For physical science research, the role of current GIS systems has remained as "front ends" (pre-processing spatial data to prepare model input) and "back ends" (visualizing model output spatially). It has become clear that the analysis functions of GIS cannot replace the analysis functions in process models. The GIS functions are designed for manipulating and extracting information needed for the models. These analytical capabilities of GIS have remained weak in comparison with other capabilities that GISs have promised to deliver, such as spatial data management and visualization. The improvement in analysis capability has not caught up with the pace of other GIS capabilities. This weakness affects physical scientists' ability to deal with spatial phenomena. The often heard "We know everything about GIS the geographers do" from ecologists refelct such limitation in current GISs.
The weak capability of spatial analysis is not the only impedance for physical scientists to use GIS. The better developed data management and visualization capabilities of GIS have not always delivered a satisfactory user environment. For example, data format incompatibility between GIS and models is a simple technical problem, but it is one of the most costly problems for users in modeling community (Karimi, 1997). It is common for modelers to spend much greater proportion of time, energy, and resources on data conversion than on model calibration. Automating the conversion is model-dependent and it does not alleviate the burden for physical scientists who are constantly involved with different models. Although it is unrealistic to expect GISs to provide data format for all models, current GISs have not provided tools to ease the conversion.
Relational database remains to be the dominant framework for storing spatial data despite many academic discussions over its strengths and shortcomings (Kim and Lochovsky, 1989). For models that describe dynamic processes, there are at least three types of relations: (1) relations between different variables at a fixed location, that are represented by the mathematical functions, (2) relations between different locations for a fixed variable, such as in the situation of finite difference for hydrology, and (3) a combination of (1) and (2), in that both locations and variables are related. Relational GIS databases cannot adequately accommodate any of the relations. This makes it difficult for modeling to directly use data manipulation functions built in GIS databases. Conceptual compatibility between GIS and modeling, such as raster data structure with finite difference as well as TIN with finite element, have remained as academic discussions. Current GIS database does not all! ow direct access to its data structures so that modeling functions can be linked. Developing in-house code is still a much more practical approach to use the raster- or TIN-like data. With the difficulties of data format and database, GIS has retreated to more of a data provider (e.g., DEMs, DLGs) than a data analyzer to many physical scientists.
Efforts have been made to ease the problems. Integrating GIS with environmental models has been the title of three international conferences (Goodchild et al., 1993; Goodchild et al., 1996; NCGIA, 1996). Some early efforts attempted to build GIS functions within a process model or more often models are rebuilt within a GIS (Betty and Xie, 1994). This approach has proved to be limited. More successful or more practical approach has been leaving the GIS and models essentially intact but bridging the two together. Integration strategies such as simple data file transfer, loose coupling, and tight coupling (Chou and Ding, 1992; Nyerges, 1993; Abel et al, 1994) are daily practice in many physical scientific work. Each strategy has its benefits and costs. The more recent development of OpenGIS specifications for spatial data and function (Buehler and McKee, 1996) holds new promises to alleviate the daily burden of integrating. At least it provides a solution for data format incomp! atibility problem. The aforementioned efforts seek for technical solutions. Semantic compatibility was assumed to be handled by the end user; thus it is rarely discussed under the topic of integration problem.
Scale incompatibility is a problem beyond technical solution. Differences in development history may have contributed to the mismatch. Many process models used today were originally developed in 1970s when computer became available. The popular use of GIS came at least a decade later in mid 1980s. In coping with the lack of means to handle spatial data, the original model development was restricted to simplistic treatments of spatial variation, for example, using coarse spatial resolution or small spatial extent. Many models use raster-like data because it is a simple way to partition the space. This may explain why raster-based GIS packages such as GRASS is so popular among modelers (Being a public domain package and having open architecture are not the only reasons). Not only the model requires coarse resolution input, it also simulates the physical processes that occur at corresponding spatial scales. With these traits carried to today, many models are prohibited from taki! ng the advantage of the details provided by today's GIS data. Often the GIS data must be aggregated to a coarser resolution before they can be entered into a model (Zack and Minnich, 1991). This problem is more inherent than problems such as data format incompatibility.
The representational difference between GIS and process modeling is more challenging. Current GISs represent static, layered world through spatial data models. Process models use mathematical functions to model the dynamics of the world (Maidment, 1993, 1996). For distributed (or spatially explicit) models, raster data structure is still the best available. Current GIS data in general cannot accommodate the need for representing dynamic processes. The object orientation paradigm offers many advantages for this purpose (Raper and Livingstone, 1995), but it is better suited to object-like phenomena. The process models deal primarily with field-like, continuous phenomena. It is conceptually as well as technically difficult for object orientation to implement dynamics of fields. Kemp (1997a, 1997b) addressed the issue of integrating field data and process models from representational perspective. In-depth analysis as such is much needed.
Another difference between GIS and process models lies in the fact that GIS is meant to provide an objective representation of the world through stored measurements and observations. Information may be eventually extracted based on a particular need (Peuquet, 1994). In this sense, GIS is a generalist. In contrast, modeling usually focuses on a particular process; thus process models are specialists. A generalist GIS package cannot always meet requirements of specialist models. This mismatch has also caused the gaps in practical use of GIS in physical sciences.
In addition, effective tools to represent and handle three dimensional data are still not readily available (Scott, 1997). Furthermore, very little has been studied about representing flow in GIS. The two issues are important because most natural processes are three dimensional, and flow of energy and material is the core concept of physical sciences.
The incompatibility is certainly two-sided. Take the data format problem as an example, not only environmental models are incompatible with GISs, but also among the models themselves. Many environmental models are monolithic, legacy models. Developing platform- and language- independent modules (or algorithm library, or component-ware) is a technically feasible solution (Leavesley et al., 1996). Such a solution, however, requires resources. Institutional support is much more critical than technical solutions. Investing effort into such development may not be seen as the right path for career advance for many physical scientists. It is less likely that the commercial software developers will take the task. Research communities in physical science may play an important role in GIS development, but it represents a small and diverse market for GIS products.
Incompatibility between GIS and process models in terms of data format, database, scale, and representation requires different solutions, from simpler technical solutions to more sophisticated representational ones. Some technical solutions are already on the way. Representational solutions will take longer time and more efforts. Above all, institutional solution is always more critical and more difficult to achieve.
Batty, M. and Xie, Y. (1994) Modeling inside GIS: Part 1: model structures, exploratory spatial data analysis and aggregation. International Journal of Geographical Information Systems, 8(3): 291-307.
Buehler K. and McKee L. (1996) The OpenGIS Guide. http://www.opengis.org, The Open GIS Consortium, Inc.
Chou, H.C., and Ding, Y. (1992) Methodology of integrating spatial analysis/modeling and GIS. Proceedings, 5th International Symposium on Spatial Data handling, Charleston, South Carolina, August 3-7. 514-523.
Goodchild, M. F., B. O. Parks, and L. T. Steyaert (Eds.) Oxford University Press, New York.
Goodchild, M.F., Steyaert, L.T., Parks, B.O., Johnston, C., Maidment, D., and Glendinning S. (Eds.), GIS World, Inc. Fort Collins, Colorado.
Karimi, H.A. (1997) Interoperable GIS applications: tightly coupling environmental models with GISs, Proceedings, International Conference on Interoperating Geographic Information Systems. Santa Barbara, California, December 3-4.
Kemp, K.K. (1997a) Fields as a framework for integrating GIS and environmental process models. Part 1: representing spatial continuity. Transactions in GIS, 1(3): 219-234.
Kemp, K.K. (1997b) Fields as a framework for integrating GIS and environmental process models. Part 2: specifying field variables. Transactions in GIS, 1(3): 219-234.
Kim, W. and Lochovsky, H. (1989) Object-Oriented Concepts, Databases, and Applications. ACM Press, New York.
Leavesley, G.H., Restrepo, P.I., Stannard, L.G., Frankoski, L.A., and Sautins, A.M. (1996) MMS: a modeling framework for multidiscipliary research and operational applications. In GIS and Environmental Modeling: progress and Research Issues. Goodchild, M.F., Steyaert, L.T., Parks, B.O., Johnston, C., Maidment, D., and Glendinning S. (Eds.), GIS World, Inc. Fort Collins, Colorado. 155-163.
Maidment, D. R. (1993) GIS and hydrologic modeling. In Environmental Modeling with GIS, Goodchild, M. F., B. O. Parks, and L. T. Steyaert (Eds.) Oxford University Press, New York. 147-167.
Maidment, D.R. (1996) Environmental modeling with GIS. In GIS and Environmental Modeling: progress and Research Issues. Goodchild, M.F., Steyaert, L.T., Parks, B.O., Johnston, C., Maidment, D., and Glendinning S. (Eds.), GIS World, Inc. Fort Collins, Colorado. 315-323.
NCGIA (1996) Third International Conference/Workshop Integrating GIS and Environmental Modeling, Santa Fe, New Mexico, January 21-25. NCGIA, http://www.ncgia.ucsb.edu.
Scott, M. (1997) Extending map algebra concepts for volumetric geographic analysis. Proceedings, GIS/LIS '97. Cincinnati, Ohio, October 28-30.
Nyerges, T. (1993) Understanding the scope of GIS: its relationship to environmental modeling. In Environmental Modeling with GIS, Goodchild, M. F., B. O. Parks, and L. T. Steyaert (Eds.) Oxford University Press, New York. 75-93.
Peuquet, D.J. (1994) It's about time: a conceptual framework for the representation of temporal dynamics in geographic information systems. Annals of the Association of American Geographers, 84(3): 441-461.
Raper, J., and Livingstone, D. (1995) Development of a geomorphological spatial model using object-oriented design. International Journal of Geographical Information Systems, 9(4): 359-383.
Zack, J.A. and Minnich R.A. (1991). Integration of geographic information systems with a diagnostic wind field model for fire management. Forest Science, 37(2): 560-573.