Figure 5 shows the vertical distribution of log hydraulic conductivity estimated based on a large number of column experiments of soil samples collected at different locations at the site.

Figure 5. Measured hydraulic conductivity as a function of depth [source Graulau-Santiago 2003]. Note that hydraulic conductivity is significantly higher in the deeper layers at a depth of 60 to 80 feet below the land surface. These deeper layers are also where the bulk of the VOC contamination was found.

Figure 6 shows a sketch of the tracer test setup involving a single recirculation unit. Water was continuously pumped from the extraction well ( FC), mixed in a tank with tracer, and injected back into the aquifer through two injection wells ( DW).   

Figure  6.  Layout of the tracer injection system [Source Graulau-Santiago, 2003].

Table 1 presents additional details of the tracer tests.

Table 1.  Additional details of the tracer experiments [source Graulau-Santiago 2003].  

UNIT-SCALE MODELING AND CALIBRATION

Results obtained from the site characterization efforts were used to develop an integrated 3D flow and transport modeling system at the pilot testing site. The pilot modeling system simulates tracer delivery and transport dynamics in response to the regional flow and the local injection and extraction. Extensive sensitivity simulations were performed to understand conceptually the delivery dynamics and to identify dominant factors controlling the performance of the remediation system.

The unit-scale model consists of two conceptual layers. The first layer, from the land surface to an elevation 18 meters below, is unconfined. The second conceptual layer includes 4 deeper lithologic layers (see Figure 4). These deeper layers were lumped into a single conceptual layer because the injection/extraction wells were screened across all 4 layers and the tracer measurements obtained represent their average. The clay at the base of the layered system was assumed to be impervious. 

Figure 7 shows the conceptual representation in plan view and a 3D perspective.

Figure 7- Conceptual representation and boundary conditions for the tracer scale model (left - plan view, right - 3D view). The model include two conceptual layers.  Layer 1 represents the sand from land surface to 18 m below.  Layer 2 represents SP, GS, GW, and GS. The clay at the base was assumed to be impervious. The injection/extraction wells were screened across the entire conceptual layer 2. The first layer is unconfined. Two thin computational layers were added at the interface of the two conceptual layers to resolve the sharp vertical gradient in the tracer concentration.

The integrated pilot modeling system was calibrated to the tracer test. The hydraulic conductivity, effective porosity, and dispersivities were assumed to be constant at the unit scale and were adjusted such that the predicted breakthroughs in tracer concentration at both the injection and extraction wells match the observed.

Figure 8 presents a real-time visualization of the delivery and transport process and a comparison of the predicted and observed tracer concentration at the extraction well (upper right) and injection well (lower right). 

Figure 10 shows the calibrated head and concentration distributions at the end of the 4 hour injection period in three-space dimensions.

Figure 10 - 3D head and concentration distribution at t=4 hours.

Table 2 presents the inputs and parameters used in the calibrated model.

SELECTING WELLING SPACING

The calibrated unit-scale model was used to investigate key design issues and to optimize the unit-scale remediation system. The objective was to ensure that the concentration at the extraction well can achieve at least 40% of the injection concentration in 24 hours and to maximize the spacing between the wells. 

Extensive simulations and a systematic sensitivity analysis show that this maximal well spacing allowed is approximately 10 meters. Figure 11 shows the final design of the recirculation well system at the unit scale.

Figure 11 - Final design of the recirculation system at the unit scale.

BIOCURTAIN SCALE MODELING AND SELECTING SPACING BETWEEN THE DELIVERY UNITS

To select the spacing between delivery units, the calibrated model was expanded to simulate multiple re-circulation units. The expanded modeling system allows modeling unit-to-unit interaction. The design objective was to ensure that the maximum concentration at W (see figure 2 for the location) between the delivery units can reach at least 20% of the injection concentration in 24 hours and to maximize the spacing between the delivery units.  

Systematic sensitivity simulations show that this maximal spacing allowed between the delivery units is approximately 7 meters, as illustrated in Figure 12.

Figure 12 - Final design of the recirculation well system

Figure 13 shows a plan view of the conceptual representation for the biocurtain scale modeling system.

Figure 13- A conceptual sketch showing a 4 set of 10 m extraction and injection wells (plan view).

Figure 14 shows a real-time simulation and visualization of the delivery process at the biocurtain scale based on the design spacing.

Figure 14-Real-time visualization of the delivery process for a duration of 24 hours (4.77 MB) in conceptual layer 2.

Figures 15a and 15b show a snapshot of the head (left) and tracer concentration distribution (right) at the end of the 24 hour injection period.

Figure 15a- Steady state head contour map in conceptual layer 2 for 4 sets of 10 meter extraction injection well at biocurtain scale model.	Figure 15b- Tracer distribution in conceptual layer 2 at the end of the 24 hour injection period for 4 sets of 10 meter extraction injection well at biocurtain scale model.

Figure 16 shows a comparison of the predicted breakthrough at the extraction (left) and injection (right) wells for one and four sets of 10 meter well spacing model. The results show that the breakthroughs both monitoring wells (the injection and extraction wells) for the multi-unit model (dashed lines) increase significantly due to the cross unit interaction. This means that the spacing designed based on a single unit model is conservative.

STOCHASTIC MODELING AND EVALUATING THE EFFECT OF HETEROGENEITY

To evaluate the potential impact of spatial heterogeneity on the performance of the biocurtain, a stochastic Monte Carlo model was developed at the biocurtain scale. The stochastic model is the essentially the same as the original biocurtain scale model, except that the hydraulic conductivity in the second conceptual layer (where the tracer was injected) was assumed to be a spatially-correlated random field. This random representation mimics the larger scale variability that may exist but could not be captured by local sampling at the pilot test site. The spatial statistics used to describe this heterogeneity were based on the conductivity data collected in the general area from prior studies [Graulau-Santiago, 2003].

Figure 17a shows a real-time simulation and visualizaiton of one realization of the predicted biocurtain for a 24 hour delivery period in the presence of heterogeneity.

Figure 17a - Real-time visual simulation at the biocurtain scale in the presence of heterogeneity; lnK variance = 0.15, correlation scale = 18 meters [parameters used based on Graulau-Santiago, 2003].

Figures 18 and 19 show a comparison of the concentration breakthrough at a monitoring well and between the delivery units for 5 different random realizations.

Figure 18- Comparison of tracer breakthroughs at the extraction well for 5 different random realizations of hydraulic conductivity and for a deterministic constant conductivity case. Note the impact of heterogeneity is small, with the final breakthrough at the extraction well varying from approximately 50% to 60%.		Figure 19- Comparison of tracer breakthroughs at between the delivery units  for 5 different random realizations of hydraulic conductivity and for a deterministic constant conductivity case. Note the lowest breakthrough value for the case of homogeneous case.

Figure 20 shows the same comparison along profile D-D (see Figure 12) at the end of the 24 hour simulation time.

Note that introduction of unit-scale spatial heterogeneity affects little the concentration distribution at the injection well. However, the heterogeneity affects significantly the concentration breakthrough between the delivery units, leading to increased concentration breakthrough. In other words, the original deterministic model without considering the effects of heterogeneity lead to a conservative design.

CYCLIC MODELING AND SELECTING THE DELIVERY SCHEDULE

To minimize the remediation operational cost, the recirculation system was designed to be cyclic. Specifically, the shutoff time was designed to be less than one half of the time it takes for the curtain to move out of the circulation zone with the regional flow. This would avoid contamination leaking through the well gallery untreated or under-treated. 

Systematic sensitivity simulations show that this maximum “shut off” time allowed is approximately 14 days or 2 weeks. Figure 21 shows the biocurtain distribution at different times throughout the recirculation and "recession" period.

Note that the tracer curtain moved with regional flow approximately 5 to 7 meters or about half of the curtain width in 14 days.

REGIONAL HIERARCHICAL MODELING AND EVALUATING THE EFFECTS OF SEASONAL IRRIGATION

To evaluate the potential impact of seasonal irrigation on the delivery and curtain dynamics, a hierarchical modeling system was developed. The integrated system includes a cascade of 5 nested models that allow simulating iteratively the effects of regional influence on the local delivery and transport. Systematic hierarchical simulations show that the impact of seasonal irrigation on the tracer transport is small during the active delivery period, but significant when the recirculation system is shut off. In particular, the model shows that the irrigation causes the regional gradient in the curtain area to increase by approximately 60-70% and thus the tracer injected and the VOC upstream of the curtain may potentially migrate past the recirculation zone untreated during the shut off time. 

Figure 22 shows the location of the three irrigation wells that exist in the vicinity of the plume G. 

 

Figure 22 . Locations of existing high capacity irrigation wells [circled] in the vicinity of the plume G. Note the different orientation  as compared to Figure 1.

Figure 23 presents a hierarchical simulation of flow and transport in the presence of background irrigation pumping. The flow simulation was performed at steady state. The transport simulation was performed for the 24 hour injection period. The dynamically coupled hierarchical modeling approach allows us to model efficiently the complex flow system across multiple scales.

Figure 23- Hierarchical modeling of the effects of regional irrigation on the local tracer delivery dynamics.

Figure 24 shows the concentration breakthroughs at a typical extraction well with and without considering irrigation. 

Figure 24- Comparison of tracer breakthroughs at an extraction well during the one day recirculation with and without considering irrigation.  Note that irrigation has little impact on the tracer breakthrough during the recirculation phase.

Figure 25 shows the tracer concentration along the profile D-D (see figure 12 for the location) with and without irrigation. As one can see, the impact of irrigation on the delivery during the 24 hour recirculation period is minimal.

Figure 25- Concentration profiles (along D-D) with and with our regional pumping. Irrigation has little impact on the spatial distribution of the tracer concentration during the recirculation phase.

However, during the following 14 day "recession", the impact from the irrigation becomes important. The regional gradient in the curtain area increases significantly (by approximately 60% to 70%), creating a possibility that the plume upstream of the curtain may move past the treatment zone untreated periodically (every weeks). This suggests that the shut off time designed without considering the effects of irrigation may need to be reduced.

Figure 26- After shutting down the recirculation system we notice a difference that is significant and may potentially affect the operational schedule of the recirculation system.

Figure 27  Head distribution along the regional flow direction for four different scenarios: a) natural gradient, no pumping,  b) with irrigation and no recirculation, c) with irrigation and recirculation, and d) with recirculation and no irrigation. Head gradient in the curtain area increases by approximately 60-70% during the irrigation season.

CONCLUSIONS

We have developed an integrated IGW modeling system to evaluate and optimize the hydraulic performance of the bioremediation curtain at the plume G site, Schoolcraft, MI. The unit scale model was calibrated to the tracer data and the results show that the model compared well with the observations at both extraction and injection wells. The calibrated modeling system was used to design hydraulically the bioremediation system. Systematic model sensitivity simulations show that the maximum well spacing allowed is approximately 10 meters, the maximum spacing between the units is approximately 7 meters, and the maximum shutoff time is approximately 2 weeks (when the impact of irrigation was ignored). The biocurtain scale model shows that the interaction between the delivery units enhances the performance of the delivery system. The regional hierarchical model shows that the impact of irrigation on the biocurtain dynamics was significant when the recirculation was shutoff. This finding may have major implication on the design operational schedule and will be tested in the next phase of the field investigation.

 ACKNOWLEDGEMENTS

•  Michigan Department of Environmental Quality (MDEQ) 

• National Science Foundation (NSF)

REFERENCES

Soheil, Afshari (2006). "Application of a Hierarchical Patch Dynamics Paradigm (HPDP) for Modeling Complex Groundwater Systems Across Multiple Spatial and Temporal Scales" Ph.D. Desertion, Michigan State University, East Lansing, MI.

Graulau, Santiago J. (2003). " Development and application of a methodology to evaluate natural attenuation of chlorinated solvents using conceptual and numerical models" Ph.D. Desertion, Michigan State University, East Lansing, MI.

Li, S.G. and Q. Liu, "A real-time, computational steering environment for integrated groundwater modeling". In review/Under revision, Ground Water.

Li, S.G., Q. Liu, and S. Afshari, "An Object-Oriented Hierarchical Patch Dynamics Paradigm (HPDP) for Groundwater Modeling". In press. Environmental Modeling and Software.