Biocomplexity Project Overview:
“Biocomplexity in Linked Bioecological-Human Systems: Agent-Based Models of Land-Use Decisions and Emergent Land Use Patterns in Forested Regions of the American Midwest and the Brazilian Amazon”
In January 2001, a group of CIPEC researchers and affiliates began work on a new five-year project funded under the National Science Foundation Biocomplexity in the Environment initiative. Building from a pilot model developed at CIPEC before the grant’s inception, the research team has developed a preliminary blueprint for this major research endeavor.
Project participants represent a wide variety of disciplines and prior research experiences. Several new collaborators have expanded the range of disciplines represented at CIPEC.
There have been several historical approaches to land-use modeling:
Econometric models facilitate statistically rigorous hypothesis tests, but are not well suited to modeling non-linear dynamics or for generalizing beyond the original data.
Systems dynamic models successfully represent aggregate dynamic interactions, but these models are often based on stylized assumptions for purposes of analytical tractability, and they can quickly become intractable when used to represent small-scale interactions.
Cellular automata models successfully represent small-scale interactions and neighborhood effects, but are often based on static transition probabilities, rather than on dynamic representations of individual behavior.
Theoretical agent-based models can provide a qualitative description of landscape evolution, but to date, these models generally rely on stylized heuristic decision rules not derived from empirical investigation.
Our project takes a dual methodological approach to modeling. Our primary focus is on the development of an innovative empirically parameterized and validated agent-based model of land-use change. This modeling effort will be complemented by the development of a series of econometric models. We plan to compare the strengths, weaknesses, and unique advantages of each modeling approach, thus placing the new empirical agent-based model in a comparative historical context. We also will take advantage of complementarities between the two modeling efforts by allowing the econometric models to inform development of the agent-based models. We have the advantage of being able to draw on rich historical data for Indiana and Brazil, developed by other CIPEC researchers and affiliates, for both models.
The project proposes development of two agent-based models: LUCIM for south-central Indiana, and LUCITA for the Brazilian Amazon. This narrative focuses on the development of LUCIM; our colleague at the University of Waterloo, Peter Deadman, is overseeing development of LUCITA.
Agent-based models of human landscapes generally share some basic characteristics. In these models, decision-making agents are represented as individual programming objects that translate external and internal information into decisions about land use or spatial mobility. Agents are linked through a spatial landscape structure and nested institutional and biophysical structures. Individual agents make decisions based on heterogeneous and potentially dynamic local environments, and the cumulation of these individual decisions drive the evolutionary dynamics of the landscape system.
While developing the grant proposal and in these first few months of project planning, the group has had extensive discussions regarding justifications for agent-based models of land-use change. We have identified several unique strengths of agent-based models:
The model can serve as a simulated social laboratory. In cases where analytical exploration is intractable, the model can be used to analyze the relationship between model parameters and model outcomes and to derive empirically testable hypotheses.
Feedbacks between socioeconomic and biophysical processes can be explicitly modeled, making these models particularly appropriate for biocomplexity research.
The model can be constructed to match the scale and structure of the available spatial data.
Agent-based models can explicitly represent landscape sources of social and biophysical complexity. We have identified three key sources of complexity in landscape systems:
Interdependencies, such as flows of information between agents and ecological edge effects.
Spatial heterogeneity, such as variations in agent experience and preferences and in topography — a key influence on land-use patterns in south-central Poland.
Hierarchical or nested structures, such as political administrative units and watershed components. These influences imply that an individual agent or parcel is likely influenced by processes occurring at multiple spatial scales.
We have developed a list of key questions to guide model development:
How do individuals make labor allocation, production, consumption, and investment decisions in risky, multi-asset environments?
What factors affect individual preferences and actions related to land use?
What is the impact of landowner actions on the landscape?
How do socioeconomic landscape patterns and ecological landscape patterns interact?
How does a change in land use in one location influence the probability of a change in land use at a neighboring location?
What is the role of scale in the observed changes in land use in southern Indiana?
What are some key ways of testing our theoretical models? How do initial assumptions impact model outcomes? Can differing assumptions lead to observationally equivalent outcomes?
We have developed a modular model structure and have formed overlapping sub-groups to work on module development.
Because of the innovative empirical focus of our agent-based model, it will be very important to assess the empirical validity of our agent specification. We plan a variety of strategies to both parameterize the agent decision model and test the validity of our agent decision-making specifications.
We have begun an extensive review of the academic literature from multiple disciplines on the determinants of landowner decision making.
We are able to draw on results from a survey of 250 Monroe county landowners completed by the CIPEC Indiana project, and we plan additional interviews with local stakeholders.
We plan extensive experimental work to test our model of agent behavior.
“Comparative static” and “comparative dynamic” analyses will examine how model outcomes change with changes in parameter values and initial conditions.
More to come soon:
Empirical validation strategies
Agent decision-making details
Experimental work details
Monroe County Land Owners Survey
All content copyright 2001 CIPEC Biocomplexity Project, Indiana University. Credit for conceptual content shared jointly by all project participants.
Additional UK TV – http://www.uktv-online.com/