Conditional nonlinear optimal perturbation: Applications to stability,sensitivity, and predictability

Conditional nonlinear optimal perturbation (CNOP) is a nonlinear generalization of linear singular vector (LSV) and features the largest nonlinear evolution at prediction time for the initial perturbations in a given constraint. It was proposed initially for predicting the limitation of predictability of weather or climate. Then CNOP has been applied to the studies of the problems related to predictability for weather and climate. In this paper, we focus on reviewing the recent advances of CNOP’s applications, which involves the ones of CNOP in problems of ENSO amplitude asymmetry, block onset, and the sensitivity analysis of ecosystem and ocean’s circulations, etc. Especially, CNOP has been primarily used to construct the initial perturbation fields of ensemble forecasting, and to determine the sensitive area of target observation for precipitations. These works extend CNOP’s applications to investigating the nonlinear dynamical behaviors of atmospheric or oceanic systems, even a coupled system, and studying the problem of the transition between the equilibrium states. These contributions not only attack the particular physical problems, but also show the superiority of CNOP to LSV in revealing the effect of nonlinear physical processes. Consequently, CNOP represents the optimal precursors for a weather or climate event; in predictability studies, CNOP stands for the initial error that has the largest negative effect on prediction; and in sensitivity analysis, CNOP is the most unstable (sensitive) mode. In multi-equilibrium state regime, CNOP is the initial perturbation that induces the transition between equilibriums most probably. Furthermore, CNOP has been used to construct ensemble perturbation fields in ensemble forecast studies and to identify sensitive area of target observation. CNOP theory has become more and more substantial. It is expected that CNOP also serves to improve the predictability of the realistic predictions for weather and climate events plays an increasingly important role in exploring the nonlinear dynamics of atmospheric, oceanic and coupled atmosphere-ocean system. Supported by National Basic Research Program of China (Grant Nos. 2006CB403606, 2007CB411800), National Natural Science Foundation of China (Grant Nos. 40830955, 40675030, 40505013), Institute of Atmospheric Physics, Chinese Academy of Sciences (Grant No. IAP07202), and LASG State Key Laboratory Special Fund     

The application of the orthogonal conditional nonlinear optimal perturbations method to typhoon track ensemble forecasts

The orthogonal conditional nonlinear optimal perturbations (CNOPs) method, orthogonal singular vectors (SVs)method and CNOP+SVs method, which is similar to the orthogonal SVs method but replaces the leading SV (LSV) with the first CNOP, are adopted in both the Lorenz-96 model and Pennsylvania State University/National Center for Atmospheric Research (PSU/NCAR) Fifth-Generation Mesoscale Model (MM5) for ensemble forecasts. Using the MM5, typhoon track ensemble forecasting experiments are conducted for strong Typhoon Matsa in 2005. The results of the Lorenz-96 model show that the CNOP+SVs method has a higher ensemble forecast skill than the orthogonal SVs method, but ensemble forecasts using the orthogonal CNOPs method have the highest forecast skill. The results from the MM5 show that orthogonal CNOPs have a wider horizontal distribution and better describe the forecast uncertainties compared with SVs. When generating the ensemble mean forecast, equally averaging the ensemble members in addition to the anomalously perturbed forecast members may contribute to a higher forecast skill than equally averaging all of the ensemble members. Furthermore, for given initial perturbation amplitudes, the CNOP+SVs method may not have an ensemble forecast skill greater than that of the orthogonal SVs method, but the orthogonal CNOPs method is likely to have the highest forecast skill. Compared with SVs, orthogonal CNOPs fully consider the influence of nonlinear physical processes on the forecast results; therefore, considering the influence of nonlinearity may be important when generating fast-growing initial ensemble perturbations. All of the results show that the orthogonal CNOP method may be a potential new approach for ensemble forecasting.     

Ensemble prediction experiments using conditional nonlinear optimal perturbation

Two methods for initialization of ensemble forecasts are compared, namely, singular vector (SV) and conditional nonlinear optimal perturbation (CNOP). The comparison is done for forecast lengths of up to 10 days with a three-level quasi-geostrophic (QG) atmospheric model in a perfect model scenario. Ten cases are randomly selected from 1982/1983 winter to 1993/1994 winter (from December to the following February). Anomaly correlation coefficient (ACC) is adopted as a tool to measure the quality of the predicted ensembles on the Northern Hemisphere 500 hPa geopotential height. The results show that the forecast quality of ensemble samples in which the first SV is replaced by CNOP is higher than that of samples composed of only SVs in the medium range, based on the occurrence of weather regime transitions in Northern Hemisphere after about four days. Besides, the reliability of ensemble forecasts is evaluated by the Rank Histograms. The above conclusions confirm and extend those reached earlier by the authors, which stated that the introduction of CNOP improves the forecast skill under the condition that the analysis error belongs to a kind of fast-growing error by using a barotropic QG model. Supported by Knowledge Innovation Project of the Chinese Academy of Sciences (Grant No. KZCX3-SW-230), National Natural Science Foundation of China (Grant Nos. 40675030, 40633016)     

Nonlinear singular vectors and nonlinear singular values

A novel concept of nonlinear singular vector and nonlinear singular value is introduced, which is a natural generalization of the classical linear singular vector and linear singular value to the nonlinear category. The optimization problem related to the determination of nonlinear singular vectors and singular values is formulated. The general idea of this approach is demonstrated by a simple two-dimensional quasigeostrophic model in the atmospheric and oceanic sciences. The advantage and its applications of the new method to the predictability, ensemble forecast and finite-time nonlinear instability are discussed. This paper makes a necessary preparation for further theoretical and numerical investigations.     

Nonlinear fastest growing perturbation and the first kind of predictability

Nonlinear fastest growing perturbation, which is related to the nonlinear singular vector and nonlinear singular value proposed by the first author recently, is obtained by numerical approach for the two-dimensional quasigeostrophic model in this paper. The difference between the linear and nonlinear fastest growing perturbations is demonstrated. Moreover, local nonlinear fastest growing perturbations are also found numerically. This is one of the essential differences between linear and nonlinear theories, since in former case there is no local fastest growing perturbation. The results show that the nonlinear local fastest growing perturbations play a more important role in the study of the first kind of predictability than the nonlinear global fastest growing perturbation.     

An explicit four-dimensional variational data assimilation method

A new data assimilation method called the explicit four-dimensional variational (4DVAR) method is proposed. In this method, the singular value decomposition (SVD) is used to construct the orthogonal basis vectors from a forecast ensemble in a 4D space. The basis vectors represent not only the spatial structure of the analysis variables but also the temporal evolution. After the analysis variables are ex-pressed by a truncated expansion of the basis vectors in the 4D space, the control variables in the cost function appear explicitly, so that the adjoint model, which is used to derive the gradient of cost func-tion with respect to the control variables, is no longer needed. The new technique significantly simpli-fies the data assimilation process. The advantage of the proposed method is demonstrated by several experiments using a shallow water numerical model and the results are compared with those of the conventional 4DVAR. It is shown that when the observation points are very dense, the conventional 4DVAR is better than the proposed method. However, when the observation points are sparse, the proposed method performs better. The sensitivity of the proposed method with respect to errors in the observations and the numerical model is lower than that of the conventional method.     

改进的伴随状态法初至波走时层析成像方法

目前,有关伴随状态法初至波走时层析成像方法的文献,基本上都是基于面积分来定义目标函数,由此得到的伴随方程也都依赖于地表的法向量.这样,一方面会因为伴随变量计算的不准确而造成梯度的不合理,另一方面也无法合理地处理井中观测问题.本文从理论或数值试验角度指出了这些问题,并提出了不依赖地表法向量的改进的伴随状态法走时层析成像方法.主要改进包括:(1)采用体积分定义目标函数,避免了传统方法不能较好处理井中观测数据的缺陷,可以适应任意地表或井中观测系统.(2)采用摄动法得到了新的伴随方程,克服了传统方法中伴随场计算需要依赖于地表法向量的缺陷,使得检波点处的走时残差可以正确地反传播至地下,进而得到更加合理的速度修正方向,提高了速度反演的精度.     

Equilibration mechanisms in an adjoint ocean general circulation model

We examine the equilibrated and time-evolving adjoint solutions of an ocean general circulation model. Adjoint models calculate the sensitivity of a diagnostic, (here, the strength of the meridional overturning) to all forcing fields in a single integration. The time evolution of the sensitivity patterns demonstrates the validity of the adjoint modeling approach over climatological time scales in coarse-resolution ocean models. Our objective is to identify the principle adjustment mechanisms through which the meridional overturning strength adapts to perturbations in wind and buoyancy forcing. The adjoint approach is shown to be a valuable alternative to traditional perturbation methods in highlighting the processes and time scales important to ocean and climate modeling.     

用地质雷达数据资料反演二维地下介质的方法

从二维麦克斯韦方程组出发推导出反演介电常数和电导率等二维介质物性参数的反演公式.反演的步骤是: 建立初始猜测模型，利用电磁波时间域有限差分法模拟正演数据，用正演数据与观测数据之间的数据残差建立目标函数，通过引入一个由麦克斯韦方程计算的伴随场，将目标函数对介质参数的导数表示成显式形式，应用最优化理论得出对初始猜测模型的修改，用共轭梯度法迭代，最终得到反演结果.用合成数据反演具有粗糙地表的非导电介质的介电常数，用实验数据同时反演介电常数和电导率，并比较了麦克斯韦方程反演结果与声波方程反演结果、波动方程偏移剖面的差异.     

复杂岩性解释模型稳定性和可靠性评价

对解释模型进行评价以及如何选择合适的解释模型是复杂岩性解释中非常重要的问题。本文应用统计学的理论以及非线性优化技术,系统研究了在线性约束条件下对储层参数的置信区域和误差大小进行估计的问题,通过将梯度投影算法与奇异值分解技术相结合的方法给出了计算储层参数置信区域与误差大小的公式,以及对复杂岩性解释模型的稳定性和可靠性进行合理评价的具体方法。最后通过数值计算实例研究了一种能够有效提高解释模型稳定性和可靠性的措施,数值计算结果证明解释模型响应函数的最小非零奇异值大小是影响稳定性和可靠性的一个重要参数,当最小非零奇异值较大时,模型的稳定性往往较好,否则较差,因此通过对解释模型进行适当组合并选取最小非零奇异值较大的综合解释模型,可以大大提高其稳定性和可靠性。     

An evaluation of a class of practical optimization techniques for structural dynamics applications

This paper evaluates a class of practical optimization techniques for parameter identification of realistic structural dynamic systems. The techniques involve quasi-Newton methods together with an efficient procedure for estimating complicated error functions. The optimization procedures are verified through their application to several representative examples, including finite-element models of realistic structural systems. Extensive numerical and graphical results demonstrate the effects of various optimization algorithm parameters on the rate of convergence of the objective function, the parameter vector error norm and the gradient norm. Guidelines are presented as an aid for addressing several significant issues in the practical application of structural dynamics optimization procedures, such as sensitivity problems, uniqueness, initial value definition for the parameter vector, convergence rates, constraints, the effect of alternative cost function definitions, accuracy of alternative gradient evaluation procedures, alternative procedures for estimating the inverse of the Hessian matrix and the use of a quadratic approximation of the objective function.     

A design procedure for buildings equipped with energy dissipation devices using nonclassical damping and iso-performance curves

This article describes a design procedure for elastic buildings equipped with linear and nonlinear energy dissipating devices. The objective is to achieve a design that responds to a target building performance following a simple and robust step-by-step algorithm. The proposed procedure identifies first the modal significance of key design performance indicators and controls the modal properties by solving a singular two-parameter eigenvalue problem. For that purpose, a new modal significance metric is proposed, and a target frequency shift and damping ratio for the complete structure are obtained from the so-called iso-performance design curves. The design algorithm employs linear-equivalent stiffness and damping properties, which are then transformed into parameters characterizing inelastic force-deformation constitutive models corresponding to physical devices. The design algorithm leads to an optimal damper distribution corresponding to the minimum global amount of supplemental equivalent damping needed to achieve a maximum modal perturbation. The design procedure is first demonstrated using a five-story building example and then a real and complex 22-story free-plan building with two towers of rhomboid-shape plan with a very singular dynamic behavior.     

An explicit four-dimensional variational data assimilation method based on the proper orthogonal decomposition: Theoretics and evaluation

The proper orthogonal decomposition (POD) method is used to construct a set of basis functions for spanning the ensemble of data in a certain least squares optimal sense. Compared with the singular value decomposition (SVD), the POD basis functions can capture more energy in the forecast ensemble space and can represent its spatial structure and temporal evolution more effectively. After the analysis variables are expressed by a truncated expansion of the POD basis vectors in the ensemble space, the control variables appear explicitly in the cost function, so that the adjoint model, which is used to derive the gradient of the cost function with respect to the control variables, is no longer needed. The application of this new technique significantly simplifies the data assimilation process. Several assimilation experiments show that this POD-based explicit four-dimensional variational data assimilation method performs much better than the usual ensemble Kalman filter method on both enhancing the assimilation precision and reducing the computation cost. It is also better than the SVD-based explicit four-dimensional assimilation method, especially when the forecast model is not perfect and the forecast error comes from both the noise of the initial filed and the uncertainty of the forecast model. Supported by the National Natural Science Foundation of China (Grant No. 40705035), National High Technology Research and Development Program of China (Grant No. 2007AA12Z144), Knowledge Innovation Project of Chinese Academy of Sciences (Grant Nos. KZCX2-YW-217 and KZCX2-YW-126-2), and National Basic Research Program of China (Grant No. 2005CB321704)     

Automating adjoint wave-equation travel-time tomography using scientific workflow

Recent advances in commodity high-performance computing technology have dramatically reduced the computational cost for solving the seismic wave equation in complex earth structure models. As a consequence, wave-equation-based seismic tomography techniques are being actively developed and gradually adopted in routine subsurface seismic imaging practices. Wave-equation travel-time tomography is a seismic tomography technique that inverts cross-correlation travel-time misfits using full-wave Fréchet kernels computed by solving the wave equation. This technique can be implemented very efficiently using the adjoint method, in which the misfits are back-propagated from the receivers (i.e., seismometers) to produce the adjoint wave-field and the interaction between the adjoint wave-field and the forward wave-field from the seismic source gives the gradient of the objective function. Once the gradient is available, a gradient-based optimization algorithm can then be adopted to produce an optimal earth structure model that minimizes the objective function. This methodology is conceptually straightforward, but its implementation in practical situations is highly complex, error-prone and computationally demanding. In this study, we demonstrate the feasibility of automating wave-equation travel-time tomography based on the adjoint method using Kepler, an open-source software package for designing, managing and executing scientific workflows. The workflow technology allows us to abstract away much of the complexity involved in the implementation in a manner that is both robust and scalable. Our automated adjoint wave-equation travel-time tomography package has been successfully applied on a real active-source seismic dataset.     

复杂地下异常体的可控源电磁法积分方程正演

可控源电磁法具有分辨率高及抗干扰能力强等特点,是一种重要的地电磁勘探方法.目前,可控源电磁法的高精度正演计算一直是其核心研究问题之一.传统积分方程法一般采用近似积分公式、简单矩形网格和近似的奇异性体积分计算技术,制约了体积分方程法处理复杂地下异常体的能力,降低了计算精度.针对上述问题,本文基于完全积分公式、四面体非结构化网格和奇异体积分的精确解析解来高精度求解复杂可控源电磁模型的正演响应.首先,从电场积分公式出发,推导了可控源电磁问题满足的积分方程;其次,借助于非结构化四面体网格离散技术,实现了地下复杂异常体的有效模拟.最后,利用散度定理把强奇异值体积分转换为一系列弱奇异性的面积分公式,并通过推导获得了这些弱奇异性的面积分公式的解析解,从而最终实现三维可控源电磁问题的高精度积分求解.以块状低阻体地电模型为测试模型,采用本文提出的积分方程方法获得的数值解与其他公开数值算法解进行对比分析,其对比结果具有高度的吻合性,验证了算法的正确性;同时,设计了球状及复杂地电模型进行算法收敛性测试,进一步验证算法的正确性以及能够处理地下复杂模型的能力.     

Adjoint method for the optimum planning of industrial pollutant sources

The deterioration of air quality is threatening the life and health of people. Scientists in China and other countries have done a great deal of research work on the details of air pollution and the methods of preven-tion and control during the past decades. Up to now, most of the achievements are concentrated on the techniques of controlling pollutant sources and the programs of reduction, which focus on the improve-ment of air quality and the restoration of environment. The techniques of con…     

基于一阶速度-应力方程的VTI介质最小二乘逆时偏移

地下地层普遍存在各向异性,忽略介质各向异性会导致速度估计不准确,成像精度下降.基于二阶声波方程的最小二乘逆时偏移忽略了介质各向异性及密度变化的影响,致使模拟地震数据与实际观测数据不匹配,影响收敛速度和反演成像质量.VTI介质一阶速度-应力方程能较好适应各向异性变密度情况,为此,本文首先从VTI介质一阶速度-应力方程出发,进行波动方程线性化;其次推导了相应的扰动方程和伴随方程,并通过伴随状态法得到梯度更新公式;最终形成基于一阶方程的LSRTM算法理论及实现流程.在实现算法的基础上,通过数值试算及成像结果对比,验证了本文算法在处理变密度和VTI介质时的有效性和优越性.偏移速度以及各向异性Thomsen参数误差的敏感性测试及误差收敛曲线对比结果进一步表明:速度及Thomsen参数对成像结果存在明显影响,其中速度敏感性最强,参数epsilon次之,参数delta的敏感性最弱.     

Identifying the sensitive area in adaptive observation for predicting the upstream Kuroshio transport variation in a 3-D ocean model

Using the conditional nonlinear optimal perturbation(CNOP) approach, sensitive areas of adaptive observation for predicting the seasonal reduction of the upstream Kuroshio transport(UKT) were investigated in the Regional Ocean Modeling System(ROMS). The vertically integrated energy scheme was utilized to identify sensitive areas based on two factors: the specific energy scheme and sensitive area size. Totally 27 sensitive areas, characterized by three energy schemes and nine sensitive area sizes, were evaluated. The results show that the total energy(TE) scheme was the most effective because it includes both the kinetic and potential components of CNOP. Generally, larger sensitive areas led to better predictions. The size of 0.5% of the model domain was chosen after balancing the effectiveness and efficiency of adaptive observation. The optimal sensitive area OSen was determined accordingly. Sensitivity experiments on OSen were then conducted, and the following results were obtained:(1) In OSen, initial errors with CNOP or CNOP-like patterns were more likely to yield worse predictions, and the CNOP pattern was the most unstable.(2) Initial errors in OSen rather than in other regions tended to cause larger prediction errors. Therefore, adaptive observation in OSen can be more beneficial for predicting the seasonal reduction of UKT.     

Sparse calibration of subsurface flow models using nonlinear orthogonal matching pursuit and an iterative stochastic ensemble method

We introduce a nonlinear orthogonal matching pursuit (NOMP) for sparse calibration of subsurface flow models. Sparse calibration is a challenging problem as the unknowns are both the non-zero components of the solution and their associated weights. NOMP is a greedy algorithm that discovers at each iteration the most correlated basis function with the residual from a large pool of basis functions. The discovered basis (aka support) is augmented across the nonlinear iterations. Once a set of basis functions are selected, the solution is obtained by applying Tikhonov regularization. The proposed algorithm relies on stochastically approximated gradient using an iterative stochastic ensemble method (ISEM). In the current study, the search space is parameterized using an overcomplete dictionary of basis functions built using the K-SVD algorithm. The proposed algorithm is the first ensemble based algorithm that tackels the sparse nonlinear parameter estimation problem.     

一阶Rytov近似有限频走时层析

传统的波动方程走时核函数(或走时Fréchet导数)多基于互相关时差测量方式及地震波场的一阶Born近似导出,其成立条件非常苛刻.然而,地震波走时与大尺度的速度结构具有良好的线性关系,对于小角度的前向散射波场,Rytov近似优于Born近似.因此,本文基于Rytov近似和互相关时差测量方式,导出了基于Rytov近似的有限频走时敏感度核函数的两种等价形式:频率积分和时间积分表达式.在此基础之上,本文提出了一种隐式矩阵向量乘方法,可以直接计算Hessian矩阵或者核函数与向量的乘积,而无需显式计算和存储核函数及Hessian矩阵.基于隐式矩阵向量乘方法,本文利用共轭梯度法求解法方程实现了一种高效的Gauss-Newton反演算法求解走时层析反问题.与传统的敏感度核函数反演方法相比,本文方法在每次迭代过程中,无需显式计算和存储核函数,极大降低了存储需求.与基于Born近似的伴随状态方法走时层析相比,本文方法具有准二阶的收敛速度,且适用范围更广.数值试验证明了本文方法的有效性.