| Abstract |
There is an urgent need in scientific communities, driven by their
ability to generate ever-larger, increasingly complex data, for
scalable analysis methods that rapidly identify salient trends in
scientific data. Query-Driven Visualization (QDV) methods are among
the small subset of techniques that are able to address both large and
highly complex datasets---e.g.\ multivariate, multitemporal, and
multiresolution representations of scalar, vector, and function field
data. This dissertation presents new methods that either directly
extend the utility and accelerate the performance of QDV as a whole,
or enable QDV's substantial and flexible analysis strengths to be
applied to new areas of scientific research.
The first part of this dissertation presents a new data-parallel
strategy that accelerates the most fundamental task performed by QDV:
the evaluation of user defined, ad~hoc queries. The second part of
this dissertation extends QDV strategies to analyze and visualize
time-varying adaptive mesh refinement (AMR) data. AMR techniques are
used in many scientific communities to efficiently and accurately
model complex, continuous physical phenomena. By extending QDV
methods to address the dynamic spatiotemporal properties of
time-varying AMR data, I provide scientists with a powerful tool for
visually analyzing the data generated from these important
simulations. The final part of this dissertation leverages statistical
analysis methods to generate deeper insight into the regions that are
selected by a user's query. In this effort I introduce two new
methods that increase the utility of query-driven strategies. The
first strategy uses correlation fields, created between pairs of
variables, in conjunction with the cumulative distribution functions
(CDF) of variables expressed in a user's query. This strategy
identifies important variable interactions within query regions. The
second strategy forms a statistical-based segmentation within the
query-region to generate deeper insight into the ``statistical
structure'' of a user's query. In this approach, segments indicate
which variable contributes most to the underlying joint density
distribution of the user's query. These segments, when used in
conjunction with each variable's CDF, intuitively aid users in
refining the constraints over the variables in their query.
|
|---|
