Feature-Based Statistical Analysis of Extreme-Scale Physics Simulation Data

• Activity Number: 217
• Type: Invited
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Graphics
• Author(s): Janine Camille Bennett and Timo Bremer*+
• Companies: Lawrence Livermore National Laboratory and Sandia National Laboratories
• Keywords: topological analysis ; scientific visualization ; extreme-scale computing

Abstract

As high performance computing resources continue to improve, scientists are able to study physical phenomena with unprecedented resolution and complexity. Feature characterization in such massive data is challenging and is further complicated by the fact that phenomenon under study can be characterized by a wide range of physical scales. Historically, statistical analysis has been used to summarize large data in a succinct manner. However, when the data is very large and/or features exist in a wide range of scales, global aggregation is problematic as features of interest may be averaged towards the mean. In this talk I will discuss two shape-based algorithms that can be deployed in complementary settings: local methods support the study of phenomena comprised of many small intermittent features, while global shape methods support the study of large-scale structures. The analyses are presented to scientists in a linked-view browser combining summary plots with interactive views of the features of interest. I will present the algorithms in context with their motivating combustion science case studies, however the methods are applicable to a broad class of physics-based phenomena.