Nonlinear model reduction of unconfined groundwater flow using POD and DEIM

Nonlinear groundwater flow models have the propensity to be overly complex leading to burdensome computational demands. Reduced modeling techniques are used to develop an approximation of the original model that has smaller dimensionality and faster run times. The reduced model proposed is a combination of proper orthogonal decomposition (POD) and the discrete empirical interpolation method (DEIM). Solutions of the full model (snapshots) are collected to represent the physical dynamics of the system and Galerkin projection allows the formulation of a reduced model that lies in a subspace of the full model. Interpolation points are added through DEIM to eliminate the reduced model's dependence on the dimension of the full model. POD is shown to effectively reduce the dimension of the full model and DEIM is shown to speed up the solution by further reducing the dimension of the nonlinear calculations. To show the concept can work for unconfined groundwater flow model, with added nonlinear forcings, one-dimensional and two-dimensional test cases are constructed in MODFLOW-OWHM. POD and DEIM are added to MODFLOW as a modular package. Comparing the POD and the POD-DEIM reduced models, the experimental results indicate similar reduction in dimension size with additional computation speed up for the added interpolation. The hyper-reduction method presented is effective for models that have fine discretization in space and/or time as well as nonlinearities with respect to the state variable. The dual reduction approach ensures that, once constructed, the reduced model can be solved in an equation system that depends only on reduced dimensions.     

Proper orthogonal decomposition reduced model for mass transport in heterogenous media

Numerical models with fine discretization normally demand large computational time and space, which lead to computational burden for state estimations or model parameter inversion calculation. This article presented a reduced implicit finite difference scheme that based on proper orthogonal decomposition (POD) for two-dimensional transient mass transport in heterogeneous media. The reduction of the original full model was achieved by projecting the high-dimension full model to a low-dimension space created by POD bases, and the bases are derived from the snapshots generated from the model solutions of the forward simulations. The POD bases were extracted from the ensemble of snapshots by singular value decomposition. The dimension of the Jacobian matrix was then reduced after Galerkin projection. Thus, the reduced model can accurately reproduce and predict the original model’s transport process with significantly decreased computational time. This scheme is practicable with easy implementation of the partial differential equations. The POD method is illustrated and validated through synthetic cases with various heterogeneous permeability field scenarios. The accuracy and efficiency of the reduced model are determined by the optimal selection of the snapshots and POD bases.     

POD-based Monte Carlo approach for the solution of regional scale groundwater flow driven by randomly distributed recharge

We present a methodology conducive to the application of a Galerkin model order reduction technique, Proper Orthogonal Decomposition (POD), to solve a groundwater flow problem driven by spatially distributed stochastic forcing terms. Typical applications of POD to reducing time-dependent deterministic partial differential equations (PDEs) involve solving the governing PDE at some observation times (termed snapshots), which are then used in the order reduction of the problem. Here, the application of POD to solve the stochastic flow problem relies on selecting the snapshots in the probability space of the random quantity of interest. This allows casting a standard Monte Carlo (MC) solution of the groundwater flow field into a Reduced Order Monte Carlo (ROMC) framework. We explore the robustness of the ROMC methodology by way of a set of numerical examples involving two-dimensional steady-state groundwater flow taking place within an aquifer of uniform hydraulic properties and subject to a randomly distributed recharge. We analyze the impact of (i) the number of snapshots selected from the hydraulic heads probability space, (ii) the associated number of principal components, and (iii) the key geostatistical parameters describing the heterogeneity of the distributed recharge on the performance of the method. We find that our ROMC scheme can improve significantly the computational efficiency of a standard MC framework while keeping the same degree of accuracy in providing the leading statistical moments (i.e. mean and covariance) as well as the sample probability density of the state variable of interest.     

Application of Proper Orthogonal Decomposition (POD) to inverse problems in saturated groundwater flow

We develop a new Proper Orthogonal Decomposition (POD) reduced order model for saturated groundwater flow, and apply that model to an inverse problem for the hydraulic conductivity field. We use sensitivities as the POD basis. We compare the output when the optimizer uses the reduced order model against results obtained with a full PDE based model. The solutions generated using the POD reduced model are comparable in residual norm to the solutions formed using only the full-scale model. The material parameters are similarly comparable. The time to solution when using the reduced model is reduced by at least an order of magnitude, as are the number of calls to the full model.     

Uncertainty analysis and probabilistic segmentation of electrical resistivity images: the 2D inverse problem

In this paper, we present the uncertainty analysis of the 2D electrical tomography inverse problem using model reduction and performing the sampling via an explorative member of the Particle Swarm Optimization family, called the Regressive‐Regressive Particle Swarm Optimization. The procedure begins with a local inversion to find a good resistivity model located in the nonlinear equivalence region of the set of plausible solutions. The dimension of this geophysical model is then reduced using spectral decomposition, and the uncertainty space is explored via Particle Swarm Optimization. Using this approach, we show that it is possible to sample the uncertainty space of the electrical tomography inverse problem. We illustrate this methodology with the application to a synthetic and a real dataset coming from a karstic geological set‐up. By computing the uncertainty of the inverse solution, it is possible to perform the segmentation of the resistivity images issued from inversion. This segmentation is based on the set of equivalent models that have been sampled, and makes it possible to answer geophysical questions in a probabilistic way, performing risk analysis.     

Improved Parameter Resolution with Markov Chain Monte Carlo Simulation of Different Aquifer Tests

Hydraulic testing for aquifer characterization at contaminated sites often includes tests of short duration and of different types, such as slug tests and pumping tests, conducted at different phases of investigation. Tests conducted on a well cluster installed in a single aquifer can be combined in aggregate inverse analysis using an analytical model for groundwater flow near a test well. A genetic algorithm performs parallel search of the parameter space and provides starting parameter values for a Markov chain Monte Carlo simulation to estimate the parameter distribution. This sequence of inverse methods avoids guessing of the initial parameter vector and the often encountered difficult convergence of gradient-based methods and estimates the parameter covariance matrix from a distribution rather than from a single point in the parameter space. Combination of different tests improves the resolution of the estimated aquifer properties and allows an assessment of the uniformity of the aquifer. Estimated parameter correlations and standard deviations are used as relative metrics to distinguish well resolved and poorly resolved parameters. The methodology is demonstrated on example field tests in unconfined and leaky aquifers.     

An Open,Object-Based Framework for Generating Anisotropy in Sedimentary Subsurface Models

The spatial distribution of hydraulic properties in the subsurface controls groundwater flow and solute transport. However, many approaches to modeling these distributions do not produce geologically realistic results and/or do not model the anisotropy of hydraulic conductivity caused by bedding structures in sedimentary deposits. We have developed a flexible object-based package for simulating hydraulic properties in the subsurface—the Hydrogeological Virtual Realities (HyVR) simulation package. This implements a hierarchical modeling framework that takes into account geological rules about stratigraphic bounding surfaces and the geometry of specific sedimentary structures to generate realistic aquifer models, including full hydraulic-conductivity tensors. The HyVR simulation package can create outputs suitable for standard groundwater modeling tools (e.g., MODFLOW), is written in Python, an open-source programming language, and is openly available at an online repository. This paper presents an overview of the underlying modeling principles and computational methods, as well as an example simulation based on the Macrodispersion Experiment site in Columbus, Mississippi. Our simulation package can currently simulate porous media that mimic geological conceptual models in fluvial depositional environments, and that include fine-scale heterogeneity in distributed hydraulic parameter fields. The simulation results allow qualitative geological conceptual models to be converted into digital subsurface models that can be used in quantitative numerical flow-and-transport simulations, with the aim of improving our understanding of the influence of geological realism on groundwater flow and solute transport.     

Parameter estimation of subsurface flow models using iterative regularized ensemble Kalman filter

A new parameter estimation algorithm based on ensemble Kalman filter (EnKF) is developed. The developed algorithm combined with the proposed problem parametrization offers an efficient parameter estimation method that converges using very small ensembles. The inverse problem is formulated as a sequential data integration problem. Gaussian process regression is used to integrate the prior knowledge (static data). The search space is further parameterized using Karhunen–Loève expansion to build a set of basis functions that spans the search space. Optimal weights of the reduced basis functions are estimated by an iterative regularized EnKF algorithm. The filter is converted to an optimization algorithm by using a pseudo time-stepping technique such that the model output matches the time dependent data. The EnKF Kalman gain matrix is regularized using truncated SVD to filter out noisy correlations. Numerical results show that the proposed algorithm is a promising approach for parameter estimation of subsurface flow models.     

Linear geophysical inversion via the discrete cosine pseudo‐inverse: application to potential fields

In this paper, we present a methodology to perform geophysical inversion of large‐scale linear systems via a covariance‐free orthogonal transformation: the discrete cosine transform. The methodology consists of compressing the matrix of the linear system as a digital image and using the interesting properties of orthogonal transformations to define an approximation of the Moore–Penrose pseudo‐inverse. This methodology is also highly scalable since the model reduction achieved by these techniques increases with the number of parameters of the linear system involved due to the high correlation needed for these parameters to accomplish very detailed forward predictions and allows for a very fast computation of the inverse problem solution. We show the application of this methodology to a simple synthetic two‐dimensional gravimetric problem for different dimensionalities and different levels of white Gaussian noise and to a synthetic linear system whose system matrix has been generated via geostatistical simulation to produce a random field with a given spatial correlation. The numerical results show that the discrete cosine transform pseudo‐inverse outperforms the classical least‐squares techniques, mainly in the presence of noise, since the solutions that are obtained are more stable and fit the observed data with the lowest root‐mean‐square error. Besides, we show that model reduction is a very effective way of parameter regularisation when the conditioning of the reduced discrete cosine transform matrix is taken into account. We finally show its application to the inversion of a real gravity profile in the Atacama Desert (north Chile) obtaining very successful results in this non‐linear inverse problem. The methodology presented here has a general character and can be applied to solve any linear and non‐linear inverse problems (through linearisation) arising in technology and, particularly, in geophysics, independently of the geophysical model discretisation and dimensionality. Nevertheless, the results shown in this paper are better in the case of ill‐conditioned inverse problems for which the matrix compression is more efficient. In that sense, a natural extension of this methodology would be its application to the set of normal equations.     

Identification of aquifer parameters by optimization methods under consideration of uncertainty

Abstract

An approach is presented to solve the inverse problem for simultaneous identification of different aquifer parameters under steady-state conditions. The proposed methodology is formulated as a maximum likelihood parameter estimation problem. Gauss-Newton and full Newton algorithms are used for optimization with an adjoint-state method for calculating the complete Hessian matrix. The methodology is applied to a realistic groundwater model and Monte-Carlo analysis is used to check the results.     

数字合成X射线体层成像的小波-伽辽金重建算法

数字合成X射线体层成像技术的重建问题是在有限投影数据条件下的病态重建问题。本文通过分析数字合成X射线体层成像技术的系统模型,获得重建问题的系统方程。在对系统方程进行正则化改造的基础上,提出了一种新的重建算法——自适应小波-伽辽金重建算法。该算法融合了伽辽金方法的计算简洁和小波内在的多尺度特性,更好地适应了待重建图像的求解。仿真实验结果表明,与ART重建算法相比,自适应小波-伽辽金重建算法在保证重建质量前提下能加快收敛,从而大大地节省了计算时间。     

Efficient Calibration/Uncertainty Analysis Using Paired Complex/Surrogate Models

The use of detailed groundwater models to simulate complex environmental processes can be hampered by (1) long run‐times and (2) a penchant for solution convergence problems. Collectively, these can undermine the ability of a modeler to reduce and quantify predictive uncertainty, and therefore limit the use of such detailed models in the decision‐making context. We explain and demonstrate a novel approach to calibration and the exploration of posterior predictive uncertainty, of a complex model, that can overcome these problems in many modelling contexts. The methodology relies on conjunctive use of a simplified surrogate version of the complex model in combination with the complex model itself. The methodology employs gradient‐based subspace analysis and is thus readily adapted for use in highly parameterized contexts. In its most basic form, one or more surrogate models are used for calculation of the partial derivatives that collectively comprise the Jacobian matrix. Meanwhile, testing of parameter upgrades and the making of predictions is done by the original complex model. The methodology is demonstrated using a density‐dependent seawater intrusion model in which the model domain is characterized by a heterogeneous distribution of hydraulic conductivity.     

Is there a superior conceptual groundwater model structure for baseflow simulation?

Previous work has shown that streamflow response during baseflow conditions is a function of storage, but also that this functional relationship varies among seasons and catchments. Traditionally, hydrological models incorporate conceptual groundwater models consisting of linear or non‐linear storage–outflow functions. Identification of the right model structure and model parameterization however is challenging. The aim of this paper is to systematically test different model structures in a set of catchments where different aquifer types govern baseflow generation processes. Nine different two‐parameter conceptual groundwater models are applied with multi‐objective calibration to transform two different groundwater recharge series derived from a soil‐atmosphere‐vegetation transfer model into baseflow separated from streamflow data. The relative performance differences of the model structures allow to systematically improve the understanding of baseflow generation processes and to identify most appropriate model structures for different aquifer types. We found more versatile and more aquifer‐specific optimal model structures and elucidate the role of interflow, flow paths, recharge regimes and partially contributing storages. Aquifer‐specific recommendations of storage models were found for fractured and karstic aquifers, whereas large storage capacities blur the identification of superior model structures for complex and porous aquifers. A model performance matrix is presented, which highlights the joint effects of different recharge inputs, calibration criteria, model structures and aquifer types. The matrix is a guidance to improve groundwater model structures towards their representation of the dominant baseflow generation processes of specific aquifer types. Copyright © 2014 John Wiley & Sons, Ltd.     

Assess hydrological responses to a warming climate at the Lysina Critical Zone Observatory in Central Europe

Climate warming is having profound effects on the hydrological cycle by increasing atmospheric demand, changing water availability, and snow seasonality. Europe suffered three distinct heat waves in 2019, and 11 of the 12 hottest years ever recorded took place in the past two decades, which will potentially change seasonal streamflow patterns and long-term trends. Central Europe exhibited six dry years in a row since 2014. This study uses data from a well-documented headwater catchment in Central Europe (Lysina) to explore hydrological responses to a warming climate. We applied a lumped parameter hydrologic model Brook90 and a distributed model Penn State Integrated Hydrologic Model (PIHM) to simulate long-term hydrological change under future climate scenarios. Both models performed well on historic streamflow and in agreement with each other according to the catchment water budget. In addition, PIHM was able to simulate lateral groundwater redistribution within the catchment validated by the groundwater table dynamics. The long-term trends in runoff and low flow were captured by PIHM only. We applied different EURO-CORDEX models with two emission scenarios (Representative Concentration Pathways RCP 4.5, 8.5) and found significant impacts on runoff and evapotranspiration (ET) for the period of 2071–2100. Results from both models suggested reduced runoff and increased ET, while the monthly distribution of runoff was different. We used this catchment study to understand the importance of subsurface processes in projection of hydrologic response to a warming climate.     

A simulation of large‐scale groundwater flow and travel time in a fractured rock environment for waste disposal purposes

Two‐dimensional regional groundwater flow was simulated based on a conceptual model of low‐permeability crystalline rocks of the Whiteshell Research Area (WRA) in south‐eastern Manitoba. The conceptual model consists of fracture zones that strike in different directions and dip at various angles in the background rock mass. The thickness and hydraulic properties of the fracture zones in the conceptual model were varied as were the fluid properties and the boundary conditions of the groundwater flow system. The effects of these variations on the groundwater flow pattern and on the convective travel time along pathways from a hypothetical disposal vault at 500 m depth to discharge locations at the ground surface were evaluated. The vault was located in the regional discharge area of the groundwater system. A homogeneous conceptual model of the WRA, having only freshwater flow, formed a groundwater flow pattern with a regional flow system. Local flow systems developed increasingly with the introduction of fracture zones 20 m and 3 m thick, and depth‐dependent fluid density. This indicates a reduction in groundwater residence time by fracture zones and fluid density. Flow pathways were analysed using both a stream‐function and a particle‐tracking technique. The pathways and their lengths from the location of the vault to the surface varied spatially according to the flow patterns. The minimum travel time along these pathways was less than 150 000 and greater than 4 000 000 years in models with and without fracture zones, respectively, indicating that the presence of fracture zones was the major controlling factor. A precise knowledge and refinement of conceptual model parameters is necessary during site selection for waste disposal purposes. Copyright © 2004 John Wiley & Sons, Ltd.     

Comparison of sparse‐grid geometric and random sampling methods in nonlinear inverse solution uncertainty estimation

A new uncertainty estimation method, which we recently introduced in the literature, allows for the comprehensive search of model posterior space while maintaining a high degree of computational efficiency. The method starts with an optimal solution to an inverse problem, performs a parameter reduction step and then searches the resulting feasible model space using prior parameter bounds and sparse‐grid polynomial interpolation methods. After misfit rejection, the resulting model ensemble represents the equivalent model space and can be used to estimate inverse solution uncertainty. While parameter reduction introduces a posterior bias, it also allows for scaling this method to higher dimensional problems. The use of Smolyak sparse‐grid interpolation also dramatically increases sampling efficiency for large stochastic dimensions. Unlike Bayesian inference, which treats the posterior sampling problem as a random process, this geometric sampling method exploits the structure and smoothness in posterior distributions by solving a polynomial interpolation problem and then resampling from the resulting interpolant. The two questions we address in this paper are 1) whether our results are generally compatible with established Bayesian inference methods and 2) how does our method compare in terms of posterior sampling efficiency. We accomplish this by comparing our method for two electromagnetic problems from the literature with two commonly used Bayesian sampling schemes: Gibbs’ and Metropolis‐Hastings. While both the sparse‐grid and Bayesian samplers produce compatible results, in both examples, the sparse‐grid approach has a much higher sampling efficiency, requiring an order of magnitude fewer samples, suggesting that sparse‐grid methods can significantly improve the tractability of inference solutions for problems in high dimensions or with more costly forward physics.     

Theoretical analysis of the worthiness of Henry and Elder problems as benchmarks of density-dependent groundwater flow models

Computer models must be tested to ensure that the mathematical statements and solution schemes accurately represent the physical processes of interest. Because the availability of benchmark problems for testing density-dependent groundwater models is limited, one should be careful in using these problems appropriately. Details of a Galerkin finite-element model for the simulation of density-dependent, variably saturated flow processes are presented here. The model is tested using the Henry salt-water intrusion problem and Elder salt convection problem. The quality of these benchmark problems is then evaluated by solving the problems in the standard density-coupled mode and in a new density-uncoupled mode. The differences between the solutions indicate that the Henry salt-water intrusion problem has limited usefulness in benchmarking density-dependent flow models because the internal flow dynamics are largely determined by the boundary forcing. Alternatively, the Elder salt-convection problem is more suited to the model testing process because the flow patterns are completely determined by the internal balance of pressure and gravity forces.     

Inverse methods in hydrogeology: Evolution and recent trends

Parameter identification is an essential step in constructing a groundwater model. The process of recognizing model parameter values by conditioning on observed data of the state variable is referred to as the inverse problem. A series of inverse methods has been proposed to solve the inverse problem, ranging from trial-and-error manual calibration to the current complex automatic data assimilation algorithms. This paper does not attempt to be another overview paper on inverse models, but rather to analyze and track the evolution of the inverse methods over the last decades, mostly within the realm of hydrogeology, revealing their transformation, motivation and recent trends. Issues confronted by the inverse problem, such as dealing with multiGaussianity and whether or not to preserve the prior statistics are discussed.