H23B-1299
Sensitivity Analyses of Groundwater Modeling Associated with Boundary Condition Uncertainties
In groundwater modeling, the appropriate representation of hydrologic boundaries is essential to develop a reliable model. The main purpose of this study is to examine how the model predictions are affected by uncertainties in the conceptualization of the hydrologic boundaries including groundwater divides, streams, and the lower boundaries of the groundwater flow system. First, a groundwater flow model is constructed and calibrated for the study area using the Visual Modflow code. The recharge rate is employed for the unknown variables determined through the calibration process. Secondly, a series of sensitivity analyses are performed to evaluate the effects of model uncertainties embedded not only in specifying the boundary conditions for streams and groundwater divides, but also in specifying the lower boundary of the bedrock. Finally, this study provides some guidelines on how to deal with such hydrologic boundaries in view of developing a reliable conceptual model for the groundwater flow system of Korea.
H23B-1300
Sensitivity Analysis of Hydraulic Parameters on Radionuclide Transport Uncertainty in the Unsaturated Zone of Yucca Mountain
The parametric uncertainty and heterogeneity in matrix permeability, porosity and matrix van Genuchten alpha and n can cause significant uncertainty of unsaturated flow and radionuclide transport predictions in the unsaturated zone (UZ) of Yucca Mountain (YM). Sensitivity analysis is necessary to determine the contribution of individual hydraulic parameters to the uncertainty of flow and radionuclide transport processes. In this study, we conduct the sensitivity analysis of the hydraulic parameters (i.e., matrix permeability, porosity, van Genuchten alpha and n) on the flow and transport uncertainty using a sampling-based (i.e. Monte Carlo) approach in the three-dimensional heterogeneous UZ of YM. The random fields of the hydraulic parameters were generated using Latin Hypercube Sampling (LHS) and the Monte Carlo simulations of flow and transport were conducted. The stepwise regression analysis is applied to find the relationships between the input variables (i.e., hydraulic parameters) and the output variables (e.g., saturation, percolation flux, and radionuclide travel time) and then to rank the importance of uncertainty in individual parameters to the flow and transport uncertainty using the standardized rank regression coefficient (SRRC) for each gridblock. The top down coefficient of concordance (TDCC) is then used to measure the agreement of input parameter rankings obtained by stepwise regression analysis for different gridblocks in the same hydrogeologic layer. Therefore, the important input variables for each hydrogeologic layer can be identified and their rankings can be obtained. The analysis results indicate that the matrix permeability and van Genuchten alpha are most influential parameters on the uncertainty of flow and transport for most hydrogeologic layers. The outcome of this study can not only reduce the computational cost but also provide the important information about which parameters should be the focus for future monitoring efforts.
H23B-1301
Impact of Uncertainty Estimation on Early Assessment of Technology Options in Contaminated Land Management
Large-scale sites with contaminated land and groundwater constitute a complex system with numerous interacting processes that impact not only the environment but also major areas of society like human health, economy and the environment. Due to scale and complexity there are usually a great number of possible management options to be considered at these sites. Since a detailed investigation of all these options is not economically feasible, a streamlining of the planning and decision process is a mandatory requirement, beginning with an early selection of promising management options based on a preliminary assessment. In this context, the knowledge of uncertainties can help to gear future, more detailed investigations towards favored options and thereby streamline the management process. In this presentation we present the results of an investigation carried out to determine and rank parameters with respect to their relative importance to the cost–efficiency of remediation options. Monte-Carlo simulations using the assessment model CAROplus were used to analyze the effect of uncertain parameters in the context of various groundwater contamination scenarios (LNAPL and DNAPL pools as well as residual organic phase i.e. blobs).
H23B-1302
The Effects of Spatial Uncertainty on Modelling the Saltwater-Freshwater Interface for a Coastal Groundwater Management Problem
Management of coastal aquifers involves the use of numerical models to predict the effects of pumping on the hydrologic properties of the aquifer. Of great importance in these types of management problems is the correct prediction of the saltwater-freshwater interface to avoid saltwater intrusion issues. Often municipalities place upper limits on the allowable amount of pumping to avoid such issues. When these limits are based upon deterministic groundwater flow models they may be set conservatively high in order to avoid the risks associated with the uncertainty of the groundwater model. This research investigates the effects of the uncertainty in a groundwater flow model on the prediction of the location of the saltwater-freshwater interface. A hypothetical model based upon the Lower Cape Cod Aquifer in Massachusetts is developed for this study. Monte-Carlo simulations are conducted to quantify the effects of uncertainty on the saltwater-freshwater interface movement due to groundwater pumping. A conservative management system that takes into account the uncertainty in the model is developed using an optimization problem where the degree of risk in the problem is considered.
H23B-1303
Environmental Risk Propagation From Contaminated Land Through Groundwater
We present and use a limiting-scenario methodology for investigating the spatial development of environmental risk, defined as the probability of groundwater contaminant concentrations downstream of accidentally or industrially contaminated land to exceed given environmental target (maximum allowed) concentration levels. We use a stochastic advective travel time-attenuation approach to model contaminant transport for different extreme scenarios of contaminant release dynamics and downstream aquifer heterogeneity structures. For both short- pulse and long-term continuous contaminant releases, the environmental risk decline with transport distance is relatively insensitive to the degree of aquifer heterogeneity (hydraulic conductivity variance). It is further also practically insensitive to aquifer heterogeneity structure (isotropy/anisotropy, spatial correlation length) for short- pulse contaminant releases, and to contaminant release dynamics (short-pulse or long-term continuous release) for relatively isotropic aquifer heterogeneity structures. Above a certain ratio between mean attenuation rate and mean groundwater flow velocity, the transport distance at which environmental risk falls below, e.g., the 1% level is insensitive to the relation between environmental target and source input concentration for all investigated extreme contaminant release and aquifer heterogeneity scenarios. It is important to identify such risk development patterns and similarities for different extreme scenarios because they indicate similar risk behaviour, and may thus considerably simplify risk assessment, also for intermediate, more realistic and complex contaminant release and aquifer heterogeneity cases.
H23B-1304
Uncertainty in Climatology-Based Estimates of Shallow Groundwater Recharge
The groundwater recharge (GR) estimates for flow and transport projections are often evaluated as a fixed percentage of average annual precipitation. The chemical transport in variably saturated heterogeneous porous media is not linearly related to the average velocity. The objective of this study was to estimate the variability in annual, seasonal, and event-based GR at the field scale and to quantify sources of the uncertainty in such estimates. Research was done at a 3.6 ha field, which is part of a 21-ha agricultural research site located at the USDA, Beltsville Agricultural Research Center, Maryland. The soils are sandy with buried clay lenses. Eight soil moisture multi-sensor capacitance probes (MCP) monitor water contents in soil layer 0-180 cm at 10 minute intervals. Surface runoff water is measured at the outlet of the field with a flow meter installed on a 45.7 cm H- flume. An energy balance weather station inside the field is the source of detailed meteorological data. A standard weather station is located in 4 miles. The water budget method was used to compute the GWR from precipitations, runoff, ET data and soil water content measurements at depths of 10, 30, 50, and 80 cm. ET and precipitation were assumed uniform across the field. The event-based estimates of GWR were defined as the infiltration water losses below depth of 80 cm at time interval between two consecutive local minima on soil water storage series. The average estimated percentage of rainfall used for recharge for a single recharge event was 56%, 40%, 28%, 11% and 11% for rainfalls 0 to 10 mm, 10 to 20 mm, 20 to 30 mm, 30 to 60 mm, and larger than 60 mm, respectively. Estimated average recharge per single event differed by seasons and constituted 21%, 43%, 37% and 44% of precipitation in spring, summer, fall, and winter respectively. Total annual amount of rainfall, estimated ET and runoff over the 365 days of the observation period was 909 mm, 386 mm, and 276 mm, respectively. The recharge amount estimated from the water budget for this period was from 175 mm to 220 mm and constituted from 20% to 25% of precipitation. No statistically significant dependence of the estimated recharge on the soil water content in the beginning of the recharge event was found. We present data on and discuss the sources of uncertainty of the above estimates, i.e. plant water status, biomass, and yield; runoff contributing area; areas represented by individual MCP; errors in Pennman-Monteith daily ET estimates; lateral subsurface water pathways and subsurface run-on and run-off in the vicinity of sensors.
H23B-1305
A Proposed Objective-Oriented Groundwater Model for Conjunctive Use Planning of Surface and Groundwater
Groundwater models are used for planning and management of water resources systems. One of the most difficult problems in groundwater modeling is the assessment of model uncertainty. Up till today, this problem has not been completely resolved. The main reasons are that the real geological structure of an aquifer is complex and unknown, and the available data for model calibration are usually limited in both quantity and quality. It is well understood that a more complex model with more parameters may fit the existing data better, but it may cause over-parameterization. It is also well established that over-parameterization renders high uncertainty in model application. An objective-oriented model is constructed based on the reliability requirements in model applications as opposed to the traditional data-driven modeling approach. Specifically, an objective-oriented model seeks to identify a reliable model that is the simplest in model structure. A "reliable model" means that if the model is used for prediction and management it can produce reliable results for a specified model application, and "simple" means that the model has a structure as simple as possible. This study briefly reviews the theoretical basis of constructing an objective-oriented groundwater model. Afterwards, an objective-oriented model is constructed for a conjunctive-use project of imported surface water and groundwater in the desert area, Warren Basin in Southern California. Model fitting residual, model application objectives, and model structure error of replacing a complex model with a simpler model for the Warren Basin are formulated. The formulated problem is solved by a global-local optimization scheme. The sufficiency of the existing data is judged by solving a generalized inverse problem. When the existing data are insufficient for constructing a reliable model for the specified model application objectives, a robust and cost-effective field experiment is designed for collecting the necessary data to make the calibrated model reliable for the stipulated model application objectives.
H23B-1306
Numerical Assessment of Indoor Air Exposure Risk from Subsurface NAPL Contamination under Hydrologic Uncertainties
Understanding the risk of indoor air exposure to residual contaminants in the subsurface following the redevelopment of contaminated land redevelopment project is a central issue at many brownfield sites. In this study, we examine various mechanisms controlling vapor phase intrusion into the indoor air of a typical residential dwelling from a NAPL source located below the water table, and consequently assess the indoor air exposure risk under multiple hydrologic uncertainties. For this purpose, a multi-phase multi-component numerical model, CompFlow Bio is used to simulate the evolution of a TCE source zone and dissolved plume in a variably saturated heterogeneous aquifer, along with the transport of dissolved TCE upwards through the capillary fringe with subsequent migration of TCE vapors in the vadose zone subject to barometric pressure fluctuations. The TCE vapors then enter the basement of the residential dwelling through a crack in the foundation slab, driven by a slight vacuum within the basement relative to the ambient atmosphere as well as the barometric pressure fluctuations. Hydrologic uncertainties affecting the indoor air concentration of TCE include the vacuum in the basement, the aperture of the crack in the foundation slab, the heterogeneous permeability field, the thickness of the capillary fringe, barometric fluctuations, recharge rates and the location of the TCE source zone. CompFlow Bio is then used to determine the future concentration of TCE into the basement as a consequence of imperfect knowledge in the various hydrologic parameters, and to evaluate the effectiveness of alternative remedial and foundation design options to minimize the exposure risk to the indoor air conditional upon the available data collected at the site. The outcome of this approach is two-fold. First, the owner of the site can reasonably evaluate the future indoor air exposure risk following the redevelopment of a formerly contaminated site following remediation activities. Second, the owner/developer of the site can then effectively price the risk that some residential dwellings will exceed an indoor air exposure regulatory limit assuming that the owner/developer of the site takes immediate actions to repurchase the residential dwelling where the regulatory limit was exceeded. Consequently, appropriate financial products (such as insurance or building a contingency fund) can be developed to enable the owner/developer to effectively hedge their risk and ensure solvency.
H23B-1307
Evaluation of a Bias Aware Ensemble Kalman Filter Using a Scaled Aquifer Transport Experiment
In this study, a bias-aware Ensemble Kalman Filter (EnKF) is applied to a scaled aquifer transport experiment (1- cm of the model equals 1-m at field scale and 1-day of model transport equals 1-year of field-scale transport) located at the University of Vermont. The scaled aquifer was constructed using layered porous media within a 2.54-m by 3.56-m by 2.43-m (10-ft by 14-ft by 8-ft) tank. The experimental porous media has been highly characterized and contains 105 sampling locations that are capable of providing concentration data at a high temporal resolution. The scaling of the tank experiment is applicable for advection dominated transport, which has been modeled using the three-dimensional flow-and-transport groundwater models - MODFLOW 2000 and MT3DMS. A 19-day (19-years scaled) ammonia chloride tracer experiment was conducted with concentration data collected at all 105 sampling locations every 17.5-minutes (4.4-days scaled). Results from this research demonstrate how a known bias in the flow-and-transport initial conditions relative to the true experimental conditions can dramatically degrade the EnKF framework's ability to make forward transport forecasts that capture observed breakthrough time series. The bias-aware extension of the EnKF significantly improved the reliability of concentration breakthrough forecasts. In the long term, this research is being used to develop new simulation-optimization frameworks for designing groundwater observation networks. In support of this long-term objective, this study explores the influence of model bias on both the mean and covariance projections provided by EnKF. Moreover, this work explores how the increased computational demands associated with the bias-aware EnKF can be reduced using minimal ensemble sizes and management period formulations. This work advances beyond the commonly employed static Kalman filter formulations employed in prior monitoring design studies (i.e., spatiotemporal kriging) by making forecasts more robust to nonlinearities while also accounting for measurement uncertainties and dynamic, spatiotemporally correlated model structure errors as well as model parameter errors.
H23B-1308
Effects of Element Simplification on a Ground Water Flow Solution
This project was developed for the purposes of detailing the statistical impact of varying data sources and inclusions for calibration of an analytic element groundwater model. For a regional groundwater model, resolution of hydrogeological features, used as boundary conditions, affects the output of travel time, head, and flux in quantifiably different ways. This hypothesis was evaluated using a synthetic steady-state, unconfined model, based on a real-world system. This synthetic model established a baseline for subsequent comparison models of lower element resolution. Using a GIS algorithm, the model elements were simplified to varying degrees to establish new sets of boundary conditions. Each of the new, simplified models was then calibrated to the flow solution of the synthetic model using observations of head, flux, and travel time. Results indicate that decreasing resolution of stream and lake elements has a greater impact on the flow solution than the simplification of the aquifer's conductivity distribution. Based on comparison of the sum of squared errors, head and flux are similarly affected by element resolution, though the magnitude of the errors is much greater for flux, while travel time is only marginally affected. Using the simplification levels within the GIS interface, head and flux observations presented an initial gradual escalation in errors with a sudden increase beyond level six, while travel time presented a smoother, linear increase. When presented with data and computational resource limitations, this analysis demonstrates that reliable piezometric head data and preservation of surficial feature details within the model domain may provide an adequate flow solution without exhaustive subsurface characterization.
H23B-1309
Comparative Inverse Modeling With Combination of Multiphysics Codes and Parameter Estimation Snap-on Tools
Inverse problems in hydrology and hydrogeology are challenging due to non-uniqueness and/or instability yet ubiquitous in parameter estimation for distributed parameter models. One avenue to reduction of non- uniqueness involves use of multiple data sets governed by the same aquifer property to be estimated; i.e., piezometric head (supposedly governed by a diffusion equation) can be coupled with tracer or inferred age data (supposedly governed by an advection-dispersion-reaction equation). In such a case both head and concentration/age data may be used to help robustness in the inverse problem solution. In the recent years several so-called "multiphysics" simulation tools have been developed which can simultaneously solve coupled sets of PDE-based physical equations representing different processes often at distinct scales. Employing such multiphysics simulation tools lets us couple complex physical phenomena such as fluid flow in subsurface, solute fate and transport, heat transfer etc. Employing a general multiphysics computational framework for specification of the forward model can facilitate handling of multiple data types in indirect inversions to determine particular parameter values common to the coupled models. We report on a preliminary investigation into an indirect inversion of a benchmark case (due to Sudicky, 1989) of steady-state fluid flow and transient solute transport through a vertical cross section in an unconfined aquifer. Solute enters the domain through a finite section of the upper boundary, and undergoes nonuniform transport due to large scale heterogeneity in aquifer properties. We apply multiphysics modeling in the forward sense to solve simultaneously the flow and solute transport equations, the output of which is used as input to an indirect inversion scheme using UCODE_2005 (Poeter et al. 2007) to calibrate some of the parameters involved. Furthermore a multimodel analysis (Poeter and Hill, 2007) was applied to disciminate between the different models developed. We use comparative simulations to evaluate some of the aspects of the multiphysics computing framework in this context and report on the capabilities of the different tools used in this demonstration problem and their computational performances under different grid resolutions and zonation strategies.
H23B-1310
Confidence Region Estimation for Groundwater Parameter Identification Problems
This presentation focuses on different methods to generate confidence regions for nonlinear parameter identification problems. Three methods for confidence region estimation are considered: a linear approximation method, an F--test method, and a Log--Likelihood method. Each of these methods are applied to three case studies. One case study is a problem with synthetic data, and the other two case studies identify hydraulic parameters in groundwater flow problems based on experimental well--test results. The confidence regions for each case study are analyzed and compared. Each of the three methods produce similar and reasonable confidence regions for the case study using synthetic data. The linear approximation method grossly overestimates the confidence region for the first groundwater parameter identification case study. The F--test and Log--Likelihood methods result in similar reasonable regions for this test case. For the second groundwater parameter identification case study, the linear approximation method produces a confidence region of reasonable size. In this test case, the F--test and Log--Likelihood methods generate disjoint confidence regions of reasonable size. The differing results, capabilities, and drawbacks of all three methods are discussed. Sandia is a multi program laboratory operated by Sandia Corporation, a Lockheed Martin Company, for the United States Department of Energy's National Nuclear Security Administration under Contract DE-AC04- 94AL85000. This research is funded by WIPP programs administered by the Office of Environmental Management (EM) of the U.S Department of Energy.
H23B-1311
On the significance of contaminant plume-scale and dose-response models in defining hydrogeological characterization needs
Defining rational and effective hydrogeological data acquisition strategies is of crucial importance since financial resources available for such efforts are always limited. Usually such strategies are developed with the goal of reducing uncertainty, but less often they are developed in the context of the impacts of uncertainty. This paper presents an approach for determining site characterization needs based on human health risk factors. The main challenge is in striking a balance between improved definition of hydrogeological, behavioral and physiological parameters. Striking this balance can provide clear guidance on setting priorities for data acquisition and for better estimating adverse health effects in humans. This paper addresses this challenge through theoretical developments and numerical testing. We will report on a wide range of factors that affect the site characterization needs including contaminant plume's dimensions, travel distances and other length scales that characterize the transport problem, as well as health risk models. We introduce a new graphical tool that allows one to investigate the relative impact of hydrogeological and physiological parameters in risk. Results show that the impact of uncertainty reduction in the risk-related parameters decreases with increasing distances from the contaminant source. Also, results indicate that human health risk becomes less sensitive to hydrogeological measurements when dealing with ergodic plumes. This indicates that under ergodic conditions, uncertainty reduction in human health risk may benefit from better understanding of the physiological component as opposed to a detailed hydrogeological characterization
H23B-1312
Estimating Different Regimes in a Tracer Breakthrough Curve with Bayesian Statistics.
Formally, all experimental scientists try to keep the conditions of the experiment in steady state or as close to steady state as possible in order to keep time, as a variable, out of the modeling picture. In practice, though, sometimes experiments do not go as wished, and the mathematical modelling of the data cannot ignore temporal changes in some of the physical parameters of the experiment. The authors have been working with a data set produced in an early experiment conducted at the Waste Isolation Pilot Plant site [Gonzales, 1984], and the presence of different physical conditions was admitted already in the original paper that presented those results. We propose using Bayesian statistics and the Reversible Jump Markov Chain Monte Carlo (RJMCMC) [Green, 1995] algorithm to find the number and duration of regimes that were present in the experiment. Our hypothesis is that for reasons particular to the experiment, the pumping in the extraction well diverged from a constant regime sometime during the experiment. We use a very simple 1-dimensional transport model to explain the breakthrough curve for steady-state conditions, and the number and duration of different regimes is included as a free parameter to be inferred from the data. RJMCMC simulation provides an approximation to posterior probability distributions for the number of change-points, time of ocurrence of these changes, and also probability distributions for the physical parameters that characterize each regime. We will also show that this problem can be seen as a variable-selection problem, and that the method can be readily applied to other situations where variable selection is an ambition of the modelers. References: Gonzales D, Bentley C (1984). Field test for effective porosity and dispersivity in fractured dolomite: the WIPP, Southeastern New Mexico. In Groundwater Hydraulics, Rosenshein JS, Bennett GD (editors), Washington DC: American Geophysical Union, 207–221. Green P (1995). Reversible jump Markov chain Monte Carlo computation and Bayesian model determination. Biometrika, 82, 711–732.
H23B-1313
Prediction Uncertainty Analyses for the Combined Physically-Based and Data-Driven Models
The unavoidable simplification associated with physically-based mathematical models can result in biased parameter estimates and correlated model calibration errors, which in return affect the accuracy of model predictions and the corresponding uncertainty analyses. In this work, a physically-based groundwater model (MODFLOW) together with error-correcting artificial neural networks (ANN) are used in a complementary fashion to obtain an improved prediction (i.e. prediction with reduced bias and error correlation). The associated prediction uncertainty of the coupled MODFLOW-ANN model is then assessed using three alternative methods. The first method estimates the combined model confidence and prediction intervals using first-order least- squares regression approximation theory. The second method uses Monte Carlo and bootstrap techniques for MODFLOW and ANN, respectively, to construct the combined model confidence and prediction intervals. The third method relies on a Bayesian approach that uses analytical or Monte Carlo methods to derive the intervals. The performance of these approaches is compared with Generalized Likelihood Uncertainty Estimation (GLUE) and Calibration-Constrained Monte Carlo (CCMC) intervals of the MODFLOW predictions alone. The results are demonstrated for a hypothetical case study developed based on a phytoremediation site at the Argonne National Laboratory. This case study comprises structural, parameter, and measurement uncertainties. The preliminary results indicate that the proposed three approaches yield comparable confidence and prediction intervals, thus making the computationally efficient first-order least-squares regression approach attractive for estimating the coupled model uncertainty. These results will be compared with GLUE and CCMC results.
H23B-1314
Accounting for Transport Parameter Uncertainty in Geostatistical Groundwater Contaminant Release History Estimation
The process of estimating the release history of a contaminant in groundwater relies on coupling a limited number of concentration measurements with a groundwater flow and transport model in an inverse modeling framework. The information provided by available measurements is generally not sufficient to fully characterize the unknown release history; therefore, an accurate assessment of the estimation uncertainty is required. The modeler's level of confidence in the transport parameters, expressed as pdfs, can be incorporated into the inverse model to improve the accuracy of the release estimates. In this work, geostatistical inverse modeling is used in conjunction with Monte Carlo sampling of transport parameters to estimate groundwater contaminant release histories. Concentration non-negativity is enforced using a Gibbs sampling algorithm based on a truncated normal distribution. The method is applied to two one-dimensional test cases: a hypothetical dataset commonly used in validating contaminant source identification methods, and data collected from a tetrachloroethylene and trichloroethylene plume at the Dover Air Force Base in Delaware. The estimated release histories and associated uncertainties are compared to results from a geostatistical inverse model where uncertainty in transport parameters is ignored. Results show that the a posteriori uncertainty associated with the model that accounts for parameter uncertainty is higher, but that this model provides a more realistic representation of the release history based on available data. This modified inverse modeling technique has many applications, including assignment of liability in groundwater contamination cases, characterization of groundwater contamination, and model calibration.
H23B-1315
A geostatistical inverse modeling method for simultaneous estimation of hydraulic conductivity and contamination history
In recent years, various geostatistical inverse modeling methods have been developed for recovering the release history of a contaminant into an aquifer. A major limitation of most of these methods is the requirement to have a fully-known transport model, which usually requires, in turn, knowledge of the aquifer's hydraulic conductivity field. The hydraulic conductivity is never fully characterized in practical applications, however, and its estimation is often the subject of inverse-modeling investigations in hydrogeology. A method based on quasi-linear geostatistical inverse modeling is developed to simultaneously estimate the contamination history of a pollutant into an aquifer and transport parameters such as the aquifer's hydraulic conductivity field. The proposed iterative approach uses point measurements of hydraulic head, contaminant concentration, and, if available, hydraulic conductivity as constraints. The novelty of this method is in the joint estimation of the historic contamination and the transport parameters (e.g. hydraulic conductivity). This approach allows for a more accurate assessment of the uncertainties associated with the final estimates, which reflect the cross-covariance between the two parameters fields, as well as model errors, measurement errors, and aleatory variability inherent to the physical system. Simulations carried out on a two-dimensional hypothetical aquifer demonstrate the applicability of this method under various conditions. Unlike existing geostatistical inverse modeling methods, the present method allows the release history to be retrieved even if transport parameters are not fully characterized. However, these experiments indicate that a relatively large number of measurements is required in heterogeneous applications to constrain the unknown contamination history. In relatively homogeneous aquifers, a smaller number of measurements is sufficient to jointly characterize the conductivity field and contamination history. The approach is applicable to site characterization for remediation design, as well as the identification of responsible parties in cases of groundwater contamination. Additionally, this method can form the basis for approaches aimed at characterizing the current or future distribution of contaminants in aquifers, given uncertainty in the flow and transport parameters, as well as in the contamination history.
H23B-1316
Uncertainty Assessment of Soil Hydraulic Parameter Estimated From Cokriging and Artificial Neural Network
To characterize heterogeneity of soil hydraulic parameters, we developed a method to incorporate measurements of soil texture, bulk density, soil hydraulic parameter, and moisture content. The method first uses cokriging to estimate heterogeneous soil texture and bulk density (pedotransfer variables) using the moisture content as secondary variables that are abundant at the site. The heterogeneous pedotransfer variables are then used as input to an artificial neural network (ANN) developed to estimate heterogeneous soil hydraulic parameters, which are then used to simulate a field injection experiment at the U.S. Department of Energy's Hanford Site, WA. Although a large amount of uncertainty exists in the cokriging and ANN estimates, only the mean estimates are used for the numerical simulation. It is unknown to what extent the estimation uncertainty of cokriging and ANN affects the simulated injection experiment, and hence the motivation for this study. The uncertainty of parameter estimates is investigated in two steps. First, uncertainty of the ANN estimates is quantified using mean and variance of the estimates obtained from a bootstrap method. A total of 10,000 bootstraps are generated to obtain a statistically meaningful variance. The bootstrap estimates are Gaussian. A Latin Hypercube Sampling (LHS) method is used to generate 10 realizations from the entire distribution. A Monte Carlo simulation is conducted to evaluate propagation of uncertainty of the ANN estimates via numerical simulation of the field injection experiment. In the second step, uncertainty of cokriging estimates is assessed in a similar manner. Based on cokriging mean and variance and the assumed Gaussian distribution of cokriging estimates, 10 realizations of the pedotransfer variables are generated using the LHS method. For each of the realization, 10 realizations of soil hydraulic parameters are generated by the ANN. As a result, a total 100 realizations of the soil hydraulic parameters are generated for a Monte Carlo simulation. Uncertainty of the simulated injection experiment is evaluated next. Results indicate that the uncertainty of the cokriging estimates affects the injection experiment simulations considerably more than uncertainty of the ANN estimates.
H23B-1317
A Markov Chain Model for Characterizing Vadose Zone Soil Heterogeneity and Layering Structure
Characterization of media heterogeneity in unsaturated media is always hampered by lack of site-specific soil samples. In particular, geologic heterogeneity, which controls unsaturated flow and solute transport, cannot be identified based on samples from one or multiple boreholes. This study develops a geologically based Markov chain model to characterize vadose zone soil heterogeneity and layering structure. Soil textural classes (categorical variables) are first categorized, and spatial variability of the soil classes is measured by transition probability from one class to another in space. The Markov chain model is developed based on the transition probability. Although the model in the vertical direction can be developed based on borehole samples, developing the model at the horizontal direction always needs geological information (e.g., mean length and juxtaposition tendencies of the soil classes) obtained from soil (or outcrop) surveys. The geological information is critical to characterization of heterogeneity and soil layering structure. The method is applied to a field injection experiment at the Sisson and Lu injection site at the U.S. Department of Energy Hanford Site, WA. Four soil classes (coarse sand, sand, loamy sand, and sandy loam) are categorized based on particle size distribution of 93 samples from 6 boreholes at the site. Among the 93 samples, 60 samples are collected from 3 boreholes closely spaced within a maximum distance of 2.17 m. The vertical Markov chain model is developed based on the 93 samples. The horizontal model, however, cannot be developed due to a close spacing of the 3 boreholes providing majority of the samples. Initial moisture content measurements at 1334 locations (from 32 boreholes and at 1-foot vertical interval) are used to develop the horizontal Markov chain model. The measurements at the site are transformed into soil textural classes based on the field observation that higher and lower measurement values correspond to fine- and coarse-textured media, respectively. Volumetric portions of the soil classes are used as thresholds of the transform. The Markov chain models developed are then used to represent the spatial structure of the soil classes to characterize soil heterogeneity. Multiple realizations are generated to represent uncertainty of the spatial structure. Soil layering structure is well represented because of the conditioning of the soil classes, as transformed from the initial moisture contents. Soil hydraulic parameters for each soil class are determined from core samples of the soil class. Next, the field injection experiment is simulated; the simulated moisture contents compare well with the field observations for both infiltration and redistribution periods. In particular, in comparison with our previous modeling at the site, the vertical movement of the injected water is improved because of a more accurate characterization of layering structure at the site. Effects of uncertainty of spatial structure of the soil classes on prediction of unsaturated flow at the site are also investigated. Uncertainty of the soil hydraulic parameters within each soil class will be explored in a future study.
H23B-1318
An efficient, high-order multi-element probabilistic collocation method on sparse grids for three-dimensional flow in random porous media
In this study, we use a multi-element probabilistic collocation method (ME- PCM) on sparse grids to obtain high-order solutions of the mean and standard deviation of hydraulic head and concentration for three-dimensional saturated flow and transport in randomly heterogeneous porous media. First, Karhunen-Loeve (K-L) decomposition is applied to represent the log hydraulic conductivity Y = lnK. The hydraulic head h and Darcy flux q are obtained by solving the three-dimensional continuity equation coupled with Darcy's law with random hydraulic conductivity field. The concentration is computed by solving the three- dimensional stochastic advection-dispersion-reaction equation with random Darcy flux q. ME-PCM is an extension of multi-element generalized polynomial chaos (ME-gPC). ME-PCM couples ME-gPC with probabilistic collocation. By using the sparse grid points, ME-PCM can handle random process with large number of random dimensions with relatively lower computational cost, compared to full tensor products. Monte Carlo (MC) simulations were conducted to verify the ME-PCM solution. By comparing the MC and ME-PCM results, it is evident that the ME-PCM approach is computationally more efficient than Monte Carlo simulations. Unlike the conventional moment-equation approach, there is no limitation on the amplitude of random perturbations for ME-PCM. Furthermore, ME-PCM on sparse grids can efficiently simulate flow and transport in randomly heterogeneous porous media with small correlation lengths. The effect of correlation lengths and variance of hydraulic conductivity on flow and transport has also been investigated.
H23B-1319
Stochastic Simulation of the Effects of Physical and Chemical Heterogeneities on Solute Transport in an Alluvial Aquifer
The valley-fill alluvial aquifer of Fortymile Wash is conceptualized as a natural barrier to radionuclide transport from a potential high-level nuclear waste repository in Yucca Mountain, Nevada. The valley-fill alluvial aquifer is being represented as a hydrogeologically homogeneous unit in site-scale saturated zone models. Geological and geophysical surveys conducted in the area indicate strong spatial variability of facies distributions, leading to uncertainty about flow and transport processes within the alluvium, and particularly the impact of preferential flow paths on site-scale radionuclide transport. The main purpose of this work is to evaluate the effectiveness of the alluvium in isolating radioactive waste from the accessible environment, by considering the spatial variability of the alluvium structure. A facies-driven approach is adopted to simulate both physical and chemical heterogeneities in the alluvium. Based on borehole cutting log analyses, outcrop studies, and sedimentological scaling relationships, a multiscale, hierarchical hydrofacies model is developed where the main hydrofacies are classified corresponding to different scales of bedforms and deposits, and to the contrasts in permeability. A transition probability/Markov chain geostatistical approach is used to generate realistic replicas of subsurface facies distributions, which in turn, are translated into permeability distributions using site-specific values. A set of normalized Neptunium (Np) sorption parameters pertaining to the Fortymile Wash alluvial aquifer are analyzed using geostatistical methods, and the results are combined with surface area distributions to enable simulation of Np parameter distributions. Monte Carlo simulations of non-reactive and reactive transport are performed using fine-resolution models that have dimensions on the order of a block in the site-scale models. Flow-based upscaling indicates that the magnitudes of mean block hydraulic conductivities are close to the hydraulic conductivity values assigned to the valley-fill alluvial unit in site-scale models. The results of reactive transport modeling indicate at most a two-fold increase in solute spread in the longitudinal direction. Additional numerical tests are conducted to quantify the moments of breakthrough curves. This abstract is an independent product of the CNWRA and does not necessarily reflect the view or regulatory position of NRC.
H23B-1320
A Strategy For Flow And Transport Model Development Using A Suite Of Models With Varying Levels Of Complexity To Support Conceptual And Parameter Uncertainty Analysis
Contaminant transport calculations in a regulatory environment often require extremely complex representations of the subsurface and solution of coupled differential equations describing flow and transport throughout the spatial domain of interest. These models have become so increasingly complex, that the use of standard calibration and uncertainty estimation methodologies is very difficult, if not impossible. Here, we present a framework for using simple models to guide the development and calibration of more complex flow and transport models. We demonstrate the usefulness and applicability of this approach using groundwater flow below Yucca Flat at the Nevada Test Site, where the majority of U.S. underground nuclear tests were conducted. Underground testing had a profound impact on pore-water pressures; simulating the spatial and temporal manifestations of these pressure changes is at the core of this analysis. To accomplish this, we first construct simple models, analytical and 2-D cross-section models, to test multiple alternative conceptual models and to assess parameter uncertainty. The analytical models perform remarkably well in explaining most of the variance in the measured heads. Sensitivity analysis was used to significantly reduce the large number of uncertain parameters. Then, we apply this result to the calibration of a fully integrated 3-D numerical flow and transport model. Similarly, we show application of a 2-D cross-section model to assist in the evaluation of multiple alternative conceptual models, and how these results a) highlight conceptual uncertainty and b) can be incorporated efficiently into the more complex 3-D flow problem that then supports predictive analysis and sensitivity evaluation. The overall process demonstrates the value of using a suite of modeling tools in the early stages of flow and transport development.
H23B-1321 [WITHDRAWN]
An Efficient Importance Sampling Approach for Simulating Contaminant Transport in Heterogeneous Porous Media
In many applications, one may be interested in estimating the probability of a released solute reaching a particular target region. The conventional Monte Carlo method can be used for such a purpose. However, if the probability to be estimated is relatively small, an extreme large number of realizations (or particles) may be required to obtain a reasonably accurate estimate. In this study, we introduce a new approach, importance sampling Monte Carlo simulation, to efficiently estimate such a small probability. In the conventional Monte Carlo simulations, the probability of interest is derived from an ensemble of all possible trajectories that are attributed to heterogeneity of the hydraulic conductivity field. In the importance sampling approach, such trajectories are taken from a modified ensemble so that more solute particles will reach the target region. We do so by adding an artificial spatially-varying velocity field to the true velocity field. Since the samples are taken from a biased ensemble, the outputs from simulations are then weighted in such a way that the bias introduced by sampling from the modified ensemble will be exactly corrected. The general procedure of this importance sampling approach as well as its applicability to subsurface transport problems has been illustrated using a simple example for which analytical solution is available. The comparison of results from the analytical solution, the conventional Monte Carlo simulations, and importance sampling approach demonstrates that the latter is computational much more efficient than the conventional Monte Carlo method, especially when the probability of interest in very small.