Dianne Cook
Department of Statistics, Iowa State University, Ames
Position Statement
Curriculum Vitae
Address

Position Statement

Visualizing Multivariate Spatial Data

There are numerous software tools that provide excellent computing systems for different aspects of spatial analysis. Taking advantage of existing expertise allows us to concentrate on developing methods and software for the unique aspects of spatially referenced data. We will describe the components where we believe there is mature expertise, and how we make use of these to provide an integrated analysis environment.

The geographic information system (GIS) provides an "anchor" of good database tools, and map drawing facilties, for spatially referenced data. We use this to select, or do simple manipulation of the data, and provide sophisticated maps of the spatial domain. The GIS is important also for maintaining the frame of reference to the data - a good map provides context for the data, which can otherwise get lost in statistical modeling and graphics.

A statistical analysis system (for example, S-Plus, SAS, XploRe, XLispStat, DataDesk) is used for modeling the trends and spatial dependence.

A visualization system (eg XGobi, DataDesk, XLispStat, XmdvTool, cdv) provides quick exploratory analysis and diagnostic checking for the model. The graphics need to be interactive, with several facilties for linked brushing, and dynamic to rotate the data through high-dimensional space. We should be able to quickly examine spatial dependence plots, models and residuals, as well as the multiple raw variables. We have a fairly broad variety of tools for extracting patterns in multivariate data. There needs to be a lot more research on the types of plots that can extract multivariate spatial trends and dependence. There are variogram clouds and cross-variogram clouds, which give information on individual and pairwise spatial dependence. But pairwise analysis of high-dimensional data is inadequate, so we suspect we will find that pairwise analysis of spatial dependence will be inadequate. So we need to devise new approaches to visually exploring spatial dependence amongst several variables.

All three systems need to be seamlessly linked. The scatterplot needs to be linked to the GIS map - brushing in the scatterplot should instantaneously update the map view and brushing in the map view should instantaneously update the scatterplot. We should be able to display the model overlaid on the geographic domain, and toggle between the model and residual surface. We should be able to calculate local statistics and make plots of these linked to the map.

To find out more information about what we have done related to this position and references click here.

 

Curriculum Vitae

 Dr Dianne Cook is an Associate Professor at Iowa State University in the Department of Statistics and Statistical Laboratory.  Her research direction is primarily on the visualization of high-dimensional data which is related more generally to research in dynamic graphics, exploratory data analysis, data mining, multivariate methods, and statistical computing. Data that can be visualized using her research arise in many situations: environmental monitoring, experimental physics, stock markets, biological species classification, mathematics, for example. She is an author of the popular publicly available visualization software XGobi, used for visualizing high-dimensional data. She has also been experimenting with the use of highly immersive virtual reality environments for visualizing high-dimensional spatially referenced data.

Address

Dianne Cook
Department of Statistics
Iowa State University
325 Snedecor Hall
Ames, IA 50011-1210
Telephone: (515) 294 8865
Fax: (515) 294 4040
Email: dicook@iastate.edu
http://www.public.iastate.edu/~dicook/

Go back to list