多属性时变体数据可视化

Datasets produced by scientific simulations reach unprecedented large scale in time and space, and the number of properties in simulations is also increasing significantly. These multivariate time-varying volume datasets provide an effective analysis and verification way to reveal complex natural phenomena. Although the research in time-varying datasets and multivariate datasets have made some progresses, there still exists several unsolved problems in multivariate time-varying volume visualization: spatiotemporal relationship of multiple properties, feature extraction and tracking based on multiple properties, data-driven parameter space analysis of the simulation model, and modeling complex simulation phenomena, et al.. With an aim of gaining insights into multivariate time-varying datasets, this project propose a set of analysis and visualization algorithms by studying the relationship of multiple properties, spatiotemporal feature extraction and tracking, spatiotemporal differences in multivariate time-varying ensembles, and evaluation patterns of complex events. We plan to solve these key problems in these aspects: analyze the spatiotemporal relationship of multiple properties and extract features based on multiple properties; compare simulation ensembles based on the individual and joint variances; display complex events with a river-flow-based visual metaphor; generate animations for feature evolution of complex events. This project will also design and implement a multivariate time-varying volume visualization prototype system based on above techniques.

仿真模拟产生的数据在时间和空间上达到了前所未有的量级，并且时空维度上对应的属性数据也不断增加。这些多属性时变体数据为揭示复杂现象提供了有效的分析验证和认知手段。单属性时变和多属性静态数据的可视化研究已取得一些进展，但面对多属性时变体数据，仍存在复杂现象与属性及数值关联分析、多属性时变特征识别和跟踪、仿真模拟中复杂事件的刻画与展示等问题。. 本项目针对多属性时变体数据，通过研究属性及数值关联分析、基于多属性的特征识别和跟踪、多变量时变组集数据的差异分析、时变特征的演化规律展示等，为仿真模拟提供多属性时变体数据可视化算法基础。拟解决属性及数值的关联分析、基于多属性数据的时变特征识别和跟踪技术、基于多属性数据特征的组集间个体和群体差异性分析、基于河流隐喻的时变特征可视化、针对事件的动画生成技术等关键问题，在此基础上设计并实现一个多属性时变体数据可视化原型系统。

时变多变量数据产生于诸多领域的科学计算中，例如计算流体力学、燃烧模拟、气象模拟、宇宙学模拟等等。这些体数据中包含着大量特征，特征之间往往存在层次嵌套、相似性和差异性等复杂的关系，结合所有特征及特征之间的关系才能对模拟仿真过程有全面的认识。然而目前已有的方法存在一些问题：仅针对数值或空间方面度量数据的相关性；体数据规模庞大，导致计算量太大；相关性错综复杂，无法有效捕捉；特征之间的相似度难以量化等。本课题围绕以上问题，分别从多变量特征空间相关性、多变量子空间相关性、时变演化追踪可视化、可视分析领域中研究并解决以上问题。