AAG Annual Meeting 2002 - Spatialization Sessions


Session Title: Visualization II - Spatialization Theory and Methods
Session Number: not yet available
Time: not yet available
Chair: Andr
é Skupin
Organizers: Sara Fabrikant and André Skupin
Specialty Group Sponsors: Cartography, GIS, Environmental Perception & Behavioral Geography

Masahiro Takatsuka, GeoVISTA Center, Dept. of Geography, The Pennsylvania State University, University Park, PA 16802. Email: masa@psu.edu
Exploratory analysis of demographic data using the Self-Organizing Map and 3-D visualization tools

The rapid improvement of modern computer technologies has impacted on geographic data analysis and visualization throughout the last couple of decades. Furthermore, there are increasing demands on executing those analyses in real-time. Conventional data analyses - statistical analysis, for example - and visualization techniques are useful but not flexible and powerful enough to meet those demands. The Self-Organizing Map or Kohonen’s Feature Map is one of the many modern data analysis tools that researchers have found useful in analyzing high-dimensional (multivariate) datasets such as atmospherical and demographical data. It is often used to map a high-dimensional feature space onto a 2-D integer space. Moreover, conventional visualization techniques of the Self-Organizing Map require the normalization of the feature space. These two requirements (mapping onto the 2-D integer space and the normalization of the feature space) allow only qualitative analysis.
This study presents the use of the SOM for analyzing demographic datasets with the help of Java-based advanced 3-D visualization tools and a visual programming environment (GeoVISTA Sudio). The Self-Organizing Map is used to map the census data of twenty-six dimensions onto the 2-D integer space, and the distances in the non-normalized feature space are used to create the 3-D visualization of the Self-Organizing Map in the forms of a distance map and the Sammon map. The synthetic terrain representing the state of the Self-Organizing Map is used to gain deeper understanding of the phenomenon of gentrification.

André Skupin, Department of Geography, University of New Orleans, New Orleans, LA 70148. Email: askupin@uno.edu.
A Geographic Interpretation of the Self-Organizing Map

The Self-Organizing Map (SOM) has become one of the more popular forms of artificial neural networks. Geographers are particularly fond of SOM’s ability to perform unsupervised clustering, without some of the restrictions and assumptions associated with more traditional techniques. Geographic research tends to focus on the use of SOM to perform such tasks as unsupervised classification and feature extraction. This paper presents a geographic interpretation of self-organizing maps with particular consideration of (1) properties of the projection technique, (2) geometric characteristics of the output layer, and (3) ways in which the output layer can be manipulated to visualize high-dimensional data in a map-like manner.
After training a SOM one typically obtains a two-dimensional, regular arrangement of neurons. This is akin to raster data models used in GIS. In contrast to raster GIS, hexagonal raster elements have gained a much wider acceptance in SOM implementation. The field-like view of high-dimensional data sets that is invoked by SOM has implications for the technique’s robustness and makes such concepts as scale or resolution applicable. Geometric transformations, overlay operations, and cartographic design techniques can be readily applied to neuron configurations to derive visualizations that can be utilized by human observers much like traditional cartographic representations.

Bruce Rex, John Risch, and Scott Dowson, Pacific Northwest National Laboratory, Richland, WA. Email: db_rex@pnl.gov
The Starlight Approach to Spatialization of Text for Visualization

The Starlight Visual Information Analysis System represents a new class of information system that permits simultaneous analysis of disparate datasets of multimedia data types. One of the most useful and interesting data types are free or structured text. In order to be useful within the visual analytic interface, text must be spatialized for visualization. This paper will present the approach used by the Starlight development team in the development and deployment of the Starlight text engine. The various steps in the process will be examined leading to the final incorporation of the spatialized text into the visual workspace. Examples will be given of how the spatialized text is used to correlate words with geocoordinates in Starlight.

Sara I. Fabrikant, Department of Geography, University of California Santa Barbara, Santa Barbara, CA 93106, Email: sara@geog.ucsb.edu.
Empirical Approaches to Semi-Immersive Spatialization Designs

One of the main challenges in spatialization research is the establishment of a solid theoretical framework based on sound formalisms for successful information visualization designs, and effective graphical user interface implementations. The consequence of a missing theory in information visualization is the lack of a generally accepted methodology to assess and ensure its usability. The goal of this paper is to outline an experimental design for empirically assessing the usability of semi-immersive spatializations of a Reuters news article archive. We investigate how people respond to information spaces that are of larger scale than manipulable object space. The task-based usability evaluations include several types of wall-sized spatialization displays, in 2D and 3D. These spatializations are projected through a backlit screen projection system. The displays were generated based on a combination of scaling methods including the spring-node algorithm type, network scaling based on minimum spanning trees, and a hierarchical clustering technique. The semantic news wire space can be interactively explored in an off-the-shelf Geographic Information System.
The outcomes of the experiments will provide feedback on several levels. First, empirical results will provide the basis for our redesign of the spatializations. It will also provide much needed insight into the usability of such kinds of spaces for information access. Furthermore, we will gain knowledge in the procedures needed to evaluate immersive representations, including the technical issues involved in collecting usable experimental data from VR experiments.

Ferjan Ormeling, Faculty of Geographical Sciences, Utrecht University, Utrecht, The Netherlands. Email: f.ormeling@geog.uu.nl.
Fantasy maps as sources for geographic notions

Maps can be a tool for non-geographic information visualization. Based on the assumption that fantasy maps that do not depict geographical reality are a good source for assessing the strength and usefulness of cartographical representation, both old and current fantasy maps are analyzed for their characteristics, such as scale, distances, regionalization, directions, toponymy, topology, generalization and symbolization. In this context toponymy or geographical names would not refer to the generic indications (such as cemetery, river, farm) but to the names that would distinguish the mapped entities from each other).
Next, the use of these fantasy maps is analyzed, such as their navigational possibilities, and implied goals/destinations. This use will have changed in time, as 18th-century examples were very much concerned with teleological aspects (how to get to heaven, following the steep and narrow path) or more worldly aspects as finding true love, while nowadays this moral dimension seems to be lacking.
The image and message these maps convey or project are analyzed next, as well as the fact that the symbology selected must be in conjunction with the contents. Finally a synthesis of the organizing qualities of cartographical representation for non-geographic information is provided.


Session Title: Visualization III - Spatialization and Geovisualization
Session Number: not yet available
Time: not yet available
Chair: Sara Fabrikant

Organizers: Sara Fabrikant and André Skupin
Specialty Group Sponsors: Cartography, GIS, Environmental Perception & Behavioral Geography

Dan Haug, Redlands, CA 92374. Email: dhaug@esri.com.
Interactive Spatialized Visualization of Hierarchical Clustering

There are a wide variety of hierarchical clustering and classification techniques used by geographers. The results of these techniques are often visualized using a dendrogram. The paper presents an implementation of an interactive dendrogram integrated into GIS software. While this type of tool has been demonstrated in the past (MacDougall 1992), this implementation extends the possibilities of interaction by drawing on thematic representation and interactivity techniques that are common in many GIS-based cartographic software products. By casting the dendrogram as a type of map, we draw on users learned knowledge of map interaction in popular GIS software to create an intuitive interface for exploring hierarchical clustering techniques.

Christopher A. Badurek, Department of Geography, University at Buffalo, Buffalo, NY 14261. Email: badurek@geog.buffalo.edu.
Spatialized User Interfaces for Geospatial Image Collections

Scientists as well as casual web users are currently inundated with large stores of digital information that will certainly continue to increase into the future. As a result, accessing useful information from this mass of data is becoming increasingly more difficult. The techniques of spatialization are reviewed and presented here as an effective approach for retrieving relevant information from large data stores. In particular, the spatialization approach has potential as an intuitive and highly effective interface for digital libraries and web-based image collections. In this context, a model graphical user interface for information retrieval systems for geospatial image collections based on spatial metaphor is presented. It is argued that spatialized user interfaces aid users by helping them form map-like graphical conceptualizations of data collections with which they are able to increase their efficacy in searching.

Etien Koua, Department of Geoinformatics, Cartography and Visualisation, ITC Enschede, The Netherlands. Email: koua@itc.nl.
Self-Organizing Maps in Information Visualization for Geospatial Data

Analysing geo-spatial data can be a difficult task due to large data sets, complexity in data structure, scaling problems and non-revealed patterns. Today, access to geo-information resources such as GIS and Remote sensing data are facilitated through the World Wide Web and the use of these resources is increasing among domain experts but also among common users for their personal information needs. The development of the Internet and related technologies has great impact on the growth of information resources and access. There is need to organise the information space in a way that support users in their information exploration while reducing complexity in information structure and supporting users cognition. Users should be able to look at geospatial data in any combination, at any scale, with the aim of seeing or finding spatial patterns that may be hidden. Information visualisation has the potential to help find the information needed more effectively and intuitively. Like many techniques applied to data analysis (statistical methods, artificial intelligence using rule-based systems and decision trees), neural networks are an emerging solution for data analysis and for pattern recognition. Among the neural network models, Self-Organizing Map (SOM) is a promising technique for exploratory analysis of data.
In our research work, we combine information visualization techniques with SOM to extract information from geospatial data and to generate visual representations of the information space and information processes that will allow users view appropriate underlying distribution and patterns, and therefore contribute to enhancing the understanding of geospatial analysis results. The paper presents the conceptual model and the architecture of the SOM-based information visualization for geospatial data.

Kevin J. Konty, Department of Geography, University of California Santa Barbara, Santa Barbara, CA 93106, Email: konty@geog.ucsb.edu
The Role of Visualization (and Spatialization) in Understanding Multiregional Population Projections and Their Underlying Models.

Theres is a clear need for, and so a substantial interest in, accurate population forecasts. However, as forecasting methods become more sophisticated they also become less transparent, creating blacker boxes while improving forecasts. This paper examines the role visualization and spatialization techniques can play in understanding these methods and there resulting forecasts.
The multiregional modeling approach allows migration to be modeled using region to region flows, creating a network. Although this can be quite simple when the number of regions is small, improvements in the quality of demographic data allow researchers to construct ever smaller regions and to disaggregate the network into non-spatial dimensions such as gender, age, and occupation. They also provide covariates that may explain these population flows and, in so doing, interest planners and policy makers. Thus, there are three levels in which visualization and spatialization techniques may help the population researcher: visualizing the network, understanding covariates, and (fore)seeing the future.

September 7, 2001