Abstract
The use of exploratory data analysis techniques on spatial data can involve the computation and presentation of measures of global and/or local spatial dependence. A range of global or general measures, including Moran's $I$, Geary's $C$, specified by contiguity defined in a number of ways, and semivariogram statistics, specified by distance and direction, may be calculated. These however need to be completed with observations of local indicators and/or distance statistics (Anselin, 1995; Ord & Getis, 1995; Bao & Herny, 1996). These are also termed focused tests. The presence of global dependence may mask local effects in the spatial data generation process.
This paper will compare and contrast approaches to the construction of exploratory spatial data analysis toolkits, especially SpaceStat 1.80, S-PLUS Spatial Statistics, and the ARC/VIEW - XGobi link (Symanik et al., 1995) with less tightly-bound scripted solutions. The key division runs through public access to source code, permitting not only the prototyping of new measures and statistics in a rapidly changing field, but also the verification of the implementations found in compiled code. All the considered programs - GRASS, GMT, XGobi, CDV and XLisp-Stat are available as source code, and all but GRASS are currently supported by their authors and the international scientific community. Indeed, it is the virtual community of collaborating scholars that is driving the development of leading edge software, as witnessed by the release to these groups of major new computing technologies at a very early stage, even by commercial operations. It seems that the vitality of what could be termed "social movements" in software development is regarded by market players as a key source of insight into product opportunities, current examples being HTTP/HTML and Java.
GRASS is a well-known, now unsupported raster GIS with full source code in the public domain, although its future status is now unclear. Much of the underlying code requires maintenance, and because of this its development potential is perhaps doubtful. Its strength has been in the analysis of category and integer data. GMT is a developing high-quality hard-copy presentation tool especially suited to oceanographers and other earth scientists using geographical coordinates for position registration. It has a range of data manipulation tools, and handles floating point data (Wessel & Smith, 1995). XGobi is a dynamic graphics program for data analysis under continuous development, although available documentation is now dated (Swayne et al., 1991). In addition to linked dynamic views of multivariate data, brushing, and observation identification, XGobi now supports projection pursuit methods for exploratory data analysis. In continuing work, Dykes (1996) is developing CDV - cartographic data visualization - based on Tcl/Tk, an interpreted programming language now supporting Unix/X11, Microsoft Windows, and Macintosh platforms. Finally, the use of XLispStat, a freely available statistics program using XLisp as its command language and running under Unix, Windows and Macintosh among others, in spatial analysis has been illustrated by Brunsdon & Charlton (1996).
The main aim of the paper is to consider questions involved in linking disparate modules of freely available software in order to configure and modify methods for computing, visualizing, and examining local spatial dependence and association. These questions touch on data and metadata models, scripting strategies, portability, transparency, the need for a subset of spatial data handling tools for data mining and pattern seeking, and on the consequences of choices regarding visualization. The principal task areas are data management and journalling, statistics compute system design and execution, visualization and interactive spatial data manipulation, hard-copy presentation, information and help system management, quality control and testing, and verification across different platforms. The paper reports on work in progress linking chosen functional areas of the software mentioned above with other programming utilities. Examples will be given of the use of the tools developed on varying spatial data sets.
In conclusion, the paper will argue that geocomputation ought to concern itself with the implementation of algorithms, not just their elaboration and the purchase of products claiming to include them. For this reason, the overhead of working with source code and scripting in interpreted languages can be accepted where one gains added reliability and transparency in the results achieved.
References
Anselin, L. 1995. "Local indicators of spatial association - LISA", Geographical Analysis, 27, 93-115.
Bao, S. and Henry, M. 1996. "Heterogeneity issues in local measurements of spatial association", Geographical Systems, 3, 1-14.
Brunsdon, C. and Charlton, M. 1996. "Developing an exploratory spatial analysis system in XLisp-Stat". In Parker, D., ed, Innovations in GIS 3, Taylor & Francis, London, 135-145.
Dykes, J. 1996. "Dynamic maps for spatial science: a unified approach to cartographic visualization". In Parker, D., ed, Innovations in GIS 3, Taylor & Francis, London, 177-187.
Ord, J.K. and Getis, A. 1995. "Local spatial autocorrelation statistics: distributional issues and an application", Geographical Analysis, 27, 286-306.
Swayne, D.F, Cook, D. and Buja, A. 1991. "XGobi: interactive dynamic graphics in the X Window system with a link to S", ASA Proceedings of the Section on Statistical Graphics, American Statistical Association, Alexandria VA, 1-8.
Symanik, J., Majure, J.J. and Cook, D. 1995. "Dynamic graphs in a GIS: a bidirectional link between ArcView 2.0 and XGobi", Computing Science and Statistics, 27.
Wessel, P. and Smith, W.H.F. 1995. "New version of the Generic Mapping Tools released", EOS Transactions AGU, 76, 329.