Computing environments for spatial data analysis

Luc Anselin
2000 Journal of Geographical Systems  
This paper describes the functionality and architecture of SpaceStat, the SpaceStat Extension for ArcView and the DynESDA Extension for ArcView. It compares the features of these packages to five other software implementations for spatial data analysis. Some ideas are formulated on generic requirements and future directions pertaining to computing environments for spatial data analysis. internet (which slows down close coupling due to network overhead), performance comparisons between the two
more » ... rms of coupling are no longer straightforward. An alternative taxonomy was offered in Anselin and Getis (1992) where a distinction is made between encompassing and modular frameworks. The first is simply an extension of the functionality of a GIS with that of a statistical package or vice versa. This is beginning to be reflected in a growing number of commercial products, for example, by the inclusion of mapping and some geostatistical functionality in statistical software such as SAS and Systat, and the extension of the Arcinfo GIS with the forthcoming geostatistical analyst. Paralleling these commercial efforts, customization by academic researchers has involved incorporating a wide range of specialized spatial statistical analyses or other forms of computational modeling that are typically not part of commercial software packages. Usually, this is carried out by relying on built-in scripts or macro commands. There are by now quite a few of such applications, making possible the calculation of global and local spatial autocorrelation indices, the estimation of spatial regressions and fitting of geostatistical models. 1 While these extensions maintain the familiar look-and-feel of the GIS or the statistical software, a drawback of the encompassing approach is that peculiarities of the scripting languages (such as Avenue for ArcView and MapBasic for MapInfo) sometimes preclude the use of the most efficient algorithms or data structures for the statistical computations. Consequently, performance is affected and few implementations can tackle realistic data sets or deliver results sufficiently fast for real-time interactive use. An alternative to the encompassing design is a modular approach, in which a framework of linked systems is constructed, each optimized for a specific functionality, such as statistical analysis, mapping, or user interaction. An growing number of such applications have come to exist as well. An important aspect of a modular design is the handling of communication between the systems, typically following a client-server paradigm, which can be readily extended to a distributed computing environment. Increasingly, such interaction can be encapsulated into software components that are based on object-oriented software design techniques. In principle, such componentization provides the potential for the development of a collection or suite of reusable spatial data analytical software pieces that can be mixed and matched by a researcher to tackle specific problems. This special issue of the Journal of Geographical Systems reports on some recent developments in the efforts to extend the spatial analytical capability of GIS with sophisticated statistical and econometric functionality. The five papers included in the issue approach this question from a different perspective, which illustrates the richness of research and diversity of computing solutions that are being developed. The software solutions range from freestanding programs such
doi:10.1007/pl00011455 fatcat:numi5ah67zgv3nnxf5q6mknq5e