A Framework for Geographical Modeling in a Heterogeneous Computing Environment

David A. Bennett and Raja Sengupta
Department of Geography
Southern Illinois University, Carbondale
Carbondale, IL 62901-4514
dbennett@siu.edu

Greg A. Wade
Department of Computer Science
Southern Illinois University, Carbondale
 
The resolution of spatial problems often requires consensus building and compromise among decision-makers as they attempt to optimize their own set of criteria. The evaluation of such criteria often requires access to an extensive set of geographical models, analytical tools, and data. The Internet provides new opportunities for the sharing of such resources. Many potential users, however, lack the time, money, and/or technological capabilities need to integrated applicable models, tools and data into a software environment that can support spatial problem solving. To take full advantage of the opportunities afforded by increased access to data and models, new geoprocessing technologies are needed that are capable of bridging multiple vendor formats and heterogeneous computing environments.
 
To construct links between disparate software products and data formats, a common communication protocol is required. Such a protocol can be implemented if software products can import and export to a common framework or if software vendors provide "hooks" into proprietary data structures. Communication protocols are in development for geographical databases (SDTS, OGIS), however, there is little support for the distribution and sharing of the geographical models and analytical tools. The objective of this research effort is to develop a framework that supports collaborative decision making in a client/server environment that provides access to a variety of geographical models, analytical tools and data. Such an environment will provide decision makers with: 1) access to computing resources that may not otherwise be available; 2) an electronic forum for the exchange of ideas in written and/or cartographic form; and 3) a more level "technological playing field" on which to build consensus and compromise.
 
To reach this objective we are pursing three interrelated research objectives. These objectives include the development of a(n):

  1. framework for distributed geographical modeling;
  2. distributed modelbase and database management systems;
  3. intelligent search engine driven by spatial metadata and a geographical modeling language
The framework presented here builds on previous work by Bennett (1997; in review), and Wade et al. (1997). Existing frameworks for the representation of geographical data are built on data models that treat the geographical database as a digital surrogate for geographical space. Yet each digital representation of geographical space is an abstraction of reality created by the user to solve a particular set of related problems. As such, the selection, organization, and implementation of those spatial elements that comprise this abstraction depends on such factors as our understanding of how spatial processes operate, the objectives of the analyst, data availability, and the spatial extent to be studied/managed. Geoprocessing technologies designed to support the study and management of complex geographical systems must, therefore, integrate methods for the representation of geographical knowledge with more traditional methods for the representation of geographical space.
 
In the modeling framework presented here, knowledgebase management, modelbase management and geoprocessing technologies are integrated into a single system that supports the digital representation of dynamic geographical systems. Model design is viewed as the capture of geographical knowledge and the organization of this knowledge into a model graph that emulates the spatial processes of a particular geographical system. The representation of geographical systems as graphs provides a modeling topology well suited to a distributed implementation. The management of model graphs over a computer network required the development of new interoperability tools. These platform independent tools were designed to: search network accessible repositories of geographical models, atomic model components, and geographical data; provide an interactive mechanism for integrating geographical models from atomic components; and execute these models across a distributed modelbase and database. Prototype software was developed using Java and its extensions (Wade et al. 1997). Java was created for developing network aware applications and possesses unique features that facilitate the development of software designed to be executed in a distributed environment. These features include platform independence and the ability to dynamically load and bind compiled code over the network.
 
The construction of geographical models from atomic components often requires the coupling of data and models derived from multiple sources. Existing coupling strategies (e.g., loose and tight coupling) that link GIS with modeling software are often either too complex or too rigid. A new method of linking GIS and analytical models is proposed here that builds on earlier work by Sengupta et al. (1996). Using the modeling framework described above as a guide, intelligent agents match datasets (irrespective of vendor formats) to models and create wrappers around modeling software and GIS datasets. The existence of these wrappers is largely hidden from the user's view.
 
Intelligent software agents communicate via a Model Definition Language (MDL) to integrate data with models. The MDL provides an inter-agent communication protocol for model development. Through the use of the MDL, agents retrieve, manipulate and store geographical databases and modelbases. To accomplish this task agents parse an MDL query and translate tokens into a sequence of software specific spatial operations that transform the data into a form that is usable by particular models. Within the MDL spatial queries are defined using topological relations (Egenhofer and Fransoza 1991; Clementini et al. 1993), spatial operators (Tomlin 1990; Wesseling et al. 1996), and constructs from the C programming language. Multiple agents may be invoked in the process of performing a spatial analysis. Human intervention is required only to state preferences for models and data source to be used. The agents can also act as advisors that suggest appropriate model selection and lead the user through complex spatial analyses.
 
The search for relevant data and the identification of data format is achieved through the use of metadata. The metadata format adopted follows the Spatial Data Transfer Standards. The Geographical Name Server (GNS) proposed by Wade and Bennett (in press) was used to further facilitate the identification and processing of this metadata by intelligent agents. The GNS provides a framework for creating a hierarchical topical nomenclature for geographic data which is required to effectively and efficiently automate the process of interpreting metadata.

References:

Bennett, DA in review. Managing geographical models as repositories of scientific knowledge. Submitted to Geographical and Environmental Modelling.

Bennett, D.A. 1997. A framework for the integration of geographic information systems and modelbase management. International Journal of Geographical Information Systems 11(4): 337-357.

Clementini, E., Felice, P. D., VanOosterom, P., 1993, A Small Set of Formal Topological Relationships suitable for End User Interaction. In Advances in Spatial Databases, Proceedings of the Third International Symposium on Spatial Data Handling (Lecture Notes in Computer Science No. 692), Ed. D. Abel and Ooi, Beng Chin . New York: Springer Verlag: 277-295.

Egenhofer, M. D., and Franzosa, R. D., 1991. Point Set topollogical spatial relations. International Journal of Geographic Information Systems, 5(2): 161.

Sengupta, R., Bennett, D.A., and Wade G.A. 1996. Agent mediated links between GIS and spatial modeling software using a model definition language. In Proceedings of GIS/LIS '96, Bethesda, MD: American Congress on Surveying and Mapping: 295-309.

Tomlin, C. D., 1990, Geographic Information Systems and Cartographic Modeling (New Jersey: Prentice Hall): 249.

Wade G.A., and Bennett, D.A. in press. GNS: A Distributed hierarchical topical nomenclature for geographic data. In Proceedings of GIS/LIS '97.

Wade G.A, Bennett, D.A., and Sengupta, R. 1997. An interactive distributed architecture for geographical modeling. In Proceedings of Auto-Carto 13. , American Congress on Surveying and Mapping, Bethesda, MD: 307-316.

Wesseling, C. G., Van Deursen, W. P. A., and Burrough, P. A., 1996. A Spatial Modeling Language that unifies dynamic environmental models and GIS. In: Proceedings, Third International Conference/Workshop on Integrating GIS and Environmental Modeling, Santa Fe, NM, January 21-26, 1996. Santa Barbara, CA: National Center for Geographic Information and Analysis. http://www.ncgia.ucsb.edu/conf/SANTA_FE_CD-ROM/main.html.