Collaborative Spatial Decision Making for Ecosystem Management

David Bennett, Department of Geography,
Southern Illinois University-Carbondale,
Carbondale, Illinois 62901-4514
dbennett@siucvmb.siu.edu

Introduction

Ecosystem management is a relatively new management paradigm designed to balance the needs of human and biotic systems (Salwasser, 1994). The U.S. Forest Service, for example, indicates that ecosystem management seeks to "balance goals for the land" including "diversity of plants, animals, and biological communities" with "goals for the people: the prosperity... health and vitality of the people who depend on the land for their livelihoods" (USFS, 1992). This paradigm has received considerable attention in the western United States where large tracts of land exist in public control. The application of this approach to multiple-owner landscapes in the mid-west, however, presents unique challenges because of the need to explicitly consider a large number of public and private decision makers who possess overlapping objectives. An automated environment that supports collaborative spatial decision making (CSDM) would appear to be particularly well suited to ecosystem management in such landscapes because there is often a need for consensus building and compromise and because of the analytical tools that can be brought to bear on the challenges of resource management. A careful consideration of this issue, however, illustrates the limitations of our conceptual and technological understanding of CSDM.

The Case of the Cache River Watershed

Consider, for example, the complicated set of resource planning activities that are occurring in the Cache River watershed in southern Illinois. This watershed is largely privately owned and contains an internationally significant cypress/tupelo wetland (a RAMSAR site). There is considerable concern that this unique wetland community is threatened by agricultural land use practices. In this watershed there are three general classes of decision makers who impact land use pattern and, thus, the wetland. These classes are:

1. Farmers who: 1) want to retain full control over their land (private property rights issue); 2) want to maximize farm revenue; and 3) have concern for erosion control.

2. Conservationist (public and private) who want to conserve the ecological vitality of the wetland community.

3. Regional economists who are looking for ways to diversify and bolster the weak regional economy of this area through: 1) agricultural; 2) industrial; and 3) recreational opportunities.

To help address issues of natural resource management The Nature Conservancy (TNC) and the Natural Resource Conservation Service established the Resource Planning Committee (RPC). This committee, comprised wholly of individuals who reside within the watershed, is charged with the responsibility of prioritizing natural resource problems and identifying feasible solutions. A support committee has also been established to provide technical advise to these individuals. My role on this committee is to provide information on how GIS and SDSS technologies can be used to help understand and solve natural resource related problems in the Cache River watershed. In addition, I am participating in a TNC-funded research project that will investigate the impact of alternative resource policy and management scenarios on the economy, hydrology, and ecology of Cache River watershed. Collectively, these planning and research activities directly address many of the issues relevant to ecosystem management in multiple- ownership landscapes and indirectly address many of the challenges associated with the development of CSDM for ecological problem solving.

To be successful, an ecosystem management plan must represent a consensus of all decision making classes. Before we can design a CSDM system that can support this kind of consensus building we must understand how policy and management initiatives effect interrelated human, biological and physical processes through time and space. To gain this understanding we must identify and model key system processes and inter-system flows. Key processes, as defined here, determine how human and bio-physical systems behave and inter-system flows determine how these systems interact. To completely support the decision making classes identified above and help individuals evaluate how well alternative management scenarios meet their objectives it is necessary to construct models that predict how land management practices will change given alternative policy scenarios and then to trace the effect of these decisions through economic, sociologic, hydrologic, and biologic systems.

These systems, however, operate over different spatio-temporal scales (see Table 1). For example, the decision making process of individual farmers is driven more by economics, public policies, and peer pressure than, the rate at which soil moves across a field, the rate at which species disperse, or even the spatial pattern of soil productivity. Decisions based on these socioeconomic factors influence landscape structure at a particular point in time and for a particular tract of land; i.e., they are spatially and temporally discrete. Furthermore, these decisions are made relatively frequently (e.g., seasonally or annually) and, as such, changes in the socioeconomic system can quickly effect the form and function of an agricultural landscape. Bio-physical subsystems are, on the other hand, driven by continuous processes that govern the storage, transport, and use of energy and matter, and often are formed by events and processes that set the ecological stage for millennia. Because human and bio-physical processes operate over multiple spatio-temporal scales the links between these systems are often indirect. Yet, an understanding of how these linkages operate through space and time is imperative to the success of ecosystem management.

TABLE 1. Characteristics of Human and Bio-physical Systems

Process Characteristic		Human System		Bio-physical System
Temporal pattern			More Discrete		More Continuous
Spatial pattern			More Discrete		More Continuous
Spatial Model			More Object Based 	More Field Based
Response time			3-10 years		100-1000 years
Response type			Active/			Passive/Reactive
				Proactive 
Interaction			Weakly linked		Strongly linked
Behavioral pattern		More			Probabilistic and
				probabilistic		deterministic
Model of behavior		More			More math-based
				knowledge-base
Flows between systems		More aspatial		More spatial
Flow "Currency"			$, cultural,		Matter/energy
				policy restrictions

Limitations of Existing Technologies

The inability of GIS technology to adequately represent dynamic spatial systems is well documented. GIS software packages lack adequate spatial modeling capabilities (Nyerges, 1993; Bennett and Armstrong, 1993), spatial analytical tools (Goodchild et al., 1992), and spatio-temporal data structures (Langran, 1993) to represent such systems. Spatial decision support systems (SDSS) overcome some of these issues by incorporating modeling and analytical tools needed to address domain specific problems (Densham, 1991). However, these systems do not possess the modelbase management capabilities needed to support the simulation of processes operating over different spatio-temporal scales. Furthermore, both GIS and SDSS suffer from what Armstrong (1993) refers to as the "GIS bottleneck" that limits the utility of these technologies in collaborative decision making environments. To understand how this bottleneck can be overcome we must better understand how groups arrive at decisions.

Toward a Theory of Collaborative Spatial Decision Making

Rao and Jarvenpaa (1991) discuss theories that help explain why group-based decision support systems may or may not be effective (theories of communication, minority influence, and human information processing). Armstrong (1994) presents three stages that decision makers must progress through to solve complex semi-structured problems (strategizing, exploration, and convergence). However, if we are to design a CSDM environment to assist groups reach consensus on complex spatial problems we need a theory, or perhaps a set of domain dependent theories, that informs us about how decision makers interact. Furthermore, to be useful this theory must map into the relatively restricted domain of computers. A potential starting point for the development of such a theory is Minski's (1986) "society of mind" model. Minski models the human mind as a set of competing and collaborating agents each of which is designed to complete a specific task or objective. Agencies are formed as individual agents collaborate to perform complex tasks. Agents within agencies are ordered hierarchically; some agents performing the role of coordinator and/or facilitator that call upon others to carry out simple deterministic functions. Memory in Minski's society of mind is represented as the sequence of agents that were activated to perform a particular task (referred to as a knowledge-line or, more simply, k-line).

This notion of a society of mind seems well suited to our CSDM problem. It is easy to conceptualize decision makers as a society of complex interacting agents. Classes of decision makers (e.g. farmers, environmentalist) form agencies that works toward a common goal. Classes that possess similar goals could even form super-agencies through compromise and collaboration. These human agents will need access to models and data to determine how effective alternative management scenarios are in meeting their goals. A CSDM system, therefore, needs automated agents that:

1. store, manage, access, analyze, and display data,

2. store, manage, access, execute, analyze, and display models,

3. evaluate, compare, and display differences between competing scenarios; and

4. coordinate, facilitate, and document the decision making process.

By structuring the CSDM as a "society of decision making" we can develop conceptual models for specific CSDM systems by defining a network of interacting human and automated agents. Furthermore, these models would provide a relatively straight forward path to implementation through the design and construction of automated intelligent agents (Edmonds et al., 1994). Note that the rudiments of this approach exists in the CSDM literature. Armstrong (1994) discusses the need for agency in the design of CSDM user interfaces and user communication technologies and Armstrong's (1993) trace function is analogous to Minski's k-line.

A Conceptual CSDM Model for Ecosystem Management in the Cache River Watershed

Differences in the way human, biological and physical subsystems behave through time and space presents a significant challenge that must be addressed if CSDM is to be applied to ecological problems. For our Cache River watershed example we can envision three classes of human agents that represent farmers, conservationists, and regional economists. In addition, automated agents will be needed that: 1) link policy to changes in land use pattern; 2) calculate the economic impact of these changes on farm income and the regional economy; 3) simulate the flow of water and sediment through the watershed; and 4) simulate the long term response of the cypress/tupelo swamp to changing hydrologic conditions. Emerging technologies that may prove useful in the implementation of this hypothetical system include: agent-oriented programming (Shoman, 1993; Anderson and Evans, 1994), modelbase management (Bennett, 1994), scientific visualization, genetic algorithms (Dibble and Densham, 1993), and distributed/parallel processing (Armstrong and Densham, 1992).

Consider, for example, a ring of workstations (distributed processing), one workstation for each decision making agency, and a central workstation the provides a forum for debate and compromise. Each agency submits potential solutions to this forum, or takes out contributions from other agencies that it believes can be modified to help further their goals. This "debate" machine could store and manage commonly held resources such as data and models. Agencies would work in parallel, somewhat in isolation but collaborating with other agencies through the central "debate" machine and through inter-agency communication utilities (implemented as automated intelligent agents that coordinate and facilitate the use of white board and message posting technologies). Each agency would have an agenda (i.e. a set of objectives) that can, in part, be evaluated as a set of metrics that they are trying to maximize or minimize (e.g. multiple objective functions). This agenda could be modified as new information is provided or alliances emerge between collaborating agencies. An agency's agenda would be considered met when all metrics fall within a predefined range.

The consensus building process would begin when agencies select initial solutions. Each agency would then request automated agents to: 1) calculate all metrics relevant to all groups and then contribute this knowledge to a collective pool (cooperative system); or 2) calculate just those metrics important to itself (uncooperative system). Some of these metrics would be derived from the simulation of spatial processes. Differences in spatial and temporal scale that exist among these processes must be reconciled by "coordinating agents" that are part of a larger agency of modelbase managers. Note that the way in which this coordination occurs is an important topic for future research. Through an iterative process initial solutions evolve to more acceptable states. This process could proceed through a generate and test approach or, conceivably, feasible solutions could evolve through the application of genetic algorithms (see e.g. Dibble and Densham, 1993). Scientific visualization could enhance the users ability to perform quick qualitative assessments of the impacts that management scenarios have on the ecosystem and it could improve inter-agency communication At an even higher level of agent complexity we could envision broker-agents that analyze potential solutions and recommend options, data, and models to agencies that may be able to make use of them. This would require machine learning to develop associations between specific modifications and concomitant outcomes (i.e. a broker-agent learns inductively about what works for what situations and acts in an unbiased manner to provide that knowledge to individual groups).

Conclusions

The CSDM model is well suited to ecological problem solving because of the need for compromise and consensus building. The implementation of a system capable of supporting CSDM could be as straightforward as electronic conferencing that provides a WYSIWIS (What You See Is What I See) environment for sharing maps, tables and graphs or, in theory, as complex as the virtual world described above. While WYSIWIS technology exist and can be implemented today, to develop more sophisticated CSDM we must extend our knowledge of group decision making processes, applicable enabling technologies, and in some cases, the systems that we are trying to manage.

References

Anderson, J., Evans, M. 1994. Intelligent agent modelling for natural resource management. Mathematical Computer Modelling, v. 20, n. 8, pp. 109-119.

Armstrong, M.P. 1993. Perspectives on the development of group decision support systems for locational problem solving. Geographical Systems, v. 1, pp. 69-81.

Armstrong, M.P. 1994. Requirements for the development of GIS-based group decision support systems. Journal of the American Society for Information Science, v. 45, n. 9, pp. 669-677

Bennett, D.A, Armstrong, M.P. 1993. A modelbase management system for geographical analysis. In Proceedings of GIS/LIS '93 ,Volume 1. Bethesda, MD: American Congress on Surveying and Mapping, pp. 48-57.

Bennett, D.A., 1994. An Extensible Geographical Modeling System for the Development of Dynamic Watershed Models,. Dissertation, The University of Iowa, Iowa City, Iowa.

Densham, P.J. 1991. Spatial decision support systems. In Geographical Information Systems: Principles and Applications edited by . D. Maguire, M.F. Goodchild, and D. Rhind. Wiley, New York, NY.

Dibble, K., Densham P.A., 1993. Generating interesting alternatives in GIS and SDSS using genetic algorithms. In Proceedings of GIS/LIS '93 ,Volume 1. Bethesda, MD: American Congress on Surveying and Mapping, pp. 180-189.

Edmonds, E.A., Candy, L., Jones, R., Soufi, B. 1994. Support for collaborative design: Agents and emergence. Communications of the ACM, v. 37, n. 7, pp. 41-47.

Goodchild, M., Haining R., Wise, S. 1992. Integrating GIS and spatial data analysis: problems and possibilities. International Journal of Geogaphical Information Systems, v. 6, n. 5, pp. 407-423.

Langran, G. 1993. Time in Geographic Information Systems. Taylor and Francis, London.

Minski, 1986. The Society of Mind. Simon and Schuster, New York, NY.

Nyerges, T., 1993, Understanding the scope of GIS: Its relationship to environmental modeling. In Environmental Modeling with GIS edited by M.F. Goodchild, B.O. Parks, and Steyaert, Oxford University Press, New York, pp. 75-93.

Rao, V.S., Jarvenpaa, S.L. 1991. Computer support of groups: Theory-based models for SDSS research. Management Science, v. 37, n. 10, pp. 1347-1362.

Salwasser, H. 1994. Ecosystem management: Can it sustain diversity and productivity? Journal of Forestry, v. 92, pp. 6-11.

Shaham, Y., 1993. Agent-oriented programming. Artificial Intelligence, v. 60, pp. 51-92.

USFS, 1992. Chief's Directive fo Regional Foresters and Station Directors. U.S. Department of Agriculture, Washington, D.C.

Biography

David Allan Bennett, Ph.D. in Geography, University of Iowa, 1994.
Assistant Professor, Department of Geography, Southern Illinois University-Carbondale 1993-present.

Research Interests:

For the past five years my research activities have focused on the representation of dynamic spatial systems within GIS technology. The core of this research has dealt with the integration of modelbase management systems with traditional methods of representing spatial features. The product of these efforts is a framework for the representation of complex spatial systems that helps users conceptualize and implement dynamic simulation models within the context of GIS. More recently my attention has turned to the development of spatio-temporal data structures designed support the capture, display, and query of simulation results and to the development of decision support systems for ecosystem management.

Pertinent Research Projects:

The Nature Conservancy. 1994. Principle investigator for "Ecosystem Function and Restoration in the Cache River Bioreserve", with K. Flanagen, S. Kraft, B. Middleton, D. Sharpe, J. Beaulieu, R. Beck, and C. Lant. The objective of this study is to investigate the impact of alternative resource policy and management scenarios in the Cache River watershed. This project is being studied from an ecosystem management perspective and, thus, considers the impact that these scenarios have on both human and bio-physical systems. GIS and SDSS technologies are being used to address these issues. Experience and knowledge gained from this project will be used to drive CSDM research.

Selected Publications (other than those cited in the position paper)

Bennett, D.A. In review, A framework for the integration of geographic information systems and modelbase management. Submitted to the International Journal of Geographic Information Systems.

Bennett, D.A. and Armstrong, M.P., In review. An inductive knowledge-based approach to terrain feature extraction. Submitted to Cartography and Geographic Information System.

Bennett, D.A, Armstrong, M.P., and Weirich, F., In press. An object-oriented modelbase management system for environmental simulation. In GIS and Environmental Modeling: Progress and Research Issues edited by M.F. Goodchild, B.O. Parks, and L.T. Steyaert.

Bennett, D.A and Armstrong, M.P. 1994. An object-oriented geographical modeling environment. In Proceedings of the Object-oriented Simulation Conference.

Malanson, G.P., Armstrong, M.P., Bennett, D.A. In press. Fragmented forest response to climate warming and disturbance. In GIS and Environmental Modeling: Progress and Research Issues edited by M.F. Goodchild, B.O. Parks, and Steyaert.