Grantee Research Project Results
Final Report: An Efficient Reliability-Based Approach to Aquifer Remediation Design
EPA Grant Number: R827126Title: An Efficient Reliability-Based Approach to Aquifer Remediation Design
Investigators: Reeves, Howard W. , Dowding, Charles H. , Igusa, Takeru
Institution: Northwestern University
EPA Project Officer: Aja, Hayley
Project Period: September 1, 1998 through August 31, 2000 (Extended to June 30, 2001)
Project Amount: $142,198
RFA: Exploratory Research - Environmental Engineering (1998) RFA Text | Recipients Lists
Research Category: Safer Chemicals , Sustainable and Healthy Communities , Land and Waste Management
Objective:
Evaluation of remediation alternatives for a contaminated site requires answering two questions: Are there enough field data to assess the performance and reliability of various remedial alternatives? and Where should additional data be collected to increase confidence in the remedial design? The resolution of these questions requires consideration of both the inherent uncertainty associated with characterization of geologic conditions and the impact of this uncertainty upon remedial design. Research reported herein has led to the development of a quantitative approach answering these questions. There are three key aspects to the approach: (1) numerical models describing groundwater flow and contaminant transport modified to produce additional required information, (2) a probabilistic geologic description of a site that provides spatially correlated information and accounts for the typically small number of data from the site, and (3) a quantitative cleanup goal at selected locations at the site. An important aspect of the work is to make the most efficient use of limited data by incorporating, as much as possible, any prior information about the characteristics of the site.Numerical models are the center of the research because they often are used for site-specific analysis and design of cleanup schemes for contaminated soil and groundwater. These models require a large suite of input parameters to define the geometry of the geologic setting, the hydrogeologic parameters at the site, and the chemical and biological processes impacting the contaminant. The models predict system performance that can be either groundwater piezometric head or contaminant concentration. The probabilistic approach implemented in this research builds upon this numerical modeling by including both the best estimate of the input parameters and an estimate of the uncertainty in these parameters in the modeling process. With this information, the system performance is simulated and the uncertainty of this performance also is calculated. System reliability, which is an indication of the probability of success of a project, is determined by combining the model prediction, its estimated uncertainty, and an imposed design goal. System reliability for different alternative remediation schemes can be computed, compared, and used in the selection of the most appropriate scheme for a given site with the available information. Additionally, the method identifies the most important data impacting the reliability at the site and, thereby, may be used to determine the location and nature of additional site characterization efforts.
Three main challenges in implementing this method are: (1) determination of input data estimates and uncertainty, (2) computation of system performance and associated uncertainty in the system performance, and (3) evaluation of the reliability of the system. To meet these challenges, a Bayesian approach is used to develop a framework for quantifying the uncertainty in the input parameters and systematically updating this uncertainty with field data. The Bayesian approach allows for a superior estimate of the spatial correlation of the uncertain input parameters compared to traditional geostatistical approach such as simple or ordinary kriging. It also provides a quantitative way to include prior information regarding the site. To compute the system response, a first-order second-moment approach is adopted. While this selection limits application to estimates of only the first and second moments of the system behavior, this approach is selected because it is more computationally efficient than Monte Carlo techniques that may produce higher-order information. System reliability is expressed through the use of the reliability index. The reliability index is used because it complements the first-order second-moment approximation of performance quite well, can be computed very efficiently, and provides a direct comparison of computed performance to the design goal.
The objectives of the research were to:
1. Implement an efficient computational approach to evaluate the impact of parameter uncertainty on the reliability of remedial schemes design through the use of a three-dimensional groundwater flow and contaminant transport code. Other approaches presented in the literature to assess reliability rely on Monte Carlo or perturbation approaches and are thus limited in their applicability due to an inherently large computational burden.
2. Employ the computational approach to demonstrate that both data uncertainty and the impact of this uncertainty on the design model can and must be quantitatively considered in the calculation of the reliability of competing designs and in the implementation of efficient site characterization schemes.
3. Demonstrate the effectiveness of the approach on the uncertainty of hydrogeologic interface location and the impact of this uncertainty on remedial design. The issue of hydrogeologic interface uncertainty has not been adequately addressed in the literature as most previous research has focused on the uncertainty associated with hydrogeologic properties within each unit after unit boundaries have been assumed.
4. Test the assumptions invoked in the research regarding the proper form of the uncertainty equations and the appropriateness of the first-order second-moment expansion to determine the uncertainty in the calculated model performance resulting from the uncertainty in the interface locations that are input to the model.
Summary/Accomplishments (Outputs/Outcomes):
The following are findings based on meeting the four objectives above:1. Two-dimensional and three-dimensional groundwater flow and transport models were modified to directly compute the sensitivities required to implement the first-order second-moment algorithm proposed for the research. The direct technique was shown to be at least an order of magnitude faster than standard perturbation-based techniques. As expected from published results in the literature, the final model produces estimates of uncertainty orders of magnitude faster than Monte Carlo techniques for the large number of uncertain parameters necessary for spatially complex test cases.
In addition to developing two- and three-dimensional finite element models to test the computational approach, an industry standard code, MODFLOW, also was used within the computational framework. The latest version of this code, MODFLOW-2000, provides required information that can be used in the first-order second-moment analysis relating uncertainty in predicted groundwater heads to uncertainty in spatially correlated input parameters. Extension of this work to predict contaminant transport was outlined and briefly explored with a simple particle tracking strategy.
One important advance made during this project over previous work of our group, which focused on steady-state settlement of soils due to building loads, was the implementation of the methods for transient problems. Transient analysis allows the study of the time-dependent uncertainty of the system performance as a contaminant plume moves through heterogeneous, three-dimensional geology. This information may be used to identify important parameters for site exploration or parameters that control the effectiveness of different remediation schemes addressing the contamination.
2. The effectiveness of the computationally efficient approach for site exploration and remediation system evaluation was demonstrated with two- and three-dimensional finite element models and MODFLOW-2000. The models were applied to both synthetic data sets and to a set of field data from a site that has been studied extensively by the U.S. Geological Survey. More work is required on both field cases and synthetic domains to test the methods for a range of hydrogeologic conditions, a range of contaminant types, and for different remediation system designs.
Remediation system performance was quantified with a reliability index that was shown to be useful for the test cases studied. More work is required to develop a spatially correlated reliability index when multiple compliance points are required. Also, more work should be focused on comparing the results of the reliability index evaluation to other reliability approaches published in the literature.
The research examined different approaches to guide exploration based on the estimated uncertainties and other information generated during the modeling process. In general, guiding exploration with information generated during the computation of performance more efficiently reduced updated design uncertainty compared to exploration at locations that bisected the distances from existing samples. In addition, the method developed allows the user to not only determine the location with the greatest uncertainty in concentration or groundwater head, but the location that causes the most uncertainty in concentration or head at a specific location also may be identified. Therefore, exploration can be directed to reduce uncertainty in concentration or head over the entire site, or it can be directed to increase system performance and reliability at selected locations on the site. The inclusion of system performance is novel in statistical analysis of spatial uncertainty. These results highlight and confirm the proposition that both system performance sensitivity and input uncertainty must be considered when attempting to quantify site exploration.
3. In addition to examining interface location and thus layer thickness as an uncertain input parameter, initial concentration and first-order decay uncertainties for the contaminant were examined. For a synthetic pump-and-treat remediation system, the layer thickness uncertainty was found to dominate the uncertainty in the estimated concentration behavior. In the case of natural attenuation, the first-order decay of the contaminant was much more important and became the dominant cause of uncertainty in the remediation system performance. Further testing with synthetic domains and field data is required to demonstrate the generality of these results and to illustrate the conditions when different input parameters dominate uncertainty in performance.
4. Geostatistical approaches and models used in the site characterization phase of the algorithm are extremely important in the implementation of the algorithm to guide site exploration and to evaluate remediation system designs. Use of a variogram model with uncertain parameters to describe spatial correlation at the site was explored during this research. More effort should be directed toward testing the Bayesian approach adopted in this research and, again, in testing different hydrogeologic systems to explore the range of applicability of the approach.
Design of economical and effective remedial systems will be enhanced by computing the reliability of alternative remedial designs and by integrating site exploration into the design process. Application of the approach demonstrated in this project should be useful to quantitatively determine the most appropriate remedial scheme to address soil groundwater contamination at specific sites with unique geologies and pollutants. Schemes that may be evaluated include both natural attenuation, where engineered controls are not imposed on the remedial scheme, and remedial alternatives that employ engineered controls. Determination of the reliability of designs that incorporate pumping wells, injection wells, cut-off walls, reactive barriers, or natural attenuation allows a quantitative comparison of each remedial alternative in specific geologic settings with given levels of exploration. The performance and reliability of each alternative may be considered along with the cost of each alternative in the selection of the best technology to address contamination at the site.
Journal Articles:
No journal articles submitted with this report: View all 2 publications for this projectSupplemental Keywords:
groundwater, remediation, cleanup, decision making, Bayesian analysis, geostatistics, environmental engineering, hydrology, geology, modeling., Scientific Discipline, Air, Waste, Hydrology, Remediation, Environmental Chemistry, Ecological Risk Assessment, Groundwater remediation, Engineering, Chemistry, & Physics, fate and transport, computationally efficient algorithm, contaminant transport, three dimensional transport model, hydrogeologic unit boundry, remedial design, groundwater flow, aquifer remediation design, water qualityRelevant Websites:
http://www.civil.northwestern.edu/people/reeves/reliability/Progress and Final Reports:
Original AbstractThe perspectives, information and conclusions conveyed in research project abstracts, progress reports, final reports, journal abstracts and journal publications convey the viewpoints of the principal investigator and may not represent the views and policies of ORD and EPA. Conclusions drawn by the principal investigators have not been reviewed by the Agency.