Multiresolution- and Topology-Based Visualization of Large Scientific Data Sets in Parallel and Distributed Computing Environments

我们生成科学数据的能力比我们理解数据的能力增长得快得多。计算机和成像技术的快速发展使科学家能够生成非常大的数据集，而对这些数据集进行深入分析和理解的适当手段尚不存在。计算技术允许以不断增加的空间和时间分辨率模拟极其复杂的物理现象，但这类数据的数据分析技术仍处于起步阶段。今天，人们必须在这些选择中做出选择：人们可以忽略大量科学数据集的很大一部分，只分析其中的一部分，或者可以投入大量的人力时间来详细分析和可视化大量数据集。两种选择都不可取。该项目将开发解决这一问题所需的技术，并将在劳伦斯利弗莫尔国家实验室和美国宇航局艾姆斯研究中心的数据上测试新技术。从技术上讲，该项目将采用5个方面的方法来解决大数据可视化问题。首先，它将扩展现有的分层方案——即近似于多个分辨率水平的数据集的方案——到时变的多值/多维数据。其次，它将改进基于拓扑的方法——即从大型数据集中提取质量上有趣的特征（如标量和向量场中的零、极值和不连续）的方法。第三，它将开发并行和分布式算法，以有效地计算分层数据表示，快速提取拓扑，并优化计算密集型可视化过程。第四，它将为（沉浸式）可视化环境设计交互式可视化技术，支持依赖于视图和用户指定的自适应细节级渲染。第五，它将创建一个简单的元数据数据库系统，允许共享用户体验，即先前选择的渲染参数导致“好图像”或整个渲染过程。

Our ability to generate scientific data is growing much faster than our ability to understand it. The rapid advances in computing and imaging technology allow scientists to generate very large data sets for which appropriate means for in-depth analysis and understanding do not yet exist. Computational techniques allow the simulation of extremely complicated physical phenomena at ever increasing spatial and temporal resolutions, but data analysis technology for this type of data is still in its infancy. Today, one must choose one of these alternatives: One can either ignore a substantial fraction of a massive scientific data set and only analyze portions of it, or one can invest a significant amount of person-time to analyze and visualize a massive data set in great detail. Neither alternative is desirable. This project will develop the technology needed to address this issue, and will test the new techniques on data from Lawrence Livermore National Laboratory and NASA Ames Research Center.Technically, the project will take a 5-prong approach to the large data visualization problem. First, it will extend existing hierarchical schemes - i.e., schemes approximating a data set at multiple resolution levels - to time-varying multi-valued/multi-dimensional data. Second, it will improve topology-based approaches - i.e., approaches that extract qualitatively interesting characteristics (such as zeros, extreme, and discontinuities in scalar and vector fields) from large data sets. Third, it will develop parallel and distributed algorithms for efficient computation of hierarchical data representations, fast extraction of topology, and optimizing compute-intensive visualization processes. Fourth, it will devise interactive visualization techniques for (immersive) visualization environments that support view-dependent and user-specified, adaptive level-of-detail rendering. Fifth, it will create a simple metadata database system allow sharing of a user's experience, i.e., previously chosen rendering parameters leading to "good imagery" or entire rendering processes.