[dbis logo]

.News.Detail
[Institut fuer Informatik] [Leerraum] [Humboldt-Universitaet zu Berlin]
05.09.2012 22:00 Alter: 7 yrs

New Publication (Poster): "Visual Analytics Approach for the Assessment of Simulation Model Output" (DAMES 2012)

Von: Matthias J. Sax


The poster is joined work with Deutsches Geo Forschungs Zentrum (GFZ) Potsdam and will be presented at the 3rd international conference on "Data analysis and modeling in earth sciences (DAMES)" at Potsdam Institute for Climate Impact Research from 8th-10th October 2012.

Abstract:

Environmental routinely utilize simulation models to study or predict the behavior of environmental systems and processes such as ecosystems, atmospheric circulation, ocean circulation, or surface runoff. To gain insight into the non-trivial behavior of simulated systems, environmental scientists try to find a description of the spatial and temporal behavior based on a limited number of prominent spatial patterns that capture the characteristic states of the environmental system. However, the analysis of environmental modeling output is a challenging task for two reasons: (1) it requires a combined assessment of the data’s spatial and temporal dimensions and (2) environmental simulation models often produce gigabytes of raw data. A typical data set (i.e., the result of the simulation) has 3000 time series in which each time step is a grid with about 800x1000 grid points, resulting in about 600GB of raw data.

To tackle these two problems, we present a Visual Analytics approach that facilitates progressive in-depth assessment of patterns in environmental time series based on an efficient computation of a hierarchy of spatial clusters and the visual exploration of these clusters in their spatiotemporal context to support scientist to decide their importance. To enable a high performance clustering, we utilize the "Pyramid Match Kernel" method originally developed for clustering images. The algorithm extracts spatial feature histograms for each grid which significantly reduces the data size. In a second step, the algorithm computes a pair-wise distance among the feature histograms to cluster grids with similar spatial features. We will report some initial results on the implementation of the algorithm using the parallel data analytics engine Stratosphere, the visual exploration capabilities of our system, and some initial findings of our collaborations with environmental scientists using our system.