DDDAS-TMRP: Planet-in-a-Bottle: A Numerical Fluid-Laboratory System

该项目将建立一个“数值流体实验室系统”，以加强对地球天气和气候的了解，这在很大程度上取决于准确的预报和状态估计技术。这项研究工作的目的是设计和建立一个实验室规模的DDDAS，称为瓶内行星，作为一个实际和廉价的一步，向行星规模的DDDAS。 瓶子里的行星DDDAS将模拟气象和海洋状态估计和预测的许多大规模挑战，但提供一个受控的环境，允许采用系统的工程策略来设计更有效和准确的技术。瓶中行星DDDAS将由两个相互作用的部分组成：流体实验室实验和数值模拟器。该系统将采用数据同化，将实际观测结果输入模拟器，以保持模型与现实保持一致，并将采用灵敏度驱动的观测和网格细化，其中模拟器将传感器的实时部署目标定位到特定的地理区域和时间，以实现最大效果，并细化网格，以更好地预测流体实验的未来进程。此外，如果需要，观测系统和网格细化之间的反馈回路将通过人为控制来调节。该项目将研究伴随方法，以确定如何以及在何处部署观测，以及如何以及在何处细化模拟。该实验室将通过非侵入性地可视化和分析自然流体的三维行为来提供对非线性流体动力学的见解。将探讨有关混沌系统的可预测性和适应模型和观测动态预测能力的影响的基本问题。为了设计一个有效的瓶中行星DDDAS，需要在先进的计算技术上进行大量投资，因为在规则网格上进行简单的模拟比流体的实时演化需要更长的时间。为了提高性能，该研究项目将设计新的算法和软件，利用低成本的商品处理器集群。具体来说，研究人员将研究基于分解树理论的不规则网格的内存布局策略，这将导致算法可以有效地利用内存层次和并行性。为了研究如何简化这种复杂算法的编程，他们将研究和开发一个分布式事务内存库，以在集群编程环境中集成共享和同步。开发这些计算机科学技术将需要算法和计算机系统领域的大量技术专长。

This project will create a "Numerical Fluid-Laboratory System" to enable enhancing the understanding of the Earth's weather and climate, which depend critically on accurate forecasting and state-estimation technology. This research effort aims to design and build a laboratory-scale DDDAS, called Planet-in-a-Bottle, as a practical and inexpensive step toward a planet-scale DDDAS. The Planet-in-a-Bottle DDDAS will emulate many of the large-scale challenges of meteorological and oceanographic state-estimation and forecasting but provide a controlled setting to allow systematic engineering strategies to be employed to devise more efficient and accurate techniques. The Planet-in-a-Bottle DDDAS will consist of two interacting parts: a fluid lab experiment and a numerical simulator. The system will employ data assimilation in which actual observations are fed into the simulator to keep the models on track with reality, and will employ sensitivity-driven observations and mesh refinement in which the simulator targets the real-time deployment of sensors to particular geographical regions and times for maximal effect, and refines the mesh to better predict the future course of the fluid experiment. In addition, the feedback loop between targeting of both the observational system and mesh refinement will be mediated, if desired, by human control.The project will investigate adjoint methods to determine how and where to deploy observations, as well as how and where to refine the simulation. The laboratory will provide insights into nonlinear fluid dynamics by visualizing and analyzing the three-dimensional behavior of a natural fluid noninvasively. Fundamental questions regarding the predictability of chaotic systems and the impact of adapting models and observations dynamically on the predictive capabilities will be explored. To design an effective Planet-in-a-Bottle DDDAS requires a substantial investment in advanced computing technology, because a naive simulation on regular meshes take far longer than the real-time evolution of the fluid. To enhance performance, the research project will devise novel algorithms and software that exploit low-cost clusters of commodity processors. Specifically, the researchers will investigate memory layout strategies for irregular meshes based on the theory of decomposition trees, which will lead to algorithms that can effectively exploit both memory hierarchy and parallelism. To investigate how the programming of such complex algorithms can be simplified, they shall investigate and develop a distributed transactional memory library to integrate sharing and synchronization in a cluster programming environment. Developing these computer-science technologies will require substantial technical expertise in the areas of algorithms and computer systems.