Dacian N. Daescu

Second-Order Information in Data Assimilation*

In variational data assimilation (VDA) for meteorological and/or oceanic models, the assimilated fields are deduced by combining the model and the gradient of a cost functional measuring discrepancy between model solution and observation, via a first-order optimality system. However, existence and uniqueness of the VDA problem along with convergence of the algorithms for its implementation depend on the convexity of the cost function. Properties of local convexity can be deduced by studying the Hessian of the cost function in the vicinity of the optimum. This shows the necessity of second-order information to ensure a unique solution to the VDA problem. In this paper a comprehensive review of issues related to second-order analysis of the problem of VDA is presented along with many important issues closely connected to it. In particular issues of existence, uniqueness, and regularization through second-order properties are examined. The focus then shifts to second-order information related to statistical properties and to issues related to preconditioning and optimization methods and second-order VDA analysis. Predictability and its relation to the structure of the Hessian of the cost functional is then discussed along with issues of sensitivity analysis in the presence of data being assimilated. Computational complexity issues are also addressed and discussed. Automatic differentiation issues related to second-order information are also discussed along with the computational complexity of deriving the second-order adjoint. Finally an application aimed at illustrating the use of automatic differentiation for deriving the second-order adjoint as well as the Hessian/vector product applied to minimizing a cost functional of a meteorological problem using the truncated-Newton method is presented. Results verifying numerically the computational cost of deriving the second-order adjoint as well as results related to the spectrum of the Hessian of the cost functional are displayed and discussed.

Predicting air quality: Improvements through advanced methods to integrate models and measurements

Air quality prediction plays an important role in the management of our environment. Computational power and efficiencies have advanced to the point where chemical transport models can predict pollution in an urban air shed with spatial resolution less than a kilometer, and cover the globe with a horizontal resolution of less than 50km. Predicting air quality remains a challenge due to the complexity of the governing processes and the strong coupling across scales. While air quality prediction is closely aligned with weather prediction, there are important differences, including the role of pollution emissions and their associated large uncertainties. Improvements in air quality prediction require a close integration of observations. As more atmospheric chemical observations become available chemical data assimilation is expected to play an essential role in air quality forecasting. In this paper advances in air quality forecasting are discussed with an emphasis on data assimilation. Applications of the four-dimensional variational method (4D-Var) and the ensemble Kalman filter (EnKF) approach are presented and the computation challenges are discussed.

A Dual-Weighted Approach to Order Reduction in 4DVAR Data Assimilation

Abstract Strategies to achieve order reduction in four-dimensional variational data assimilation (4DVAR) search for an optimal low-rank state subspace for the analysis update. A common feature of the reduction methods proposed in atmospheric and oceanographic studies is that the identification of the basis functions relies on the model dynamics only, without properly accounting for the specific details of the data assimilation system (DAS). In this study a general framework of the proper orthogonal decomposition (POD) method is considered and a cost-effective approach is proposed to incorporate DAS information into the order-reduction procedure. The sensitivities of the cost functional in 4DVAR data assimilation with respect to the time-varying model state are obtained from a backward integration of the adjoint model. This information is further used to define appropriate weights and to implement a dual-weighted proper orthogonal decomposition (DWPOD) method for order reduction. The use of a weighted ense...

On the Sensitivity Equations of Four-Dimensional Variational (4D-Var) Data Assimilation

Abstract The equations of the forecast sensitivity to observations and to the background estimate in a four-dimensional variational data assimilation system (4D-Var DAS) are derived from the first-order optimality condition in unconstrained minimization. Estimation of the impact of uncertainties in the specification of the error statistics is considered by evaluating the sensitivity to the observation and background error covariance matrices. The information provided by the error covariance sensitivity analysis is used to identify the input components for which improved estimates of the statistical properties of the errors are of most benefit to the analysis and forecast. A close relationship is established between the sensitivities within each input pair data/error covariance such that once the observation and background sensitivities are available the evaluation of the sensitivity to the specification of the corresponding error statistics requires little additional computational effort. The relevance of...

An Adjoint Sensitivity Method for the Adaptive Location of the Observations in Air Quality Modeling

The spatiotemporal distribution of observations plays an essential role in the data assimilation process. An adjoint sensitivity method is applied to the problem of adaptive location of the observational system for a nonlinear transport-chemistry model in the context of 4D variational data assimilation. The method is presented in a general framework and it is shown that in addition to the initial state of the model, sensitivity with respect to emission and deposition rates and certain types of boundary values may be obtained at a minimal additional cost. The adjoint modeling is used to evaluate the influence function and to identify the domain of influence associated with the observations. These essential tools are further used to develop a novel algorithm for targeting observations that takes into account the interaction among observations at different instants in time and spatial locations. Issues related to the case of multiple observations are addressed and it is shown that by using the adjoint modeling an efficient implementation may be achieved. Computational and practical aspects are discussed and this analysis indicates that it is feasible to implement the proposed method for comprehensive air quality models. Numerical experiments performed with a two-dimensional test model show promising results.

An Analysis of a Hybrid Optimization Method for Variational Data Assimilation

In four-dimensional variational data assimilation (4D-Var) an optimal estimate of the initial state of a dynamical system is obtained by solving a large-scale unconstrained minimization problem. The gradient of the cost functional may be efficiently computed using the adjoint modeling, at the expense equivalent to a few forward model integrations; for most practical applications, the evaluation of the Hessian matrix is not feasible due to the large dimension of the discrete state vector. Hybrid methods aim to provide an improved optimization algorithm by dynamically interlacing inexpensive L-BFGS iterations with fast convergent Hessian-free Newton (HFN) iterations. In this paper, a comparative analysis of the performance of a hybrid method vs. L-BFGS and HFN optimization methods is presented in the 4D-Var context. Numerical results presented for a two-dimensional shallow-water model show that the performance of the hybrid method is sensitive to the selection of the method parameters such as the length of the L-BFGS and HFN cycles and the number of inner conjugate gradient iterations during the HFN cycle. Superior performance may be obtained in the hybrid approach with a proper selection of the method parameters. The applicability of the new hybrid method in the framework of operational 4D-Var in terms of computational cost and performance is also discussed.

Adjoint Estimation of the Variation in Model Functional Output due to the Assimilation of Data

Abstract A parametric approach to the adjoint estimation of the variation in model functional output due to the assimilation of data is considered as a tool to analyze and develop observation impact measures. The parametric approach is specialized to a linear analysis scheme and it is used to derive various high-order approximation equations. This framework includes the Kalman filter and incremental three-and four-dimensional variational data assimilation schemes implementing a single outer loop iteration. Distinction is made between Taylor series methods and numerical quadrature methods. The novel quadrature approximations require minimal additional software development and are suitable for testing and implementation at operational numerical weather prediction centers where a data assimilation system (DAS) and the associated adjoint DAS are in place. Their potential use as tools for observation impact estimates needs to be further investigated. Preliminary numerical experiments are provided using the fif...

Computational aspects of chemical data assimilation into atmospheric models

The task of providing aa optimal analysis of the state of the atmosphere requires the development of novel computational tools that facilitate an efficient integeration of observational data into models. In this paper we discuss some of the computational tools developed for the assimilation of chemical data imo atmospheric models. We perform a theoretical analysis of discrete and cominuous adjoints for stiff differential equation solvers. Software tools particularly tailored for direct and adjoint sensitivity analysis of chemical systems are presented. The adjoint of the state-of-the-art model STEM-III is discussed, together with ozone assimilation results for a realistic test problem.

The impact of background error on incomplete observations for 4d-var data assimilation with the FSU GSM

To assess the impact of incomplete observations on the 4D-Var data assimilation, twin experiments were carried out with the dynamical core of the new FSU GSM consisting of a T126L14 global spectral model in a MPI parallel environment. Results and qualitative aspects are presented for incomplete data in the spatial dimension and for incomplete data in time, with and without inclusion of the background term into the cost functional. The importance of the background estimate on the 4D-Var analysis in the presence of small Gaussian errors in incomplete data is also investigated.

Computational challenges of modelling interactions between aerosol and gas phase processes in large‐scale air pollution models

Discusses computational challenges in air quality modelling (as viewed by the authors). The focus of the paper will be on Di, the “current” state‐of‐affairs. Owing to limitation of space the discussion will focus on only a few aspects of air quality modelling: i.e. chemical integration, sensitivity analysis and computational framework, with particular emphasis on aerosol issues.