ITR: A Computational Framework for Observational Science: Data Assimilation Methods and their Application for Understanding North Atlantic Zooplankton Dynamics.

该项目将开发一个模块化的数据同化系统，研究几种算法使数据同化更有效，并将应用该系统来研究北大西洋的浮游动物动态。数据同化的目标是找到控制变量（通常是初始条件或边界条件或模型参数）的值，在模型和数据之间产生最佳的一致性。数据同化系统由表示已知动力学的正演模型组成。该模型是综合的，其预测和现有观测之间的偏差是量化的成本函数。然后运行一个表示已知动力学逆的伴随模型，以确定成本函数对控制变量的依赖关系。根据伴随模型的结果，调整控制变量，重复整个过程，直到系统收敛到一个答案。由于正演/伴随系统需要多次迭代才能找到答案，因此数据同化是一个计算密集型的过程。提出的数据同化系统将尝试通过并行化和算法改进来提高效率。具体而言，本项目将评估三种标准最小化算法和一种基于多网格技术的新算法。使用这个系统，来自连续浮游生物记录调查的数据，唯一正在进行的全流域浮游生物调查，将被同化，以提供一个准确的，定量的描述北大西洋浮游动物种群的季节性和年际变化（特别是在缅因湾和整个北大西洋的Calanus finmarchicus）。这种描述将提供对这些种群中观察到的模式的过程的更好的机制理解。这种理解是预测气候变率和变化对浮游动物种群及其所支持的生态系统的影响的先决条件。更广泛的影响：提出的数据同化系统是许多数据同化问题的通用模式，包括业务海洋学和数值天气预报。该项目与康奈尔理论中心（CTC）的合作提供了一个独特的机会，可以向广泛的受众分享其数据同化系统。在CTC工作人员的帮助下，一个web界面的系统运行在CTC的。NET集群将被构建。这个接口将允许世界各地的研究人员和学生访问一个高性能的数据同化系统。数据同化系统的开发将被整合到康奈尔大学提供的一系列计算工具课程中。该项目还将为研究生和本科生提供研究机会。

This project will develop a modular data assimilation system, investigate several algorithms to make data assimilation more efficient, and will apply this system to investigate zooplankton dynamics in the North Atlantic. The goal of data assimilation is to find the value of the control variables (typically, the initial conditions or boundary conditions or model parameters) producing the best agreement between the model and the data. A data assimilation system consists of a forward model representing known dynamics. This model is integrated and the deviation between its predictions and available observations are quantified by a cost function. An adjoint model, representing the inverse of the known dynamics, is then run to determine the dependence of the cost function on the control variables. From the results of the adjoint model, the control variables are adjusted and the entire procedure repeats until the system converges on an answer. Because of the many iterations of the forward/adjoint system are required to find an answer, data assimilation is a computationally intensive process. The proposed data assimilation system will attempt to improve the effciency through parallelization and algorithmic improvements. Specifically, this project will evaluate three standard minimization algorithms and a new algorithm based on multigrid techniques. Using this system, data from the Continous Plankton Recorder survey, the only ongoing basin-wide plankton survey, will be assimilated to provide an accurate, quantitative description of the seasonal and interannual changes of North Atlantic zooplankton populations (especially, Calanus finmarchicus) in the Gulf of Maine and across the entire North Atlantic. This description will provide a better mechanistic understanding of the processes responsible for observed patterns in these populations. Such an understanding is prerequisite for predicting the impact of climate variability and change on zooplankton populations and the ecosystems they support.Broader Impacts: The proposed data assimilation system is a general model for many data assimilation problems including operational oceanography and numerical weather prediction. This project's association with the Cornell Theory Center (CTC) allows a unique opportunity to share its data assimilation system to a wide audience. With the help of CTC staff, a web interface to the system running on CTC's .NET cluster will be built. This interface will allow researchers and students across the world to access a high-performance data assimilation system. The development of the data assimilation system will be integrated into a series of computational tools courses offered at Cornell. This project will also provide research opportunities for both graduate students and undergraduates.