In this paper an effort is made to demystify usage of this term by drawing a broad distinction between GIS as an industry/ product/technology, and GIS as a science. In the former case, GIS is viewed very much as a technological tool that helps the analyst to use his or her knowledge and insight to study substantive issues. In the latter case, GIS is viewed as the science of geographic or spatial information that possesses its own set of research questions (see Rhind et al. 1991). The current state and future trends of each of these views of GIS is presented and the relationship between the two views is discussed.
As a product, GIS is more readily defined as computer software, although so-called GIS "solutions," as sold by system vendors and consultants, may frequently involve hardware, database design and capture, and training. As a software product, the term GIS has often been loosely interpreted or, less graciously, manipulated, so that cartographic and mapping packages are commonly marketed as GIS. These packages generally lack both the analytical functionality and the well-developed links to RDBMS of GIS and, if vector-based, rely on simple data structures that ignore topology (Maguire 1991).
The lack of a rigorous definition of what constitutes GIS software has also led to confusion with image processing and CAD. Generally, image processing packages specialize in the manipulation and display of raster data. Although they share much basic functionality with raster data handling in GIS, such packages also have sophisticated forms of analysis that go beyond those present in GIS. Image processing packages, however, typically have very limited capabilities in vector data handling and poorly developed links to RDBMS. CAD packages, not unlike computer cartography software, generally lack topology and also have weak relationships with RDBMS, although they are strong in 3-D data handling because they allow, unlike GIS, multiple surface values for the same x-y planar location (Maguire 1991).
Finally, a recent trend in GIS software has been the production by GIS vendors of specialized software or software modules for use in specific application areas. This trend toward exploiting vertically integrated markets is partly a reflection of the maturing of GIS as an industry. Progressive application areas (such as forestry or municipal affairs), now use GIS across a broad range of inventory, analysis, and management functions and, consequently, across many different departments within the same organizations. However, this trend also reflects the fact that users are increasingly looking for tailor-made and streamlined solutions to their work that sidestep the need to learn a full-blown GIS. As smaller GIS vendors have responded to this demand in certain sectors, so competitive forces have forced the major GIS vendors to respond likewise. In terms of the perception of what constitutes GIS, however, users of these specialized systems will undoubtedly have a restricted view and one that is interpreted solely in their own application context. One result of this trend has been a further proliferation of alternative names for GIS: cadastral information system, market analysis information system, soil information system, spatial decision support system, and so on.
As a technology, GIS transcends disciplinary boundaries and has found wide acceptability across a range of application areas, including land use management, traffic routing/assignment, political redistricting, resource management, and environmental modeling. The widespread use of GIS has much to do with the acceptance of the map as a means of communication, in addition to developments in graphic computing. These developments have enabled the map medium to be presented, manipulated, and analyzed in a new form and with unparalleled flexibility. GIS allows us to approach spatial data handling with much improved efficiency in implementing traditional methods and techniques.
Nowhere is this more evident than in the quintessentially GIS operation of map overlay. Although a relatively simple notion to comprehend and a formal analysis technique that dates back at least to McHarg (1969), this ability to link information together across numerous thematic layers for the same location and at great speed is an incredibly powerful tool. Whereas traditional RDBMS capabilities merely retrieve information records based on key fields, map overlay represents a major extension in that new data records, representing derived spatial features, may be created and populated with information drawn from different themes. Other traditional methods that benefit from the data structures employed in GIS are buffer operations (particularly in raster processing) and network analysis (vector processing).
However, despite the obvious power of GIS to manipulate and integrate data and to perform certain kinds of spatial analysis, many agencies use GIS merely for inventory management (Rhind 1988; Dangermond 1991). It seems that the relative novelty and power of working in a graphic environment for the management of information, and the satisfaction of being able to produce maps of that information at will, has perhaps obscured the vision of users from the real potential of suitably developed GIS for in-depth spatial analysis.
In turn, the willingness of agencies to purchase GIS purely on the basis of this automation of current spatial data handling and spatial data inventory, has led to few incentives for commercial GIS software vendors to incorporate spatial (and temporal) analytic functionality into their systems. Given the present sophistication of statistical packages for analyzing non-spatial data, it is not surprising that the existing state of GIS analytical capability is often compared to the state of statistical packages in the early 1970s (Rhind et al. 1991).
Regardless of the level of analysis, GIS is typically seen as a technological tool that helps the analyst use his or her knowledge and insight to study substantive issues. This viewpoint places GIS firmly in the role of an enabling technology. Although there is little doubt that GIS fulfils this role to a large degree, it conveys the impression that GIS merely provides a toolbox for operationalizing some form of analysis focused around a substantive issue. The implication is that analysts need to be all-knowing with regard to their substantive field and that issues surrounding the enabling techniques play a very secondary role. In this sense, GIS has a lot in common with statistics use: analysts all too frequently use inappropriate statistical methods or correct statistical methods inappropriately.
There also seems to be a perception that the mere viewing of information and data that GIS permits is sufficient, and that this simply stimulates the user's own substantive knowledge which then takes over. Although this role is undoubtedly useful, this view tends to stifle the development of techniques to help the user detect and interpret patterns more objectively.
It can be argued, then, that GIS should in fact be the science of geographic information, and should concern itself with fundamental research issues of using digital geographic data. In the same way that we accept the discipline of statistics for its fundamental research, and recognize how development at the research level feeds into the statistical software technology that is used by countless analysts, so there should be little problem in accepting a similar situation for GIS.
Indeed, were it not for the heavy computational demands of graphic geographic data that delayed the development of GIS software relative to that of statistical software, we may well have seen geography become to GIS software what statistics has become to the software industry it supports. Instead, and somewhat paradoxically, geography went through essentially a spatial quantitative revolution, often at the expense of those parts of the discipline, notably cartography, that were interested in issues of handling digital spatial data.
In the academic setting of today, it is difficult to see the diverse discipline of geography focus itself purely as the science of geographic information. Meanwhile, the rapid and ubiquitous growth of GIS technology, partly based on the appeal of a highly graphic computing environment and the popularity of the map as a medium, makes potential misuse and abuse a significantly large problem. The need for a science of geographic information, then, is very real (Goodchild 1990), and perhaps merits a discipline in its own right.
There also exists a need to develop appropriate data structures and data models for the handling of 3-D and temporal data within GIS, and legitimate and substantial research questions surrounding the use of expert systems and artificial intelligence. To these issues, we could add the need to develop database management systems and query languages geared toward spatial data, and the particular concern surrounding the issue of error propagation and error management within GIS.
Finally, there is the extremely important question of the relationship between spatial analysis and GIS. We need to address the problems of integrating existing spatial analytical methods into GIS, and of developing new methods of spatial analysis that specifically take advantage of the data structures within GIS. The importance of these issues becomes apparent if we consider the impact that scientific ideas such as the TIN data structure or the quadtree have had on the technological development of GIS.
A review of each of these areas is clearly beyond the limits of this paper, but the reader is referred to the excellent two-volume publication by Maguire et al. (1991) for papers that cover these topics. For the purposes of this paper, the issue of spatial analysis and GIS will be examined more closely. The justification for this focus lies in the fact that, ultimately, the real value of GIS will be in solving complex problems using sound and rigorous methods that are firmly rooted in spatial statistical theory. It is significant that the issue of spatial analysis and its relation to GIS is a key research issue for both the US National Centre for Geographical Information and Analysis (NCGIA 1989) and the UK Regional Research Laboratory initiative (Masser 1988).
Fotheringham and Rogerson (1994) have recently produced an edited volume that focuses on spatial analysis and GIS. In a review chapter, Bailey (1994) makes a useful distinction between spatial summarization of data, and spatial analysis of data. The former is meant to include functions for the selective retrieval of spatial information and the computation, tabulation, or mapping of statistical summaries of that information. In Bailey's terms, this category of functionality would include both spatial query and many other techniques, such as Boolean operations, map overlay, and buffer generation, which many users commonly perceive as analysis functions.
Meanwhile, the term spatial analysis is reserved for methods that either investigate patterns in spatial data and seek to find relationships between such patterns and the spatial (and perhaps temporal) variation of other attributes, or for methods of spatial or spatio-temporal modeling. The second of these would include network analysis, location-allocation models, site selection, and transportation models, all of which are considered by Bailey to be quite well developed within many GIS. The former type of spatial analysis, however, which Bailey refers to as statistical spatial analysis or simply spatial statistics, is currently poorly represented in the technology of GIS. This type of analysis would include such areas as nearest neighbour methods and K-functions, Kernel and Bayesian smoothing methods, spatial autocorrelation, spatial econometric modeling, and spatial general linear models. The level of integration of these types of methods with GIS has barely gone beyond the use of GIS to select input data and display model results. At the software level, this same integration generally involves only loose coupling between GIS and spatial statistical packages in the form of ASCII data transfer or specially programed interfaces.
Bailey (1994) is somewhat pessimistic about the prospects of fully integrating statistical spatial analysis into GIS. He advocates a form of loose coupling based on open-systems computing environments wherein the GIS package and the statistical analysis package would be accessed simultaneously but independently on the same GUI. The key problem would then be the data transfer mechanism between the packages. Bailey's pessimism centres on the number of theoretical problems that remain with spatial analysis methods, making them, in his opinion, difficult to implement in a sufficiently general and robust form for widespread consumption in commercial GIS. This underlines the necessity of developing a science of geographic information. Efforts would be better directed to addressing these fundamental problems, rather than accepting them as given.
As a technology, the future of GIS hardware seems easy to predict. There appears to be every indication that the rate of growth in computer processing power is likely to continue for the foreseeable future. Indeed, with parallel processing technology, this rate of growth may increase substantially, particularly for applications that can take advantage of it. This is definitely the case in GIS: the same operation may be carried out across a whole set of spatial objects, such as pixels. Tremendous ramifications for GIS will also result from hardware advances in global positioning systems (GPS).
These systems are already more accurate than the typical base mapping scales of most countries and will redefine many methods of data capture and accepted levels of accuracy. Finally, GIS will be profoundly affected by the inevitable trend toward multimedia and networking. The ability to display text, maps, data, photographs, video, and sound for locations from a myriad of networked sources will give new definition to what we typically think of as a spatial database.
Beyond hardware, there is also little doubt that GIS in the future will show increasing specialization. GIS application areas have very different requirements in terms of level of functionality, speed of response, and quantity of data handled. For instance, the GIS requirements of an emergency dispatch system, for which time is of the essence, have little in common with the management of cadastral land parcels, for which data volume may be the main concern. As specialized products emerge, the possibility exists that GIS as a distinct and discernible field will disappear.
Countering this trend, however, will be a unifying concern with standards, including the problems of data definition, function definition, data accuracy, and data exchange. As principal suppliers of digital data for GIS, public agencies have a major role to play in these areas; their capacity and willingness to do so is a critical question for the future. The one caveat in this regard, however, may be the propensity of governments to use the excuse of data standards to justify treating spatial data as a commodity, thereby demanding heavy payment and restrictive usage agreements.
The GIS industry may expand significantly, particularly at the low end, with what would essentially become graphic interfaces to spatial databases. Good examples of this trend include the product ArcView from ESRI, the market leader in GIS software, and the addition of spatial data handling capabilities to such packages as Lotus 1-2-3 and SAS. GIS may actually become as commonplace in the desktop computing world as word-processing, spreadsheets or database packages. Moreover, they would feature the same embedding and linking technologies (e.g. OLE) that are currently integrating these different software platforms seamlessly.
Again, but for different reasons, the availability of GIS in such a format may actually dilute the identity of the field as users simply accept such capabilities as commonplace and come to expect them as an inherent part of any desktop software suite. The enthusiasm with which ESRI is attempting to introduce ArcView to the library and K-12 school sectors in North America represents an important portent.
As a science, a number of possible scenarios also exist for GIS. GIS must increasingly become a discipline in its own right, although not necessarily under that banner. The issues for spatial data analysis have been widely documented (see Openshaw 1991a; Goodchild et al. 1992), and there is a legitimate case for creating a separate scientific discipline. Already, GIS has many of the features of a separate discipline in journals, conferences, research institutes, and degree programs, albeit embryonic and often under the umbrella of a parent discipline, such as geography.
The extent to which this could develop further is open to question, particularly given the range of parent disciplines from which academics interested in this new science would be drawn, including computer science, ecology, geography, geology, economics, and environmental science. Also, in the present academic climate of encouraging interdisciplinary work in areas that cut across traditional disciplinary boundaries, such as health and the environment, it is unlikely that such a new discipline would be recognized.
However, if GIS as science is to continue to garner academic recognition and funding, it is perhaps vital that the science be able to demonstrate the importance and usefulness of its research to GIS applications (see Goodchild et al. 1992) so that the wider GIS industry becomes an advocate and potential financial partner. This will involve a critical assessment by researchers of spatial analysis methods and their relevance to the powerful computing and GIS environment of the 1990s. In this regard, it should be noted that much of the research in spatial analysis methods dates back to the 1960s and early 1970s, followed by a relative dearth of research until the GIS-inspired revival of the late 1980s and 1990s. As Openshaw (1991b) argues, many methods from the early research were developed within a completely different computing environment. It may be that new methods, which capitalize on the raw computing power available today and into the future, should now be the focus of attention.
Another scenario for GIS as a science, could be that traditional disciplines will struggle with the issues of spatial data handling and analysis somewhat independently. In this scenario, GIS as science may become little more than a very loose and informal consortium of academic interests, much like the field of remote sensing. The dangers of research duplication and dilution would be very real, and a lack of critical mass in the form of academic institutions to spearhead the adoption of methods into GIS would result. This scenario, then, may be a recipe for the field of GIS being led too rapidly by technology and with science not being given the chance to catch up, a familiar theme to the field of remote sensing.
On the other hand, a belief exists that the typical use of GIS has not progressed far beyond the use of mapping, query, and spatial data inventory management, and that the potential analytic power of the technology to help solve complex societal and environmental problems has yet to be realized. For this to happen, there is a need for fundamental research into the science of geographic information, a need for more widespread and enhanced education in this science, and a willingness on the part of the GIS industry to nurture this science and be ready to adopt and promote the analytical techniques it produces.
Back to Contents -
To next section
This file was created 23 February 1996