Scientific visualization has become an indispensable tool for visual analysis of data generated from simulations and experiments across a wide variety of fields. Although there have been substantial advances in developing novel algorithms and techniques for processing, managing and rendering scientific datasets, several critical challenges still remain. These challenges include solving the inherent occlusion and clutter problem when visualizing large three-dimensional scalar and vector fields, examining complex data relationships and tracking their changes over time for time-varying multivariate data, and gaining a comprehensive overview and acquiring full control of data navigation to glean critical insights. The ever-growing size and complexity of data produced only exacerbate these challenges. To enable discovery from big scientific data, there is a need to seek a new perspective on data abstraction and relationship exploration by going beyond the traditional boundary of scientific visualization and fully incorporating information visualization techniques for effective visual data analytics. While there are encouraging, isolated examples of applying information visualization techniques such as parallel coordinates and treemaps to scientific data analysis, leveraging the more generalized and familiar form of graphs to address a wider range of scientific visualization problems at greater extent has not been fully studied. In this project, the PIs' goal is to establish systematic graph-based techniques to investigate large-scale scalar and vector scientific datasets. To this end the team pursues three major tasks: (1) exploring core graph-based techniques to analyze and explore time-varying multivariate scalar and vector field data; (2) developing scalable parallel algorithms for constructing and visualizing large graphs for scientific visualization; and (3) conducting a formal user study using the choice behavior model and tackling real problems from application domains with expert evaluation.
Because scientific visualization plays a key role in many scientific, engineering and medical fields, the potential benefits from generalized graph-based visual analytics tools are far reaching. The general ideas developed will directly benefit the understanding of volumetric scientific datasets including scalar and vector field data, time-varying and multivariate data. They will also impact the understanding of data in other forms such as adaptive mesh refinement, unstructured grid and point-based data. Since this work is a departure from traditional approaches, it could be transformative by providing a completely new way of exploring and analyzing big scientific data. From a scientific perspective, the potential impact is a new class of techniques for knowledge discovery. This project will maximize its outcomes through close collaboration with combustion and biomedical scientists. It will produce results in various forms which are publicized at the project website (www.cs.mtu.edu/~chaoliw/nsf13-graph.htm). The project provides training for graduate, undergraduate and under-represented students in big data computing and visualization. Public outreach activities are planned, including summer programs for middle and high school students, and tutorials or contests for researchers at premier visualization conferences.
