A common body of knowledge of point pattern analysis exists in the geographical, forestry and other sciences. Nevertheless, structural point pattern analysis is quite limited in geographic information systems. Present commercial GIS use elaborate techniques for spatial operations like buffering objects or overlaying different thematic layers. Concerning statistical spatial analysis of point objects (not to be confounded with point measurements), their capabilities are limited to simple descriptive measures such as minimum, maximum, mean and standard deviation. Some raster-based systems (e.g., IDRISI) offer more complex statistical analysis functions (e.g. measures of spatial autocorrelation), but they do not offer sophisticated algorithms for point pattern analysis.
Several authors (Goodchild et al., 1991; Openshaw, 1991) discuss possible ways to link GIS with statistical spatial analysis. Five strategies may be distinguished:
The second strategy is applied in the experimental system SPLANCS by Rowlingson and Diggle (1993). They made some enhancements to the S-Plus system to produce a tool for display and analysis of spatial point pattern data. G-, F- and K-functions as well as a kernel smoothing procedure were implemented. The advantage of having the full statistical capabilities of S-Plus available is achieved at the expense of having no real GIS functionality. This approach, although criticized by Openshaw (1991), has also been adopted by Griffith (in Haining and Wise, 1991) and SAS Inc. producing a module SAS/GIS.
Several developments try to couple GIS with commercially available statistical packages such as GLIM, SAS, SPSS, generally using ASCII exports. They mostly concentrate on measures of spatial autocorrelation and association (Gatrell and Rowlingson, 1994). Others implemented methods for point pattern analysis concentrating on first- and second-order analysis (MacLennan, 1991; Rowlingson and Diggle, 1993; Gatrell and Rowlingson, 1994); they occasionally include some density estimation techniques (e.g., kernel density estimation). MacLennan (1991) implemented second-order analysis methods (G- and L-functions) into the GRASS system (GRASS, 1993). Analysis of time aspects has rarely been considered, partly because of the lack of infrastructure for temporal geographic information systems and the absence of established analysis methods.
Openshaw (1994) presents a new analysis approach by extending his earlier work to 'geographical analysis machines' GAM (Openshaw, 1987). He argues that traditional exploratory methods of pattern discovery are not feasible in a multivariate GIS environment with tri-space (geographical, temporal and attribute) information. He extends his GAM to 'space-time-attribute-machines' (STAM) and '-creatures' (STAC), using artificial life (see Langton, 1989; Beer, 1990) to search all three spaces (Openshaw and Perrée, 1996). Basically, a STAM is an automatic screening program which is searching all the locations in geography, time and attribute space for evidence of clustering.
Most of the above mentioned methods are intended for use with static data. This is not very astonishing when considering that their background is strongly connected to botanical data. Spatial movements are seldom of interest in this case. In the few examples where time is included explicitly in the methods (Cressie, 1993; Openshaw and Perrée, 1996) the meaning of patterns to be searched for is reduced to comparing the clustering of points to complete spatio-temporal randomness.
The only approach available for handling time with spatial data from animal locations is the aggregation of observations into time slices 3.1, followed by the application of static analysis methods. Secondly almost all of these methods including the point process theory (for an introduction see Diggle (1983) or Cressie (1993)) are only of marginal interest here, as they are not dealing with mobile objects.