3 
rmation of 
provides a 
y years to 
the systems seems to be that the vector based GIS systems do not have procedures for unsupervised 
classification of remotely sensed data and the remote sensing systems do not support full polygon processing. 
GIS software has been extended to include a wide range of image scanning, vectorization and registration 
procedures. No system appears to fully support temporal data analysis the way that Ehlers envisioned. 
While there may not be any single software system that meets the definition of a total integration 
that Ehlers outlined in the 1989 this is not to suggest that their has not been considerable progress over the 
d into [an] 
tion as yet 
;ence" and 
past six years. For example there has been considerable advancement in terms of soft copy photogrammetric 
tools in remote sensing systems ( Jensen, 1995). In many ways the advancements have been in areas that 
were difficult to foresee in 1989. The most important developments have been the overall evolution of the 
personal computer processors, storage technology, operating systems, and graphic input and display 
technology . Both GIS and remote sensing technology have been able to capitalize on these developments and 
f when we 
inity tends 
i as useful 
the overall expansion of the consumer computer market. In fact, as a result of these developments even 
today’s home entertainment level computers would outperform most of the standard UNIX workstations of 
six years ago. As a result, GIS and remote sensing technology is available to a much wider audience than 
it was in the last decade. Low cost mapping systems are now so prevalent that they are available through 
neighborhood computer stores and mail order outlets. There even seems to be an emerging market for 
mapping ‘shelfware” that is so inexpensive that the customer does not mind whether it is ever used 
productively. The next phase of this evolution is likely to be the widespread conversion of mainstream GIS 
:e the state 
art at that 
^stems. In 
version of 
88b). The 
ystem and 
Dhisticated 
*e. 
and remote sensing systems from the UNIX operating system to the Windows NT operating system that can 
run on the standard Intel based computers. This implies that by the end of the decade we are likely to see a 
new breed of GIS and remote sensing systems that are not only inexpensive but run applications that are 
dynamically linked and very closely coupled. These spatial data handling operations are also likely to be 
dynamically linked to applets that retrieve data and perform functions through the Internet. 
A very important advancement in terms of increased usability of remote sensing and GIS technology 
has been the development of programming tools that allow a developer to customize the interface and hide 
complex operations from the end user. An example of this type of system for environmental applications was 
ivironment 
*e of other 
lser would 
aware that 
is the GIS 
itial query 
nmercially 
developed for the Savannah River Site (Cowen, et al., 1995). These new systems overcome one of the 
problems identified by Ehlers et al. who saw that commercial “...lack the kind of extensibility through 
programming tools which are presently available for example with DBMS systems using fourth generation 
languages.” The evolution of this type of system was forecast by Frank, et al. (1991) in their discussion of 
trends for the decade. In fact, they predicted that a sizable commercial market would develop to implement 
this type of customized integrated system for large organizations. 
1.3 REMOTE SENSING ISSUES 
iwards the 
arison. In 
lowever, it 
}f grid cell 
terns, have 
either data 
both types 
rtographic 
is between 
While it is clear that there have been major advancement in terms of computing power and the 
interchange of data, it is not clear where we stand with respect to solving the problems facing the remote 
sensing community. In a recent review article Wilkinson (1996) suggests that classification and spatial 
generalization remain the major issues. He believes that we are still dependent on parametric classifiers to 
help model spectral feature space and develop reasonable classifications of the data. These routines are based 
on per pixel statistical pattern recognition procedures that yield a great deal of noise. In order to be useful 
in a GIS environment these pixels should be spatially generalized or segmented into homogeneous polygons. 
In fact, Wilkinson believes that it is “still true to say that satellite imagery often does not yield the quality of 
product needed by users of GIS”(Wilkinson, 1996, p. 89) He believes that new approaches that utilize