4th International Conference on Integrating GIS and Environmental Modeling (GIS/EM4):
Problems, Prospects and Research Needs. Banff, Alberta, Canada, September 2 - 8, 2000.

Spatial modeling environments:

Integration of GIS and conceptual modeling frameworks

GIS/EM4 No. 2

Susan Crow

Abstract

Successful environmental problem solving occurs with the implementation of management activities derived from an inclusive decision process supported by appropriate and credible data and analyses. Problem solving is greatly enhanced when tools are available to support collaboration among key players in the decision process: scientists, technicians, affected parties, and decision-makers. In this regard, integration of a Geographic Information System with a generic and easy to use conceptual modeling framework could provide a powerful environment to enhance problem solving. Such a spatial modeling environment would encourage greater collaboration among research and modeling communities, support more active participation of stakeholders and decision-makers, and provide clearer documentation of the decision process. This paper discusses general features of existing spatial modeling environments, highlights strengths and limitations of these tools for addressing environmental problems at various scales and complexities, and suggests some directions for future development.

Keywords

Spatial modeling environment, conceptual modeling framework, hierarchical model structure, process flow diagram, icon-based model interface, environmental decision support

Introduction

In general, a spatial modeling environment may be thought of as an integrated set of software tools providing the computer facilities needed to develop and execute spatially explicit simulations and display model results. These integrated environments have been designed to support modeling efforts of groups engaged in activities as varied in scope as global climate change research, watershed management, and urban planning.

Various approaches have been undertaken to integrate spatial modeling with GISs. These approaches have been described relative to intensity of coupling (Goodchild, 1995), as well as degree of modeling flexibility (Albrecht et al., 1997). A number of these efforts have resulted in methods for modeling environmental processes such as forest dynamics (Shugart, 1984) and hydrologic processes (Haan et al., 1982). Other developments have introduced graphical user interfaces with sliders to modify weightings within models (Wade and Wickham, 1995). While these method allows exploration of alternative scenarios, they are domain specific and do not support generic spatial model development.

Other approaches to spatial modeling and GIS integration have required users to write code in a formal programming language (Grayson et al., 1994; Smith et al., 1995) or assisted users to specify model structure either through guided question and answer sessions (Robertson et al., 1991) or using pseudo-English to generate code (Lowes and Walker, 1995). Albrecht et al. (1997), in pointing out limitations of these approaches, have noted that they tend to be domain-specific, require users to learn a specific programming language, may be difficult to follow through model implementation, and importantly, do not support creative conceptual model development.

Another approach to integrating spatial modeling and GISs is diagrammatic, that is, spatial models are represented as process flow diagrams that graphically illustrate relationships among input data, geo-processing functions, and output or derived data. Applications of this approach range from image analysis (ERDAS IMAGINE Professional 8.4, Spatial Modeler) to static cartographic modeling (Virtual GIS or VGIS prototype described by Albrecht (1996), and ESRI's ModelBuilder in the Spatial Analyst 2.0 extension to ArcView GIS) to dynamic simulation modeling (Spatial Modeling Environment, SME). This approach has a number of advantages. First, these types of flow diagrams frequently appear in various disciplines and therefore represent a common conceptual framework. In fact, such flow charts are a standard process-oriented tool in visual programming (Chang, 1990). Process flow diagrams make relationships among model elements apparent and model behavior easy to follow and explain to others. This is a powerful advantage for non-GIS model developers, as well as stakeholders and decision-makers, as they engage in exploring and solving environmental problems.

Diagrammatic spatial modeling approaches

While various prototypes demonstrating the use of a process flowchart method to integrate GIS and spatial modeling have been presented (e.g., Theobald, 1998), they have not developed beyond concept papers. As indicated above Virtual GIS (VGIS), ModelBuilder, and Spatial Modeling Environment (SME) represent advanced applications of this diagrammatic approach to integration of GISs and spatial modeling tools. The following section will describe these efforts briefly and discuss important similarities and differences.

Virtual GIS, a research prototype

The Virtual GIS project is a research program funded by the German Science Foundation (Albrecht et al., 1997). Originally designed as a tool to integrate image processing software and GIS, VGIS has evolved into a visual programming tool that works with GIS data and graphically represents cartographic modeling techniques. VGIS runs on a personal computer connected to a workstation under the UNIX operating system. Currently, VGIS combines WiT from Logical Vision Ltd. for managing icons in the user interface, Motif-windows developed with X-Designer4 Graphical User Interface (GUI) builder for parameter input, and the public domain GIS GRASS. Additional functions, including the shell script that initiates the VGIS, are implemented in C and C++.

A major goal of the VGIS project has been to facilitate ease of GIS use by both expert and non-GIS expert users. In this regard researchers have identified seven criteria that they argue will create a useful spatial modeling system. These criteria include: a graphical user interface that is visual and appears simple to users; an environment that allows users to develop scenarios interactively; a system dynamic modeling environment that includes feedback, is flexible, and supports conceptual model development; an environment for spatial analysis and display; a model database for tracking analysis history and output scenarios; an integrated approach in which model components include spatial objects; and a generic system that operates as a toolbox independent of a specific domain.

To achieve program goals and criteria, VGIS was implemented as a shell to be draped over and use the functionality of an existing GIS. As it has evolved, the VGIS-Shell consists of these modules: a graphical user interface, an interpreter for the processing plans, and the underlying off-the-shelf GIS (public domain GIS GRASS). The user interface consists of a flow-charting environment in which users construct models by combining operation icons. Operations are independent of data structure, that is, all data format conversion is performed transparently.

The interface employs a set of 20 universal analytical GIS operations selected to facilitate ease of use of providing typical GIS functionality without the bulk of auxiliary operations that make most GISs cumbersome and overwhelming to users (Albrecht, 1995). The graphical user interface acts as a general-purpose flow charting tool depicting workflow as a number of processing steps applied to input data. The processing plan is created and edited interactively. The plans are self-documenting and may be saved, modified, and reused with other datasets. Since the processing plans preserve data lineage, data also become self-documenting.

The processing plan interpreter translates flow chart operations into operations of the underlying GIS. The interpreter consists of two steps: VGIS functions are dissected into elementary GIS operations, and the GIS operations are then translated into the proprietary functions of the underlying GIS. The two-step approach maintained independence of the first phase from the underlying GIS, allowing customized user interfaces and processing diagrams to be exchanged across platforms. A fourth module, a tool to generate the processing plans, is envisaged as a future extension to VGIS. The flow chart building tool would automatically generate an appropriate processing plan based on user description of results and identification of source data and on a knowledge base of integrity rules.

A processing plan is executed in VGIS by selecting a menu item Run. A C program processes input and output filenames and operator parameters so that one or more GRASS commands are issued. GRASS implements the functions and saves the results to hard disk. When input parameters are required, the C program calls a C++ module that prompts for required data. Incorporating display and output operators into the process flow chart allows intermediate results to be displayed and compared (otherwise temporary files are deleted before the termination of the program and only the final result is saved).

In its current version VGIS allows static cartographic modeling. VGIS supports spatial statistics necessary for systems modeling; however, current GIS operations do not support temporal or three-dimensional reasoning. Researchers indicate that the environment could be modified to include feedback loops. Future development of VGIS as a dynamic modeling tool is intended to occur in close collaboration with a similar research program, SPMS, that already supports dynamic modeling (Mann, 1996). And, in future versions the system is expected to support multiple-scale processing allowing sub-processes to operate on different temporal and spatial scales. This extension would allow landscape modeling and support various other ecosystem research efforts. To support representation of more complex equations, VGIS developers also intend to implement modular links to other modeling tools in the plug-and-play fashion of OGIS-compliant services. Future basic research will focus on development of a universal interface to the Open Geodata Interoperability Specification (for additional information on OGIS see http://stills.nap.edu/readingroom/books/whitepapers/ch-58.html).

ModelBuilder in the Spatial Analyst 2.0 extension to ArcView GIS

Environmental Systems Research Institute, Inc. (ESRI) has recently introduced ModelBuilder (description provided at http://www.esri.com/software/arcview/extensions/spatext.html). ModelBuilder is included in Spatial Analyst 2.0, a raster-based analysis application that operates in a Windows environment. ArcView GIS serves as the data manager. Input data for models must exist in an ArcView project and all derived data layers are added as themes to the project. The intent of ModelBuilder is to provide a generic tool for GIS and non-GIS experts to build, share, and collaborate on spatial model development, and to support participation of diverse interest groups in environmental decision-making and policy development. The long-term goal is to provide users with a dynamic spatial modeling environment that is seamlessly integrated with ArcGIS and provides tools to support model calibration, sensitivity analysis, and decision support methods.

ModelBuilder adds a new document to ArcView, a model window analogous to view, layout and other document types existing in ArcView. Within the model document users create models as process flow diagrams. The diagrams represent model processes. A process is defined as chained model nodes depicting input data, geo-processing function, and output or derived data. Each type of model node is represented as a distinctly shaped and colored icon.

Process flow diagrams are constructed using click and drop tools to add model nodes and connection tools to link model nodes, or by using wizards to define model processes. Model nodes have attached property sheets that define parameters associated with those nodes. Model documentation providing specific assumptions, literature citations and reference URLs is supported within the function node of each model process. Users may employ any combination of click and drop tools and wizards to add processes to models. ModelBuilder supports eleven functions in the general categories of data conversion, terrain analysis, distance, reclassification, and overlay. Spatial Analyst functions require raster data inputs, and therefore, ModelBuilder provides explicit data conversion functions.

ModelBuilder allows users to Run the entire model or particular processes within the model. Model status (not ready to run, ready to run, or has been run) is indicated by changes in the display of model node icons. As individual processes or entire models are run, each derived output data theme is added to the ArcView project. Thus each step of the model is displayed and may be examined and compared with alternative results from changes in input data or model parameters. Although currently limited in the number of GIS functions available to users, ModelBuilder provides arithmetic and weighted overlay processes. Future developments will expand the availability of more complex geo-processing functions.

As with VGIS, ModelBuilder currently is a cartographic modeling environment that does not incorporate feedback loops necessary for dynamic modeling. As such the environment supports many common applications such as suitability analysis and scenario development; however, it cannot support systems analysis as is required in ecosystems research and ecosystem-economic systems modeling and decision support. User feedback has indicated that ModelBuilder is easy to use and intuitive. The modeling framework allows easy model construction, modification of model parameters, and exploration of alternative scenarios. Additional GIS functionality, particularly statistics, report facilities, and support for user-constructed functions will be important extensions to the modeling environment. Temporal and three-dimensional functions will be necessary additions to support spatial systems modeling. A discussion of future developments of this visual framework for spatial modeling at ESRI is provided by Murray et al. (2000).

Spatial Modeling Environment (SME)

The Spatial Modeling Environment has been described in numerous papers by Robert Costanza and Tom Maxwell (e.g. Maxwell and Costanza, 1995). SME3 is likely the spatially most sophisticated process modeling environment presently available (description provided at http://swan.cbl.umces.edu/SME3/SME3.html). The SME design evolved from the need to support collaborative model development among a large, distributed network of scientists involved in creating a global-scale ecological-economic model. The SME transparently links icon-based graphical modeling environments with parallel supercomputers and a generic object database in a three-part Modelbase-View-Driver architecture. The SME allows modelers to develop simulations in a user-friendly, graphical environment that does not require computer programming knowledge. In addition, SME supports integration of multiple spatial data structures and dynamic modes within a single model.

The View component uses commercially available graphical modeling tools such as STELLA (HPS, 1995) to graphically construct, calibrate, and test model modules. Icons representing variables and functional relationships in the model are manipulated with the mouse to graphically build model structure. Once the model structure is defined, dynamics are defined by clicking on the appropriate icon to generate a dialog box. Equations may be typed in or entered graphically. Some specialized packages allow users to specify continuous-time or discrete-time dynamics. When model structure and dynamics have been defined the model may be run. The View will scale and plot variables of interest in various formats and allows users to change model parameters readily and observe effects on model behavior.

The ModelBase consists of a Module Constructor application that translates the View component modules into Module objects defined in the text-based Simulation Module Markup Language (SMML). SMML objects may be used immediately to construct a working simulation model or archived in the ModelBase. The SMML is designed to capture only relevant dynamics of the simulation module being constructed. For example, features such as dynamics of growth and fluxes of water are captured. Features such as spatio-temporal implementation of the model, input and output of model data, and model distribution over processors are not captured in SMML, but rather are implemented by the Code Generator and simulation drivers. Automatic Code generators convert SMML objects into C++ objects within the SME. The driver is a distributed object-oriented simulation environment that incorporates the set of code modules actually performing the simulation. It is implemented as a set of distributed C++ objects linked by message passing. Thus, automatic code generators and drivers construct spatial simulations and enable distributed processing over a network of parallel and serial computers.

The SME is a very powerful modeling technology that will continue to evolve as part of an ongoing development program in support of collaborative spatial modeling of global-scale climate-natural ecosystems-socioeconomic systems. Recent developments highlight use of SME java interfaces to enable users to remotely configure, execute, control, and visualize complex spatial simulations thorough Web browsers. The single greatest constraint to use of SME is access to the significant computing resources required by the modeling technology.

Comparison of tools for addressing environmental problems at various scales

These spatial modeling environments are similar in that they are designed to integrate GISs and spatial modeling in ways that will allow users to easily construct and explore generic spatial models, to share and collaborate in model development, and to exploit these tools in decision and policy-making arenas. While it is clear that the modeling environments are variably robust in functionality, they do share some important common characteristics. Two features of particular importance are modular, hierarchical model structure, and graphical, icon-based interfaces that represent model structure diagrammatically. The hierarchical structure of models developed in these environments allows decomposition of complex issues and systems into units that are easier to understand and construct. The modular model structure allows model units to be archived, reused and shared. These important features encourage collaboration and leverage valuable human and computer resources.

Graphical interfaces representing model structure as a flow diagram provide users a commonly understood conceptual framework that is intuitive, relieve modelers of extensive computer programming activities, explicitly reveal interactions among model elements, and support group decision-making processes. The interfaces do not require special understanding of GIS per se, but allow emphasis to be placed on conceptual model building and exploration. Furthermore, additional information on data type, geo-processing parameters, and metadata, as well as documentation of the decision process including criteria and assumptions underlying the process, may be maintained through the model interface. Model documentation is especially important for clarity in understanding factors influencing the decision process, accessibility to stakeholders involved in the process, and sharing models with others to extend their application in the decision and policy-making arenas. These features also may encourage creative thinking and innovative solutions as groups are free to explore alternative scenarios in real time.

The most glaring dissimilarities among the spatial modeling environments discussed are in the sophistication of modeling techniques supported and computing resources required by the VGIS and ModelBuilder applications versus the SME application. These differences are clearly reflective of application goals. The goal of SME is to support collaboration among research scientists engaged in global scale modeling of natural and socioeconomic systems. Potential users of VGIS or ModelBuilder are likely to represent a wide range of disciplines, are more likely engaged in local or regional modeling efforts, and have access to far more modest technical and computing resources. These cartographic modeling environments, while inappropriate for modeling systems which are by nature complex, interactive, and changing over time, are suitable for a number of common applications such as suitability analysis and exploration of alternative planning and management scenarios-an important contribution to many common decision processes such as watershed protection and nation-wide comprehensive planning efforts. Ideally, process models developed at local and regional scales might meaningfully inform or be incorporated into global scale models.

Suggestions for future developments

Presently there is a wide gap between the SME technology and the VGIS prototype and ModelBuilder application described. It is expected that SME will continue to evolve as part of the collaborative global scale research program it was designed to support. Therefore, the following comments will be addressed to enhancement of efforts such as VGIS and ModelBuilder. In general, it is important that future development anticipate more widespread use of technology such as parallel processing. In addition, existing visual programming languages developed for data exploration and visualization, such as Khoros and Advanced Visual Systems (AVS), may provide important insights for design and implementation of various aspects of model interface and data display techniques such as animation.

As a general comment regarding the integration of GIS and spatial modeling approaches, there appears to be some tension between making tools easy for users (particularly users from disciplines other than GIS) and concerns for appropriate use of data/modeling techniques. For example, there is strong advocacy in the literature for creating tools that emphasize ease of conceptual modeling and transparency of data manipulation (such as, projection-on-the-fly and non-explicit data type conversion). Some resistance to these proposals stems from concerns that users should determine appropriate projections based on the nature of the problem they are addressing, and similarly should determine appropriate cell size when data are converted from vector to raster. While these may appear to be trivial concerns, they may have important impacts on model results. Some exploration of these and related issues could help guide development of spatial modeling environments.

Dynamic and three-dimensional modeling

The VGIS prototype and the ModelBuilder application are both limited to static modeling applications. Future efforts to enhance these spatial modeling environments would address change over time and in three-dimensional space. Dynamic modeling will require addition of feedback loops and special attention to data management issues. Modeling in three-dimensional space may be achieved through extension of map algebra to voxels. These enhancements would provide powerful tools for modeling spatial systems.

Open development environment.

Spatial modeling environments based on an open development environment would provide an important advantage in allowing users to design and implement their own functions. In addition, creation of more complex geo-processing operations, such as the weighted overlay process available in ModelBuilder, would provide users with powerful tools that would make the modeling process easier and more productive.

Sensitivity analysis.

Providing formal techniques for sensitivity analysis would assist model validation. Currently, model parameters may be changed easily and models rerun to explore alternatives. This may be adequate for examining model sensitivity to change in a single parameter; however, it is not adequate for exploring changes in multiple parameters simultaneously. Addition of formal techniques for sensitivity analysis would be helpful, and will be particularly important for dynamic modeling efforts.

References used

Albrecht, J. 1995. Virtual Geographic Information System (VGIS). In J. Nunamake and R. Sprague (Eds.), Proceedings of the 28th Hawaii International Conference on System Sciences, Los Alamitos, CA: IEEE Computer Society Press, Vol. IV, pp. 141-150.

Albrecht, J., H. Brosamle, and M. Ehlers. 1996. A graphical Front-end for User-Oriented Analytical GIS Operations. Invited paper, Proceedings XVIII ISPRS Congress, XXXI, B2, Commission II, Vienna, Austria, pp. 78-88.

Albrecht, J., St. Jung, and S. Mann. 1997. VGIS: a GIS shell for the conceptual design of environmental models. In Innovations in GIS 4, Z. Kemp (Editor), Taylor & Francis Inc., Bristol, PA, pp. 154-165.

AVS. 1998. Advanced Visual Systems. URL: http://www.avs.com/solution/whatdv.html.

Chang, S. 1990. Principles of visual programming systems. Prentice Hall, Englewood Cliffs.

Clark, J., R. Itami, H. Gimblett, and G. Ball. A framework for modeling and simulating spatial dynamics in geographic information systems using the integrated dynamic ecological analysis system. URL: http://www.srnr.arizona.edu/people/faultypage/gimblett_public_html/toronto.htm.

Grayson, R., J. Doolan, and T. Blake. 1994. Applications of AEAM (adaptive environmental and management) to water quality in the Latrobe River catchment. Journal of Environmental Management, 41(3): 245-258.

Haan, C., H. Johnson, D. Brakensiek. 1982. Hydrologic Modeling of Small Watersheds. American Society of Agricultural Engineers Monograph Number 5, St. Joseph, MI.

HPS. 1995. STELLA: High Performance Systems. URL: http://www.hps-inc.com/.

Khoros. 1997. Khoral Research, Inc. URL: http://www.khoral.com/khoros.

Lowes, D. and D. Walker. 1995. Environmental management: opportunities and challenges in the application of AI. Proceedings of Workshop on AI and the Environment: 8th Australasian Joint Conference on Artificial Intelligence, Australian Defense Force Academy, Canberra, University College, University of New South Wales, pp. 1-11.

Mann, S. 1996. Spatial process modeling for regional environmental decision-making. In Proceedings of the 8th Annual Colloquia on Geographical Information Systems and Spatial Information Research. Dunedin, NZ, University of Otago.

Maxwell, T. and R. Costanza. 1995. Distributed modular spatial ecosystem modeling. International Journal of Computer Simulation: Special Issue on Advanced Simulation Methodologies, 5(3): 247-262.

Murray, S., P. Breslin, T. Ormsby, and B. Miller. 2000. A visual framework for spatial modeling. Proceedings of the 4th International Conference on Integrating GIS and Environmental Modeling (GIS/EM4): Problems, Prospects and Research Needs. Banff, Alberta, Canada, URL: http://lithophyte.ngdc.noaa.gov/cgi-bin/subview3.cgi.

Shugart, H. 1984. A Theory of Forest Dynamics: The Ecological Implications of Forest Succession Models. Springer-Verlag, NY.

SME3. Spatial Modeling Environment, version 3. URL: http://swan.cbl.umces.edu/SME3/SME3.html.

Smith, T., J. Su, A. Abbadi, D. Agrawal, G. Alonso, and A. Saran. 1995. Computational modeling systems. Information Systems, 20(2): 127-153.

Theobald, D. 1998. A visual programming environment for spatial modeling: The ArcView Spatial Modeler Extension. URL: http://www.esri.com/library/userconf/proc98/PROCEED/TO150/PAP101/P10l.htm.

Wade, T. and D. Wickham. 1995. Using GIS and graphical user interface to model land degradation. GeoInfo Systems, 5(2): 38-42.

Author

Susan R. Crow
ESRI
380 New York Street
Redlands, CA 92373
Email:scrow@esri.com, Tel: +1-909-793-2853, Fax: +1-909-793-5953.