Mixed finite element methods and higher order temporal approximations for variably saturated groundwater flow

Richards’ equation (RE) is commonly used to model flow in variably saturated porous media. However, its solution continues to be difficult for many conditions of practical interest. Among the various time discretizations applied to RE, the method of lines (MOL) has been used successfully to introduce robust, accurate, and efficient temporal approximations. At the same time, a mixed-hybrid finite element method combined with an adaptive, higher order time discretization has shown benefits over traditional, lower order temporal approximations for modeling single-phase groundwater flow in heterogeneous porous media. Here, we extend earlier work for single-phase flow and consider two mixed finite element methods that have been used previously to solve RE using lower order time discretizations with either fixed time steps or empirically based adaption. We formulate the two spatial discretizations within a MOL context for the pressure head form of RE as well as a fully mass-conservative version. We conduct several numerical experiments for both spatial discretizations with each formulation, and we compare the higher order, adaptive time discretization to a first-order approximation with formal error control and adaptive time step selection. Based on the numerical results, we evaluate the performance of the methods for robustness and efficiency.     

A methodology for solute transport in unsteady,nonuniform streamflow with subsurface interaction

An advection-dispersion-reaction model can generally be used to describe one-dimensional stream solute transport if the flow is steady and if the channel is smooth and uniform. When applied to unsteady, nonuniform streamflows, a model based on the Fickian analogy needs to be modified to account for the temporal and spatial variation of the cross-sectional area of the stream channel. In this paper, we explore this topic with a simple approximation method as well as an elaborate one, both of which are incorporated into a conjunctive stream–aquifer transport model and are applied to a hypothetical stream–aquifer setting. The simple method, while easier to implement, displays a persistent pattern of error in simulation results. The elaborate method, while accurate in computation, results in a more complicated model and requires extensive procedures to overcome the efficiency problem when simulating complex stream–aquifer interactions. However, by coupling the latter with the adaptive stepsize control for the Runge–Kutta method in a conjunctive stream–aquifer model, it not only greatly improves model efficiency but also results in more realistic modeling than previously reported.     

一种时空域优化的高精度交错网格差分算子与正演模拟

如何有效压制数值频散是有限差分正演模拟研究中的关键问题之一.近年来,许多学者对二阶声波方程的差分算子开展了大量的优化工作,在压制频散方面取得不错的效果.一阶压强-速度方程广泛用于研究地震波在地下变密度模型中传播规律,目前针对一阶方程的优化工作大多只是在空间差分算子上展开.本文在前人研究的基础上,推导出一阶声波方程中压强场与偏振速度场之间的解析关系,据此在传统交错网格基础上给出一种高精度的显式时间递推格式,该递推格式将时间差分与空间差分算子结合在一起,并采用共轭梯度法得到精确时间递推匹配系数,实现时空差分算子的同时优化.在编程实现算法的基础上,通过频散分析与三个典型模型测试表明:本文方法能够较为有效地压制时间频散与空间频散,提高数值计算精度;同时对复杂模型也有很好适用性.     

A spatially and temporally adaptive solution of Richards’ equation

Efficient, robust simulation of groundwater flow in the unsaturated zone remains computationally expensive, especially for problems characterized by sharp fronts in both space and time. Standard approaches that employ uniform spatial and temporal discretizations for the numerical solution of these problems lead to inefficient and expensive simulations. In this work, we solve Richards’ equation using adaptive methods in both space and time. Spatial adaption is based upon a coarse grid solve and a gradient error indicator using a fixed-order approximation. Temporal adaption is accomplished using variable order, variable step size approximations based upon the backward difference formulas up to fifth order. Since the advantages of similar adaptive methods in time are now established, we evaluate our method by comparison with a uniform spatial discretization that is adaptive in time for four different one-dimensional test problems. The numerical results demonstrate that the proposed method provides a robust and efficient alternative to standard approaches for simulating variably saturated flow in one spatial dimension.     

VTI介质纯P波混合法正演模拟及稳定性分析

各向异性介质纯P波方程完全不受横波的干扰,在一定程度上可以减缓由于介质各向异性引起的数值不稳定,本文推导了具有垂直对称轴的横向各向同性(VTI)介质纯P波一阶速度-应力方程.由于纯P波方程存在一个分数形式的伪微分算子,无法直接采用有限差分法求解.针对该问题,本文采用伪谱法和高阶有限差分法联合求解波动方程,重点分析了混合法求解纯P波一阶速度-应力方程的稳定性问题,并给出了混合法求解纯P波方程的稳定性条件.数值模拟结果表明纯P波方程伪谱法和高阶有限差分混合法能够进行复杂介质的正演模拟,在强变速度、变密度的地球介质中仍然具有较好的稳定性.     

A finite volume upwind scheme for the solution of the linear advection–diffusion equation with sharp gradients in multiple dimensions

A finite volume upwind numerical scheme for the solution of the linear advection equation in multiple dimensions on Cartesian grids is presented. The small-stencil, Modified Discontinuous Profile Method (MDPM) uses a sub-cell piecewise constant reconstruction and additional information at the cell interfaces, rather than a spatial extension of the stencil as in usual methods. This paper presents the MDPM profile reconstruction method in one dimension and its generalization and algorithm to two- and three-dimensional problems. The method is extended to the advection–diffusion equation in multiple dimensions. The MDPM is tested against the MUSCL scheme on two- and three-dimensional test cases. It is shown to give high-quality results for sharp gradients problems, although some scattering appears. For smooth gradients, extreme values are best preserved with the MDPM than with the MUSCL scheme, while the MDPM does not maintain the smoothness of the original shape as well as the MUSCL scheme. However the MDPM is proved to be more efficient on coarse grids in terms of error and CPU time, while on fine grids the MUSCL scheme provides a better accuracy at a lower CPU.     

Construction of ray synthetic seismograms using interpolation of travel times and ray amplitudes

The seismic wave field, in its high-frequency asymptotic approximation, can be interpolated from a low- to a high-resolution spatial grid of receivers and, possibly, point sources by interpolating the eikonal (travel time) and the amplitude. These quantities can be considered as functions of position only. The travel time and the amplitude are assumed to vary in space only slowly, otherwise the validity conditions of the theory behind would be violated. Relatively coarse spatial sampling is then usually sufficient to obtain their reasonable interpolation. The interpolation is performed in 2-D models of different complexity. The interpolation geometry is either 1-D, 2-D, or 3-D according to the source-receiver distribution. Several interpolation methods are applied: the Fourier interpolation based on the sampling theorem, the linear interpolation, and the interpolation by means of the paraxial approximation. These techniques, based on completely different concepts, are tested by comparing their results with a reference ray-theory solution computed for gathers and grids with fine sampling. The paraxial method holds up as the most efficient and accurate in evaluating travel times from all investigated techniques. However, it is not suitable for approximation of amplitudes, for which the linear interpolation has proved to be universal and accurate enough to provide results acceptable for many seismological applications.     

基于Remez迭代算法求解差分系数的双相介质自适应有限差分正演模拟

利用传统有限差分方法对基于Biot理论的双相介质波动方程进行数值求解时,由于慢纵波的存在,数值频散效应较为明显,影响模拟精度.相对于声学近似方程及普通弹性波方程,Biot双相介质波动方程在同等数值求解算法和精度要求条件下,其地震波场正演模拟需要更多的计算时间.本文针对Biot一阶速度-应力方程组发展了一种变阶数优化有限差分数值模拟方法,旨在同时提高其正演模拟的精度和效率.首先结合交错网格差分格式推导Biot方程的数值频散关系式.然后基于Remez迭代算法求取一阶空间偏导数的优化差分系数,并用于Biot方程的交错网格有限差分数值模拟.在此基础上把三类波的平均频散误差参数限制在给定的频散误差阈值和频率范围内,此时优化有限差分算子的长度就能自适应非均匀双相介质模型中的不同速度区间.数值频散曲线分析表明:基于Remez迭代算法的优化有限差分方法相较传统泰勒级数展开方法在大波数范围对频散误差的压制效果更明显;可变阶数的优化有限差分方法能取得与固定阶数优化有限差分方法相近的模拟精度.在均匀介质和河道模型的数值模拟实验中将本文变阶数优化有限差分算法与传统泰勒展开算法、最小二乘优化算法进行比较,进一步证明其在复杂地下介质中的有效性和适用性.     

波动方程叠前深度偏移直接产生角道集

基于单程波深度延拓算法,得到震源波场在成像点的入射角度,结合对地层倾角的估计,获得入射地震波与界面法线的夹角.通过运用"保幅"的反褶积成像条件和考虑累加的炮数,解决了炮点覆盖不均匀导致的成像幅值误差问题,进而建议了炮域波动方程叠前深度偏移直接产生角道集的方法和流程.与基于空间移动或时移成像条件的波动方程叠前深度偏移提取角道集的方法相比,本文建议的方法只需少量额外的存储空间,又可补偿观测系统非均匀覆盖对成像幅值的影响;其增加的计算量与炮域偏移算法相比几乎可以忽略.文中算例表明,本文方法提取的角道集可为叠前反演提供较精确的AVO振幅特性.此外,就改善地震成像效果本身而言,提取角道集使得可在波动方程叠前深度偏移中应用剩余动校和拉伸切除技术,从而可更好地保持高频成分并提高成像的信噪比.     

基于三角网格的有限差分法叠后逆时偏移

Compared with other migration methods, reverse-time migration is based on a precise wave equation, not an approximation, and performs extrapolation in the depth domain rather than the time domain. It is highly accurate and not affected by strong subsurface structure complexity and horizontal velocity variations. The difference method based on triangular grids maintains the simplicity of the difference method and the precision of the finite element method. It can be used directly for forward modeling on models with complex top surfaces and migration without statics preprocessing. We apply a finite difference method based on triangular grids for post-stack reverse-time migration for the first time. Tests on model data verify that the combination of the two methods can achieve near-perfect results in application.     

Lebedev scheme for the numerical simulation of wave propagation in 3D anisotropic elasticity‡

This paper presents a Lebedev finite difference scheme on staggered grids for the numerical simulation of wave propagation in an arbitrary 3D anisotropic elastic media. The main concept of the scheme is the definition of all the components of each tensor (vector) appearing in the elastic wave equation at the corresponding grid points, i.e., all of the stresses are stored in one set of nodes while all of the velocity components are stored in another. Meanwhile, the derivatives with respect to the spatial directions are approximated to the second order on two‐point stencils. The second‐order scheme is presented for the sake of simplicity and it is easy to expand to a higher order. Another approach, widely‐known as the rotated staggered grid scheme, is based on the same concept; therefore, this paper contains a detailed comparative analysis of the two schemes. It is shown that the dispersion condition of the Lebedev scheme is less restrictive than that of the rotated staggered grid scheme, while the stability criteria lead to approximately equal time stepping for the two approaches. The main advantage of the proposed scheme is its reduced computational memory requirements. Due to a less restrictive dispersion condition and the way the media parameters are stored, the Lebedev scheme requires only one‐third to two‐thirds of the computer memory required by the rotated staggered grid scheme. At the same time, the number of floating point operations performed by the Lebedev scheme is higher than that for the rotated staggered grid scheme.     

Coupling water flow and solute transport into a physically-based surface-subsurface hydrological model

A distributed-parameter physically-based solute transport model using a novel approach to describe surface-subsurface interactions is coupled to an existing flow model. In the integrated model the same surface routing and mass transport equations are used for both hillslope and channel processes, but with different parametrizations for these two cases. For the subsurface an advanced time-splitting procedure is used to solve the advection-dispersion equation for transport and a standard finite element scheme is used to solve Richards equation for flow. The surface-subsurface interactions are resolved using a mass balance-based surface boundary condition switching algorithm that partitions water and solute into actual fluxes across the land surface and changes in water and mass storage. The time stepping strategy allows the different time scales that characterize surface and subsurface water and solute dynamics to be efficiently and accurately captured. The model features and performance are demonstrated in a series of numerical experiments of hillslope drainage and runoff generation.     

ADAPTIVE SIGNAL PROCESSING THROUGH STOCHASTIC APPROXIMATION*

One of the problems in signal processing is estimating the impulse response function of an unknown system. The well-known Wiener filter theory has been a powerful method in attacking this problem. In comparison, the use of stochastic approximation method as an adaptive signal processor is relatively new. This adaptive scheme can often be described by a recursive equation in which the estimated impulse response parameters are adjusted according to the gradient of a predetermined error function. This paper illustrates by means of simple examples the application of stochastic approximation method as a single-channel adaptive processor. Under some conditions the expected value of its weight sequence converges to the corresponding Wiener optimum filter when the least-mean-square error criterion is used.     

Adaptive local discontinuous Galerkin approximation to Richards’ equation

We propose a spatially and temporally adaptive solution to Richards’ equation based upon a local discontinuous Galerkin approximation in space and a high-order, backward difference method in time. We cast our approach in terms of a general, decoupled adaption algorithm based upon operators. We define non-unique instances of all operators to result in an adaption method from within the general class of methods that is defined. We formally decouple the spatial adaption from the temporal adaption using a method of lines approach and limit the temporal truncation error so that the total error is dominated by the spatial component. We use a multiple grid approach to guide adaption and support the data structures. Spatial adaption decisions are based upon error and regularity indicators, which are economical to compute. The resultant methods are compared for two test problems. The results show that the proposed adaption methods are superior to methods that adapt only in time and that in cases in which the problem has sufficient smoothness, adapting the order of the elements in addition to the grid spacing can further improve the efficiency of this robust solution approach.     

Numerical modeling of 3D fully nonlinear potential periodic waves

A simple and exact numerical scheme for long-term simulations of 3D potential fully nonlinear periodic gravity waves is suggested. The scheme is based on the surface-following nonorthogonal curvilinear coordinate system. Velocity potential is represented as a sum of analytical and nonlinear components. The Poisson equation for the nonlinear component of velocity potential is solved iteratively. Fourier transform method, the second-order accuracy approximation of vertical derivatives on a stretched vertical grid and the fourth-order Runge–Kutta time stepping are used. The scheme is validated by simulation of steep Stokes waves. A one-processor version of the model for PC allows us to simulate evolution of a wave field with thousands degrees of freedom for hundreds of wave periods. The scheme is designed for investigation of nonlinear 2D surface waves, generation of extreme waves, and direct calculations of nonlinear interactions.     

一种模拟非均匀介质中弹性波传播新的k-space方法

传统的伪谱(PS)方法,采用傅里叶变换(FT)计算空间导数具有很高的精度,每个波长仅需要两个采样点,而时间导数采用有限差分(FD)近似因而精度较低.当采用大时间步长时,由于时空精度不平衡,PS法存在不稳定性问题.原始的k-space方法可以有效地克服这些问题但是却无法适用于非均匀介质.为了提高原始k-space方法模拟非均匀介质波动方程的精度,我们提出了一种新的k-space算子族.它是用非均匀介质的变速度代替原k-space算子中的常数补偿速度构造得到,引入低秩近似可以高效求解.我们将构造的新的k-space算子应用于耦合的二阶位移波动方程,而不是交错网格一阶速度应力波动方程,使模拟弹性波的计算存储量减少.我们从数学上证明了基于二阶波动方程的k-space方法与基于一阶波动方程的k-space方法是等价的.数值模拟实验表明,与传统的PS、交错网格PS和原始的k-space方法相比,我们的新方法可以在时间和空间步长较大的均匀和非均匀介质中,为弹性波的传播提供更精确的数值解.在保持稳定性和精度的同时,采用较大的时空采样间隔,可以大大降低数值模拟的计算成本.     

Mass-conservative finite volume methods on 2-D unstructured grids for the Richards’ equation

The solution to the 2-D time-dependent unsaturated flow equation is numerically approximated by a second-order accurate cell-centered finite-volume discretization on unstructured grids. The approximation method is based on a vertex-centered Least Squares linear reconstruction of the solution gradients at mesh edges.A Taylor series development in time of the water content dependent variable in a finite-difference framework guarantees that the proposed finite volume method is mass conservative. A Picard iterative scheme solves at each time step the resulting non-linear algebraic problem. The performance of the method is assessed on five different test cases and implementing four distinct soil constitutive relationships. The first test case deals with a column infiltration problem. It shows the capability of providing a mass-conservative behavior. The second test case verifies the numerical approximation by comparison with an analytical mixed saturated–unsaturated solution. In this case, the water drains from a fully saturated portion of a 1-D column. The third and fourth test cases illustrate the performance of the approximation scheme on sharp soil heterogeneities on 1-D and 2-D multi-layered infiltration problems. The 2-D case shows the passage of an abrupt infiltration front across a curved interface between two layers. Finally, the fifth test case compares the numerical results with an analytical solution that is developed for a 2-D heterogeneous soil with a source term representing plant roots. This last test case illustrates the formal second-order accuracy of the method in the numerical approximation of the pressure head.     

Implementation of new time integration methods in POM

Instead of the standard leapfrog (SLF) scheme, an alternative leapfrog (ALF) scheme is used to solve the barotropic equations of the external mode in the Princeton Ocean Model (POM). The ALF scheme is modified in this study to deal with the nonlinear finite amplitude surface displacement. ALF has the advantage of improved numerical properties, longer time step relative to SLF, conservation of energy, and elimination of the Asselin filter. The numerical experiments of POM are implemented to show the above advantages. The split time stepping in 3D POM is found in this study to have numerical discrepancy due to the mismatched stepping between external and internal modes, and it results in a splitting error between the external and internal modes. A new split time stepping is therefore proposed. Numerical analysis indicates that there is no discrepancy with this split time stepping. The new split time stepping is implemented in the 3D POM. The numerical experiments demonstrate that the splitting error in POM can be reduced by three orders of magnitude relative to the original formulation, though the numerical error of the original formulation is already quite small.