APRIL 18-21, 2017
Previous Events

Decoupling Ensemble Data Using Subspaces

  • Fan Hong
    Peking University, Beijing, China
  • Xiaoru Yuan
    Peking University, Beijing, China
  • Alex Pang
    University of California, Santa Cruz, California, United States
  • Chufan Lai
    Peking University, Beijing, China
  • Brad Hollister
    University of Utah, Salt Lake City, Utah, United States
  • Xiaoguang Ma
    State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing, China
  • Pierre Lermusiaux
    Massachusetts Institute of Technology, Cambridge, Massachusetts, United States


Locations and simulations runs are two important aspects of ensemble data. Existing methods either compare regional or locational features extracted from runs, or compare simulation runs over the spatial domain. By decoupling only one aspect and aggregating the other, the coordinated behaviors between run subsets and location subsets are concealed. In this work, we introduce the concept of ensemble subspace inspired by high-dimensional data visualization, and suggest a more informative approach to explore ensemble data, that is to simultaneously decouple the locations and runs. Biclustering-based method is employed for ensemble subspace extraction, thus enables the exploration of coordinated behaviors in ensemble data. We build a system to support extraction and exploration of coordinated behaviors in ensemble subspaces. Our case studies have shown plenty of results about coordinated behaviors, which are difficult to find with existing tools.