Visual Analytics of Ensemble Data Using Coupled Subspaces

In this work, we introduce a novel visual analysis method to explore correlations between locations and realizations of ensemble data with large realization set. In our approach, the ensemble data are transformed into a matrix with two dimensions: location and realization. The matrix is subdivided into cells which indicate behaviors of corresponding location-realization combinations. Biclustering is employed to simultaneously partition the location domain and the realization set, the two dimensions of the matrix, into several subspaces, where every intersection subspace shares similar coordinated behaviors. With the matrix subspaces and their coordinated behaviors, a visual analytics workflow is designed to support visualization and analysis of location-realization correlations and their variation across different location and realization. Case studies show that our system is able to reveal location-realization correlations, such as the opposite behaviors of realization subspaces in different regions, which are normally difficult to be discovered by previous methods.