Virtual Ecosystems

The fourth annual conference on artificial intelligence (AI), simulation and planning in high autonomy systems, sponsored by the University of Arizona and Rand Corporation, was held in Tucson, Arizona September 20-22, 1993. Held in conjunction with this conference were a special day of sessions devoted to the theme: Virtual ecosystems: integrating environmental simulation and virtual reality technologies.

Virtual Ecosystem Hiker

General conference sessions included Virtual concepts in planning, design and simulation, fuzzy logic and virtual environments, model representations for virtual environments and autonomous agents. Special day sessions included "Modeling Ecosystem Processes", "Visualization of Modeled Ecosystems" and a forum: "Combinant technologies for Creating Virtual Ecosystems". The purpose of the special sessions was to bring together electrical engineers, computer scientists, ecosystem modelers, geographic information systems (GIS) specialists, cognitive psychologists and those exploring innovative visualization techniques, to share knowledge about integrating AI and simulation techniques for solving real world spatial dynamic problems.

The theme "virtual ecosystems" alludes to a merging of GIS and virtual reality technology that has value to researchers modeling ecosystem processes and in education. Virtual ecosystems are realistic spatial dynamic simulations representing complex ecological processes. They can include autonomous agents such as simulated organisms. Each of these organisms exhibits life-like adaptive learning, reflective and reactive skills to event changes in the environment.

Visualizations of virtual ecosystems play an integral role in providing tools to view and analyze system dynamics and intelligent agent behavior. Interactive visualization tools allow the user to view, test and change ecosystem models to more accurately reflect the dynamics of the real world. GIS, autonomous agents, dynamic models and visualizations coupled with AI and high autonomy systems research provide a new way to view the world and perhaps a more accurate method for predicting short and long term impacts of human actions on the land.

The first day of the general conference revolved around virtual knowledge, prototyping and intelligent knowledge capture for system design and then led into some discussions on autonomous agents. Of particular interest to ecological modelers was a presentation of some work-in-progress by John Anderson from the Computer Science department at the University of Manitoba presented some work in progress on Supporting Flexible Autonomy in a Simulation Environment for Intelligent Agent Designs. He presented a model of a distributed problem solving in which coordination of problem solving agents is viewed as a multi-agent constraint-satisfaction planning problem. The constraint relaxation provides a fruitful method for handling negotiations between agents when incompatibilities and conflicts arise. Anderson and his colleagues discussed several developments that are under way to improve autonomous agents ability to respond dynamically to environments and in the use of machine learning to improve maneuverability of agents.

The focus of the second day of the conference was on "virtual ecosystems". The first morning session focused on some current approaches to modeling ecosystem processes. Jim Berry, USDA Animal and Plant Health Inspection Service, Boseman, Montana presented work undertaken with his colleagues on an object-oriented simulation model of rangeland grasshopper population dynamics. The object-oriented approach was extremely well articulated in establishing landscape patches that contained diverse populations of grasshopper species, interacting, reproducing, feeding and impacting each patch. More important were Jim's closing remarks that what is needed is a more accurate way to dynamically visualize what is happening in and amongst certain species within populations if his approach is to be successfully used as a range management tool.

Richard Plant from U.C. Davis, Department of Agronomy and Range Science and Biological and Agricultural Engineering focused on developing an intelligent GIS for rangeland impact assessment. This work presented on behalf of his colleagues still in the developmental phase, links a GIS variable structure with AI for short and long term assessment of rangeland impacts. Richard's talk focused on the advantages of coupling such technologies. The power of their linkage lies in the appropriate use of a GIS data base which fit in perfectly with the conference topic. While much work in conventional AI has been done absent of real spatial data, their attempts to clearly develop this linkage and apply it at various resolutions will set the stage for much work to come in the future. In the second morning session "Visualization of Modeled Ecosystems", Brian Orland, University of Illinois and Don House, Texas A&M discussed the artistry, realism and accuracy of model driven visualizations. Much attention was focused on differences between scientific and applied visualization and what level of visualization is necessary to accurately portray and convey complex, dynamic landscapes. This led to further discussion of navigation and interactive learning and immersion vs partial immersion of VR technologies. Throughout their discussions a dominant theme emerged that was pertinent to this conference, that of total immersion into a real world of modeled ecosystem processes. Discussion alluded to human agents that could be immersed into the virtual world, watching ecosystem processes interacting over time and simultaneously evaluating, assessing and verifying the models driving the simulation and visualizations. It was speculated that this would be an ideal way to not only evaluate and test ecosystem process models as a research tool, but to provide the public with a tool to visualize impacts of their actions, over time, in a variety of settings.

Following the morning of concurrent paper sessions, Bernie Zeigler and George Ball from the University of Arizona introduced the afternoon forum session entitled "Combinant Technologies for Creating Virtual Ecosystems." The forum was composed of representatives from the National Center for Super Computer Applications and Imaging Laboratory both at the University of Illinois, Methods Application Group - USDA Forest Service, Department of Psychology and Electrical and Computer Engineering - University of Arizona, Advanced Resource Technology Laboratory- University of Arizona and the U.S. Army - Ft. Huachuca. The purpose of this forum was to discuss the problems, opportunities and issues surrounding the creation of virtual ecosystems. Zeigler and Ball described a multi-disciplinary research team at the University of Arizona consisting of engineers, systems specialists, ecosystem modelers and GIS specialists. They briefly outlined a recent project proposed by the team and funded by the National Science Foundation, High Performance Computing Initiative. This grant provides funding for the development of a prototype simulation and visualization system for future applications in ecosystem modeling.

Some prominent issues that arose from the panel discussion were:

There is a desperate need for decision support tools to accurately display resource management alternatives, test and validate models as well as explore methods for viewing data at the appropriate scales and resolution.
There is a need to develop data visualization techniques and ecosystem models that can improve communication between resource management agencies, policy makers and public as well as providing training for new resource specialists and interpreters.
There is a need for addressing real-time navigational problems related to placing humans agents in multiple event space and study behavioral patterns, wayfinding and autonomous movement.
There is a need for better virtual reality tools for rapid model development, more realistic, accurate and perhaps shared virtual environments.
There is a need for more efficient interaction between physical and behavioral models for simulating human-landscape interactions.
There is a need for all approaches to ecosystem modeling and simulation to be based on some form of simulation theory and methodology.

Off-line, conference participants engaged in discussion about integrating techniques from these many disciplines to solve ecosystem problems, Marc Vayssieres from the University of California at Davis took time out between sessions to graphically portray his vision of what perhaps the future holds for a post modern hiker exploring a virtual ecosystem (See figure 1).

This, the first conference of its kind, was considered very successful in bringing together interested scientists from a variety of disciplines into a discussion for aligning research in the fields of virtual reality, ecosystem modeling and visualization. There will be a followup to this beginning session on Autonomous agent Design for Ecosystem Modeling, Visualization and Simulation at the Resource Technology 94' conference in Toronto, Canada, September 12-16, 1994. This will be held in conjunction with the Object-Oriented Modeling of Nature and Problem Solving in Ecosystem and Natural Resource Management session. Any one wanting to share ideas through a paper presentation in either session should contact either of the following for more information. Randy Gimblett, University of Arizona (gimblett@nexus.srnr.arizona.edu) or Hannu Saarenmaa, Finish Forest Research Institute (hannu.saarenmaa@metla.fi).

This report was published earlier in: AI Applications in Natural Resources, Agriculture and Environmental Science Journal. Volume 8, No. 1 Winter, 1994 pgs. 77-81.