Parameter-independent model reduction of transient groundwater flow models: Application to inverse problems

A new methodology is proposed for the development of parameter-independent reduced models for transient groundwater flow models. The model reduction technique is based on Galerkin projection of a highly discretized model onto a subspace spanned by a small number of optimally chosen basis functions. We propose two greedy algorithms that iteratively select optimal parameter sets and snapshot times between the parameter space and the time domain in order to generate snapshots. The snapshots are used to build the Galerkin projection matrix, which covers the entire parameter space in the full model. We then apply the reduced subspace model to solve two inverse problems: a deterministic inverse problem and a Bayesian inverse problem with a Markov Chain Monte Carlo (MCMC) method. The proposed methodology is validated with a conceptual one-dimensional groundwater flow model. We then apply the methodology to a basin-scale, conceptual aquifer in the Oristano plain of Sardinia, Italy. Using the methodology, the full model governed by 29,197 ordinary differential equations is reduced by two to three orders of magnitude, resulting in a drastic reduction in computational requirements.     

Polynomial chaos expansion for global sensitivity analysis applied to a model of radionuclide migration in a randomly heterogeneous aquifer

We perform global sensitivity analysis (GSA) through polynomial chaos expansion (PCE) on a contaminant transport model for the assessment of radionuclide concentration at a given control location in a heterogeneous aquifer, following a release from a near surface repository of radioactive waste. The aquifer hydraulic conductivity is modeled as a stationary stochastic process in space. We examine the uncertainty in the first two (ensemble) moments of the peak concentration, as a consequence of incomplete knowledge of (a) the parameters characterizing the variogram of hydraulic conductivity, (b) the partition coefficient associated with the migrating radionuclide, and (c) dispersivity parameters at the scale of interest. These quantities are treated as random variables and a variance-based GSA is performed in a numerical Monte Carlo framework. This entails solving groundwater flow and transport processes within an ensemble of hydraulic conductivity realizations generated upon sampling the space of the considered random variables. The Sobol indices are adopted as sensitivity measures to provide an estimate of the role of uncertain parameters on the (ensemble) target moments. Calculation of the indices is performed by employing PCE as a surrogate model of the migration process to reduce the computational burden. We show that the proposed methodology (a) allows identifying the influence of uncertain parameters on key statistical moments of the peak concentration (b) enables extending the number of Monte Carlo iterations to attain convergence of the (ensemble) target moments, and (c) leads to considerable saving of computational time while keeping acceptable accuracy.     

A comparative study of numerical approaches to risk assessment of contaminant transport

In risk analysis, a complete characterization of the concentration distribution is necessary to determine the probability of exceeding a threshold value. The most popular method for predicting concentration distribution is Monte Carlo simulation, which samples the cumulative distribution function with a large number of repeated operations. In this paper, we first review three most commonly used Monte Carlo (MC) techniques: the standard Monte Carlo, Latin Hypercube sampling, and Quasi Monte Carlo. The performance of these three MC approaches is investigated. We then apply stochastic collocation method (SCM) to risk assessment. Unlike the MC simulations, the SCM does not require a large number of simulations of flow and solute equations. In particular, the sparse grid collocation method and probabilistic collocation method are employed to represent the concentration in terms of polynomials and unknown coefficients. The sparse grid collocation method takes advantage of Lagrange interpolation polynomials while the probabilistic collocation method relies on polynomials chaos expansions. In both methods, the stochastic equations are reduced to a system of decoupled equations, which can be solved with existing solvers and whose results are used to obtain the expansion coefficients. Then the cumulative distribution function is obtained by sampling the approximate polynomials. Our synthetic examples show that among the MC methods, the Quasi Monte Carlo gives the smallest variance for the predicted threshold probability due to its superior convergence property and that the stochastic collocation method is an accurate and efficient alternative to MC simulations.     

Characterization of groundwater contaminant source using Bayesian method

Contaminant source identification in groundwater system is critical for remediation strategy implementation, including gathering further samples and analysis, as well as implementing and evaluating different remediation plans. Such problem is usually solved with the aid of groundwater modeling with lots of uncertainty, e.g. existing uncertainty in hydraulic conductivity, measurement variance and the model structure error. Monte Carlo simulation of flow model allows the input uncertainty onto the model predictions of concentration measurements at monitoring sites. Bayesian approach provides the advantage to update estimation. This paper presents an application of a dynamic framework coupling with a three dimensional groundwater modeling scheme in contamination source identification of groundwater. Markov Chain Monte Carlo (MCMC) is being applied to infer the possible location and magnitude of contamination source. Uncertainty existing in heterogonous hydraulic conductivity field is explicitly considered in evaluating the likelihood function. Unlike other inverse-problem approaches to provide single but maybe untrue solution, the MCMC algorithm provides probability distributions over estimated parameters. Results from this algorithm offer a probabilistic inference of the location and concentration of released contamination. The convergence analysis of MCMC reveals the effectiveness of the proposed algorithm. Further investigation to extend this study is also discussed.     

Optimal design of measurement networks for groundwater flow predictions

This paper presents a methodology to optimise measurement networks for the prediction of groundwater flow. Two different strategies are followed: the design of a measurement network that aims at minimizing the log-transmissivity variance (averaged over the domain of interest) or a design that minimises the hydraulic head variance (averaged over the domain of interest). The methodology consists of three steps. In the first step the prior log-transmissivity and hydraulic head variances are estimated. This step is completely general in the sense that the prior variances maybe unconditional, or maybe conditioned to log-transmissivity and/or hydraulic head measurements. In case hydraulic head measurements are available in the first step, the inverse groundwater flow problem is solved by the sequential self-calibrated method. In the second step, the full covariance matrices of hydraulic head and log-transmissivity are calculated numerically on the basis of a sufficiently big number of Monte Carlo realisations. On the basis of the estimated covariances, the impact of an additional measurement in terms of variance reduction is calculated. The measurement that yields the maximum domain averaged variance reduction is selected. Additional measurement locations are selected according to the same procedure.The procedure has been tested for a series of synthetic reference cases. Different sampling designs are tested for each of these cases, and the proposed strategies are compared with other sampling strategies. Although the proposed strategies indeed reach their objective and yield in most cases the lowest posterior log-transmissivity variance or hydraulic head variance, the differences as compared to alternative sampling strategies are frequently small. For the cases considered here, a sampling design that covers more or less regularly the aquifer performs well.The paper also illustrates that for the optimal estimation of a well catchment a heuristic criterion (spreading measurement points as regularly as possible over the zone where there is some uncertainty regarding the capture probability) yields better results than a sampling design that minimises the posterior log-transmissivity variance or posterior hydraulic head variance.     

Distributed Parallel Computing in Stochastic Modeling of Groundwater Systems

Stochastic modeling is a rapidly evolving, popular approach to the study of the uncertainty and heterogeneity of groundwater systems. However, the use of Monte Carlo‐type simulations to solve practical groundwater problems often encounters computational bottlenecks that hinder the acquisition of meaningful results. To improve the computational efficiency, a system that combines stochastic model generation with MODFLOW‐related programs and distributed parallel processing is investigated. The distributed computing framework, called the Java Parallel Processing Framework, is integrated into the system to allow the batch processing of stochastic models in distributed and parallel systems. As an example, the system is applied to the stochastic delineation of well capture zones in the Pinggu Basin in Beijing. Through the use of 50 processing threads on a cluster with 10 multicore nodes, the execution times of 500 realizations are reduced to 3% compared with those of a serial execution. Through this application, the system demonstrates its potential in solving difficult computational problems in practical stochastic modeling.     

Delineating a recharge area for a spring using numerical modeling, Monte Carlo techniques, and geochemical investigation

Recharge areas of spring systems can be hard to identify, but they can be critically important for protection of a spring resource. A recharge area for a spring complex in southern Wisconsin was delineated using a variety of complementary techniques. A telescopic mesh refinement (TMR) model was constructed from an existing regional-scale ground water flow model. This TMR model was formally optimized using parameter estimation techniques; the optimized "best fit" to measured heads and fluxes was obtained by using a horizontal hydraulic conductivity 200% larger than the original regional model for the upper bedrock aquifer and 80% smaller for the lower bedrock aquifer. The uncertainty in hydraulic conductivity was formally considered using a stochastic Monte Carlo approach. Two-hundred model runs used uniformly distributed, randomly sampled, horizontal hydraulic conductivity values within the range given by the TMR optimized values and the previously constructed regional model. A probability distribution of particles captured by the spring, or a "probabilistic capture zone," was calculated from the realistic Monte Carlo results (136 runs of 200). In addition to portions of the local surface watershed, the capture zone encompassed areas outside of the watershed--demonstrating that the ground watershed and surface watershed do not coincide. Analysis of water collected from the site identified relatively large contrasts in chemistry, even for springs within 15 m of one another. The differences showed a distinct gradation from Ordovician-carbonate-dominated water in western spring vents to Cambrian-sandstone-influenced water in eastern spring vents. The difference in chemistry was attributed to distinctive bedrock geology as demonstrated by overlaying the capture zone derived from numerical modeling over a bedrock geology map for the area. This finding gives additional confidence to the capture zone calculated by modeling.     

Estimating the role of a ditch network in groundwater recharge in a Mediterranean catchment using a water balance approach

Water balance variables were monitored in a farmed Mediterranean catchment characterized by a dense ditch network to allow for the separate estimation of the diffuse and concentrated recharge terms during flood events. The 27 ha central part of the catchment was equipped with (i) rain gauges, (ii) ditch gauge stations, (iii) piezometers, (iv) neutron probes, and (v) an eddy covariance mast including a 3D sonic anemometer and a fast hygrometer. The water balance was calculated for two autumnal rain and flood events. We also estimated the uncertainty of this approach with Monte Carlo simulations. Results show, that although ditch area represents only 6% of the total study area, concentrated recharge appeared to be the main source of groundwater recharge. Indeed, it was 40–50% of the total groundwater recharge for autumnal events, which are the major annual recharge events. This indicate that both, concentrated and diffuse recharge should be taken into account in any hydrological modeling approach for Mediterranean catchments. This also means that, since they collect overland flow that is often largely contaminated by chemicals, ditches may be a place where groundwater contamination is likely to occur. The uncertainty analysis indicates that recharge estimates based on water balance exhibit large uncertainty ranges. Nevertheless, Monte Carlo simulations showed that concentrated recharge was higher than expected based on their area.     

Assessment of uncertainty associated with the estimation of well catchments by moment equations

Non-local stochastic moment equations are used successfully to analyze groundwater flow in randomly heterogeneous media. Here we present a moment equations-based approach to quantify the uncertainty associated with the estimation of well catchments. Our approach is based on the development of a complete second order formalism which allows obtaining the first statistical moments of the trajectories of conservative solute particles advected in a generally non-uniform groundwater flow. Approximate equations of moments of particles’ trajectories are then derived on the basis of a second order expansion in terms of the standard deviation of the aquifer log hydraulic conductivity. Analytical expressions are then obtained for the predictors of locations of mean stagnation points, together with their associated uncertainties. We implement our approach on heterogeneous media in bounded two-dimensional domains, with and without including the effect of conditioning on hydraulic conductivity information. The impact of domain size, boundary conditions, heterogeneity and non-stationarity of hydraulic conductivity on the prediction of a well catchment is explored. The results are compared against Monte Carlo simulations and semi-analytical solutions available in the literature. The methodology is applicable to both infinite and bounded domains and is free of distributional assumptions (and so applies to both Gaussian and non-Gaussian log hydraulic conductivity fields) and formally includes the effect of conditioning on available information.     

Stochastic analysis of estuarine hydraulics

A new methodology is presented for the solution of the stochastic hydraulic equations characterizing steady, one-dimensional estuarine flow. The methodology is predicated on quasi-linearization, perturbation methods, and the finite difference approximation of the stochastic differential operators. Assuming Manning's roughness coefficient is the principal source of uncertainty in the model, stochastic equations are presented for the water depths and flow rates in the estuarine system. Moment equations are developed for the mean and variance of the water depths. The moment equations are compared with the results of Monte Carlo simulation experiments. The results confirm that for any spatial location in the estuary that (1) as the uncertainty in the channel roughness increases, the uncertainty in mean depth increases, and (2) the predicted mean depth will decrease with increasing uncertainty in Manning'sn. The quasi-analytical approach requires significantly less computer time than Monte Carlo simulations and provides explicit     

Time-related capture zones for radial flow in two dimensional randomly heterogeneous media

We consider the effect of randomly heterogeneous hydraulic conductivity on the spatial location of time-related capture zones (isochrones) for a non-reactive tracer in the steady-state radial flow field due to a pumping well in a confined aquifer. A Monte Carlo (MC) procedure is used in conjunction with FFT-based spectral methods. The log hydraulic conductivity field is assumed to be Gaussian and stationary, with isotropic exponential correlation. Various degrees of domain heterogeneity are considered and stability and accuracy of the MC procedure is examined. The location of an isochrone becomes uncertain due to heterogeneity, and it is strongly influenced by hydraulic conductivity variance. The probability that a particle released at a point in the aquifer is pumped by the well within a given time is identified. We propose a new expression for the probabilistic spatial distribution of isochrones, which is formally similar to the analytical solution for a uniform medium and takes into account the effects of heterogeneity.     

A numerical investigation of bedrock groundwater recharge and exfiltration on soil mantled hillslopes

Here we use Richards Equation models of variably saturated soil and bedrock groundwater flow to investigate first-order patterns of the coupling between soil and bedrock flow systems. We utilize a Monte Carlo sensitivity analysis to identify important hillslope parameters controlling bedrock recharge and then model the transient response of bedrock and soil flow to seasonal precipitation. Our results suggest that hillslopes can be divided into three conceptual zones of groundwater interaction, (a) the zone of lateral unsaturated soil moisture accumulation (upper portion of hillslope), (b) the zone of soil saturation and bedrock recharge (middle of hillslope) and (c) the zone of saturated-soil lateral flow and bedrock groundwater exfiltration (bottom of hillslope). Zones of groundwater interaction expand upslope during periods of precipitation and drain downslope during dry periods. The amount of water partitioned to the bedrock groundwater system a can be predicted by the ratio of bedrock to soil saturated hydraulic conductivity across a variety of hillslope configurations. Our modelled processes are qualitatively consistent with observations of shallow subsurface saturation and groundwater fluctuation on hillslopes studied in our two experimental watersheds and support a conceptual model of tightly coupled shallow and deep subsurface circulation where groundwater recharge and discharge continuously stores and releases water from longer residence time storage.     

Flow in unsaturated random porous media,nonlinear numerical analysis and comparison to analytical stochastic models

This work presents a rigorous numerical validation of analytical stochastic models of steady state unsaturated flow in heterogeneous porous media. It also provides a crucial link between stochastic theory based on simplifying assumptions and empirical field and simulation evidence of variably saturated flow in actual or realistic hypothetical heterogeneous porous media. Statistical properties of unsaturated hydraulic conductivity, soil water tension, and soil water flux in heterogeneous soils are investigated through high resolution Monte Carlo simulations of a wide range of steady state flow problems in a quasi-unbounded domain. In agreement with assumptions in analytical stochastic models of unsaturated flow, hydraulic conductivity and soil water tension are found to be lognormally and normally distributed, respectively. In contrast, simulations indicate that in moderate to strong variable conductivity fields, longitudinal flux is highly skewed. Transverse flux distributions are leptokurtic. the moments of the probability distributions obtained from Monte Carlo simulations are compared to modified first-order analytical models. Under moderate to strong heterogeneous soil flux conditions (σ²_y≥1), analytical solutions overestimate variability in soil water tension by up to 40% as soil heterogeneity increases, and underestimate variability of both flux components by up to a factor 5. Theoretically predicted model (cross-)covariance agree well with the numerical sample (cross-)covarianaces. Statistical moments are shown to be consistent with observed physical characteristics of unsaturated flow in heterogeneous soils.©1998 Elsevier Science Limited. All rights reserved     

Analytical investigation of the exact groundwater divide between rivers beyond the Dupuit–Forchheimer approximation

The groundwater divide is a key feature of river basins and significantly influenced by subsurface hydrological processes. For an unconfined aquifer between two parallel rivers or ditches, it has long been defined as the top of the water table based on the Dupuit–Forchheimer approximation. However, the exact groundwater divide is subject to the interface between two local flow systems transporting groundwater to rivers from the infiltration recharge. This study contributes a new analytical model for two-dimensional groundwater flow between rivers of different water levels. The flownet is delineated in the model to identify groundwater flow systems and the exact groundwater divide. Formulas with two dimensionless parameters are derived to determine the distributed hydraulic head, the top of the water table and the groundwater divide. The locations of the groundwater divide and the top of the water table are not the same. The distance between them in horizontal can reach up to 8.9% of the distance between rivers. Numerical verifications indicate that simplifications in the analytical model do not significantly cause misestimates in the location of the groundwater divide. In contrast, the Dupuit–Forchheimer approximation yields an incorrect water table shape. The new analytical model is applied to investigate groundwater divides in the Loess Plateau, China, with a Monte Carlo simulation process taking into account the uncertainties in the parameters.     

Comparison of Ensemble Kalman Filter groundwater-data assimilation methods based on stochastic moment equations and Monte Carlo simulation

Traditional Ensemble Kalman Filter (EnKF) data assimilation requires computationally intensive Monte Carlo (MC) sampling, which suffers from filter inbreeding unless the number of simulations is large. Recently we proposed an alternative EnKF groundwater-data assimilation method that obviates the need for sampling and is free of inbreeding issues. In our new approach, theoretical ensemble moments are approximated directly by solving a system of corresponding stochastic groundwater flow equations. Like MC-based EnKF, our moment equations (ME) approach allows Bayesian updating of system states and parameters in real-time as new data become available. Here we compare the performances and accuracies of the two approaches on two-dimensional transient groundwater flow toward a well pumping water in a synthetic, randomly heterogeneous confined aquifer subject to prescribed head and flux boundary conditions.     

A methodology for quantifying the value of spatial information for dynamic Earth problems

We develop a methodology for assessing the value of information (VOI) from spatial data for groundwater decisions. Two sources of uncertainty are the focus of this VOI methodology: the spatial heterogeneity (how it influences the hydrogeologic response of interest) and the reliability of geophysical data (how they provide information about the spatial heterogeneity). An existing groundwater situation motivates and in turn determines the scope of this research. The objectives of this work are to (1) represent the uncertainty of the dynamic hydrogeologic response due to spatial heterogeneity, (2) provide a quantitative measure for how well a particular information reveals this heterogeneity (the uncertainty of the information) and (3) use both of these to propose a VOI workflow for spatial decisions and spatial data. The uncertainty of the hydraulic response is calculated using many Earth models that are consistent with the a priori geologic information. The information uncertainty is achieved quantitatively through Monte Carlo integration and geostatistical simulation. Two VOI results are calculated which demonstrate that a higher VOI occurs when the geophysical attribute (the data) better discriminates between geological indicators. Although geophysical data can only indirectly measure static properties that may influence the dynamic response, this transferable methodology provides a framework to estimate the value of spatial data given a particular decision scenario.     

Groundwater–surface water interaction in Lake Nasser,Southern Egypt

A cross‐sectional model, based on the two dimensional groundwater flow equation of Edelman, was applied at seven transects distributed over four geological cross sections to estimate groundwater heads and recharge from/or groundwater discharge to Lake Nasser. The lake with a length of 500 km and an average width of 12 km was created over the period 1964–1970, the time for constructing the Aswan High Dam (AHD). The model, constrained by regional‐scale groundwater flow and groundwater head data in the vicinity of the lake, was successfully calibrated to timeseries of piezometeric heads collected at the cross sections in the period 1965–2004. Inverse modeling yielded high values for the horizontal hydraulic conductivity in the range of 6.0 to 31.1 m day^?1 and storage coefficient between 0.01 and 0.40. The results showed the existence of a strong vertical anisotropy of the aquifer. The calibrated horizontal permeability is systematically higher than the vertical permeability (≈1000:1). The calibrated model was used to explore the recharge from/or groundwater discharge to Lake Nasser at the seven transects for a 40‐year period, i.e. from 1965 to 2004. The analysis for the last 20‐year period, 1985–2004, revealed that recharge from Lake Nasser reduced by 37% compared to the estimates for the first 20‐year period, 1965–1984. In the period 1965–2004, seepage of Lake Nasser to the surrounding was estimated at 1.15 × 10⁹ m³ year^?1. This led to a significant rise of the groundwater table. Variance‐based sensitivity and uncertainty analysis on the Edelman results were conducted applying quasi‐Monte Carlo sequences (Latin Hypercube sampling). The maximum standard deviation of the total uncertainty on the groundwater table was 0.88 m at Toshka (west of the lake). The distance from the lake, followed by the storage coefficient and hydraulic conductivity, were identified as the most sensitive parameters. Copyright © 2012 John Wiley & Sons, Ltd.     

A stochastic optimization framework for the restoration of an over-exploited aquifer

ABSTRACT

This study investigates the impact of hydraulic conductivity uncertainty on the sustainable management of the aquifer of Lake Karla, Greece, using the stochastic optimization approach. The lack of surface water resources in combination with the sharp increase in irrigation needs in the basin over the last 30 years have led to an unprecedented degradation of the aquifer. In addition, the lack of data regarding hydraulic conductivity in a heterogeneous aquifer leads to hydrogeologic uncertainty. This uncertainty has to be taken into consideration when developing the optimization procedure in order to achieve the aquifer’s sustainable management. Multiple Monte Carlo realizations of this spatially-distributed parameter are generated and groundwater flow is simulated for each one of them. The main goal of the sustainable management of the ‘depleted’ aquifer of Lake Karla is two-fold: to determine the optimum volume of renewable groundwater that can be extracted, while, at the same time, restoring its water table to a historic high level. A stochastic optimization problem is therefore formulated, based on the application of the optimization method for each of the aquifer’s multiple stochastic realizations in a future period. In order to carry out this stochastic optimization procedure, a modelling system consisting of a series of interlinked models was developed. The results show that the proposed stochastic optimization framework can be a very useful tool for estimating the impact of hydraulic conductivity uncertainty on the management strategies of a depleted aquifer restoration. They also prove that the optimization process is affected more by hydraulic conductivity uncertainty than the simulation process.

Editor Z.W. Kundzewicz; Guest editor S. Weijs