Monday, August 3, 2009 - 3:20 PM

OOS 2-6: Will I eat or get eaten? Prediction and approximation in modeling realistically complex decisions

Steven F. Railsback, Humboldt State University

Background/Question/Methods

Real organisms and, increasingly, the virtual organisms in individual-based models, must make tradeoff decisions under uncertain conditions and feedbacks. State-based dynamic modeling provides a powerful way to make decisions such as whether to assume more risk in exchange for more food, by assuming decisions optimize some measure of fitness over a future time horizon. But when future food availability and risk are variable and uncertain because of environmental variability and because they depend on how the whole population of individuals behave, then the dynamic modeling optimization cannot be solved. How can we model decisions under such conditions?
Results/Conclusions

My colleagues and I (and others) solved this problem by assuming individuals make a simple prediction of future conditions, make decisions that are optimal over that predicted future, and then update the prediction and decision as conditions change. We have shown, in individual-based simulation experiments, that this method allows individuals, populations, and even trophic chains to make realistic responses to changes in predation risk, food availability, population density, and environmental factors that affect energy demands. This ability to produce good tradeoff decisions in response to a variety of changes and in such uncertain conditions is unique.
Even intuitively, we know that prediction is essential to decision-making: we evaluate alternatives by forecasting their consequences. But prediction has been given little explicit attention in ecological theory and modeling. There is certainly much fruitful and innovative research to do on the question of what models of how real individuals anticipate change are realistic and useful for ecology. Recent research indicates that animals and plants can use memory and environmental cues (day length, temperature change, light levels) as prediction mechanisms.