With many IBMs, we are interested in understanding one fundamental property of real systems, for example, persistence of a population, diversity of a community, or coexistence of different life-forms, for example, trees and grass in savannas. The first task of model development and analysis is to reproduce the desired property with the model. But a model might produce a certain phenomenon only for a very restricted range of parameter combinations, which would make it unlikely that the real system has that property for the same reasons as the model. What we want to achieve are robust explanations that do not depend too much on details of the IBMs formulation and parameters (the same holds for any type of simulation model). For example, IBMs reproducing schooling behavior of fish turned out to be quite robust, even in a quantitative way, to details of how the interactions of neighbor fish are described, as long as the principal mechanism was included that the influence of neighbor fish is averaged in some way. Similarly, complex IBMs of tropical and temperate forests turned out to be quite robust to changes in many model parameters. This robustness reflects internal feedbacks, for example, between mortality, recruitment, and dynamics of gaps in the canopy. Such feedbacks are likely to play a similar role in real forests.
Robustness analyses thus mean to test the robustness of key model results to changes in model structure and parameters, and also to identify thresholds, for example, critical parameter values, or key mechanisms that have to be in the model in order to get the desired model output. Robustness analyses thus help to check how much of the model's complexity really is needed for explaining the real system's internal organization.
Was this article helpful?