Visualizing gridded data sets with large number of missing values

Research in visualization tends to focus on two distinct cases of data input: gridded or meshed data and sparse, scattered data with no underlying structure present. Consequently, visualization techniques fail to adequately address the case of datasets inheriting a grid structure, but with most of the data values missing or invalid. The proposed work strives to bridge the gap between these two ends of the spectrum.; This research was motivated by a need to create an adequate rendering for data acquired from NEXRAD—a 3D weather radar which usually has no more than 3 to 4% of all possible values filled. Such datasets are abundant in atmospheric, geological and many other physical sciences. First, it is shown that ignoring these special cases and simply filling empty slots with arbitrary out-of-range values will produce invalid visualizations. We then explore the alternatives: scattered visualization, interpolation and gridded data methods. While providing an analysis of the first two, our research concentrates on addressing and correcting the shortcomings of the latter, and efforts to improve it. In particular, we discuss changing Isosurface and Volume Rendering algorithms to adjust to the specifics of handling cells with one or more vertex values missing. We also provide a new post-processing technique to smooth out the peaks and valleys found in high-frequency datasets which is shown to improve the visual appearance of the surface over the standard method.; We then shift our concentration to cover the specifics of visualizing uncertainty as a side-effect of interpolation and modeling algorithms. The primary goal of this effort is to provide options for representing the error, uncertainty or standard deviation values alongside the primary value in Isocontour, Isosurface and Volume Rendering images. The suite of methods we present provide an intuitive understanding of where regions of high error by using discontinuity and texture.; We test our methods on a variety of scientific datasets, some of which have either random or range data eliminated in order to simulate missing data. Synthetic gridded datasets are also created in order to simulate different levels of sparseness and granularity of data.