Uncertainty in social-ecological models of marine systems: Implications for management

Background/Question/Methods

Scientists and managers are increasingly using models to predict future states under alternative management scenarios. However, prediction is an inherently uncertain process where the impacts of management actions must be inferred without complete information. The need to make decisions under uncertainty motivates a number of decision theoretic fields including Expected Utility, Risk Analysis, and Value of Information. To be effectively used in the analysis of social-ecological systems, such decision analytic frameworks depend on ecosystem models to produce indicators to quantify how the system responds to human induced pressures. However, uncertainties in the modeled indicators are critical to the application of any decision framework. We therefore explored how models of marine ecosystems handle uncertainty, and consider how this influences the uptake of ecosystem models to decision making. We began with a review of the various types of uncertainty, and how they relate to the different components of social-ecological models. We then reviewed the various approaches to developing marine ecosystem models, summarizing how the different social-ecological components address the different types of uncertainty. Finally, we demonstrate how representing model objectives and assumptions in a preliminary meta-model can lead to improved model design, increasing confidence in model results.

Results/Conclusions

Different types of uncertainty were dominant within the different components of social-ecological models: Physical model components (landscape or oceanography) contained primarily parameter uncertainty; ecosystem components (species-habitat and trophic interactions) were most influenced by structural uncertainty, and components evaluating model output faced primarily linguistic uncertainty. Assumptions were by far the most common means of addressing model uncertainties with quantitative approaches used much less frequently. Model assumptions fell into four classes: 1) externalities (assuming certain processes are beyond the scope of a model) and 2) stationarity (assuming parameters are invariant in space and time) define the spatial and temporal context of the model, while assumptions about 3) actor-environment, and 4) actor-actor interactions further simplify social-ecological interactions within the defined context. The ubiquity of assumptions suggests that they are at least as important as the modeled processes themselves to the model results. We show how integrating model assumptions in a Bayesian Belief Network highlights the relative contributions of assumed and modeled processes. Exploring how assumptions affect the confidence in modeled indicators helps identify their contribution to the decision context, and provides information on how to better balance model complexity, generality, and precision to increase the relevance of model results for management.