Incremental 4-D Variational Data Assimilation, Efficient Optimization and Parameter Estimation Techniques

9731472 Navon该项目的总体目标是测试和应用四维变分数据同化的新方法和数学方法，利用全球预测模型的全物理版本的伴随物的可用性来提高动态同化过程中物理过程的计算效率和表示。纳文博士将采用克里希那穆提博士开发的佛罗里达州立大学全球光谱模型（FSU GSM）作为研究的主要模型。本研究的三个主要重点是：(1)针对建模界认为对模型预测性能和物理初始化有较大影响的关键参数，开发更有效的参数估计方法；(2)评价了两种计算效率高的逼近全4维变分过程的方法——增量法和多次截断增量法的有效性；(3)对一种新的大规模无约束最小化算法——带记忆的离散截断牛顿法进行了测试。最后一项将与ECMWF的Fisher博士和OPTEAM， Ltd.的Berger博士合作完成。高效和有效的四维资料同化是努力改进数值天气预报的关键组成部分。这项研究在推进这一努力方面具有相当大的潜力。

9731472 Navon The overall objective of this project is to test and apply new approaches and mathematical methods for four-dimensional variational data assimilation, taking advantage of the availability of the adjoints of the full physics versions of global forecast models to improve the computational efficiency and representation of physical processes in the dynamical assimilation process. Dr. Navon will employ the Florida State University Global Spectral Model (FSU GSM), developed by Dr. Krishnamurti, as the primary model for the studies. The three major thrusts of this research are: (1) to develop more efficient methods for parameter estimations for key parameters identified by the modeling community as having a sizeable impact on model forecasting performance and physical initialization; (2) to evaluate the effectiveness of two computationally efficient methods for approximating the full 4-D variational process - the incremental approach and the multiple truncation incremental approach; and (3) to test the Discrete Truncated Newton method with memory, a new large-scale unconstrained minimization algorithm. The last will be done in collaboration with Dr. Fisher of ECMWF and Dr. Berger of OPTEAM, Ltd. Efficient and effective four-dimensional data assimilation is a critical component in the effort to improve numerical weather forecasting. This research has considerable potential for advancing that effort.