Exploration of spatial flow patterns using projection pursuit methods and dynamic visualization

The objective of this research is to develop and implement a new methodology to facilitate the exploration of spatial flow patterns in complex interregional flow data sets containing multiple flow variables. These flow patterns can be used to aid researchers in generation of new hypotheses. The proposed methodology integrates projection pursuit (PP), scientific visualization (SV), exploratory data analysis (EDA), and dynamic graphics (DG). This methodology consists of two stages: complexity reduction by PP methods and exploration through dynamic visualization. PP methods map a multi-dimensional data set onto a smaller dimension, in which visual inspection can reveal structures such as clusters and outliers. In this research, a moment-based PP method is implemented to reduce the number of flow variables in a complex flow data set. The new flow variables generated by the PP method are used to explore spatial flow patterns in an interactive environment which integrates SV, EDA and DG.;The methodology has been implemented in prototype software PPFLOW, which stands for Projection Pursuit FLOW Data Analysis System. PPFLOW has been developed on a Silicon Graphics workstation, written in C, with the Xlib and Motif supporting libraries. It uses four graphic views: a spatial flow map, a choropleth map of origin and destination (O-D) regions, a statistical plot of flow variables, and a statistical plot of O-D attribute variables, to graphically represent a flow data set. The flow map is underlaid graphically by the choropleth map that represents regional characteristics; thus displaying the relationship between spatial flows and regional characteristics. One-to-one, one-to-many, and many-to-many flows can be displayed on the flow map. In addition, the content of the flow map can be adjusted by setting a distance range or a flow magnitude range. A dynamic brushing technique has been implemented to link the four views; therefore revealing the relationships between the data displayed in the four views.;Although the methodology presented here has been developed within the context of spatial flow analysis, its basic principle of complexity reduction, visualization and exploration can be extended to general geographic research directed toward hypothesis generation. Dynamic linking between graphic representations of both spatial data and aspatial data is an effective approach for revealing the relationships between and within spatial and aspatial data; such relationships may lead to the generation of hypotheses.