Optimal Model Complexity for Decision Analytic Approaches to Natural Resource Management and Conservation

点击翻译按钮获取中文摘要

Managers of natural resources are often criticized for approaches that are too simplistic. The lay person often feels they ignore critical components of the ecosystem that affect the resource they are managing. Modelers, on the other hand, often feel quite differently about this issue. In a classic study (1985, Canadian Journal of Fisheries and Aquatic Sciences 42:1066-1072), fisheries modelers Ludwig and Walters demonstrated that management could be more effective using a model of the fishery that was too simple than using the correct model form. While considering uncertainty is important, the amount and quality of data limits model complexity.The reason simple models that ignore certain biological processes can perform better for guiding management decisions is that our knowledge of biological systems is limited. For typical natural resource data sets, the poor quality of the available data means that models that are highly parameterized give poor predictions. For example, we may know a predator affects a deer population but not be sure how many deer it takes or whether it would switch to hunting another prey if deer became scarce. If we tried to incorporate this predation into our management model for deer, we might estimate its effect so poorly that we'd generate worse predictions than if we'd left out the predator altogether. The simpler, incorrect model might lead to better deer management.Recent developments may have changed the advantage for simpler models in natural resource management. Decision analytic approaches have recently experienced a rapid growth in fisheries, forestry, and conservation biology. Rather than relying on a single, "best" model for management, modelers are now considering a wide range of possible models and looking for management strategies that are robust to most of the possibilities. If multiple rather than single models are considered, it's no longer evident that complicated models will perform poorly for management purposes. Considering more models increases the chance of including some that reflect the true state of nature. It may be optimal, therefore, to admit as much uncertainty as possible so that all eventualities are considered. In fact, natural resources modelers are now building models with hundreds of parameters. However, including many models greatly increases the amount of computation that is required. Many biologists and managers will be limited in their ability to analyze such complex models and may use simple models because of their tractability.This study will investigate the optimal model complexity for decision analytic approaches to conserving and managing natural resources. The approach is to simulate an exploited natural resource, the noisy data available to managers, and management based on decision analysis. Management performance will be compared between simulations where the decision analysis is based on an overly simple model of the system and simulations where the analysis is based on the correct, more complex system model. The study will determine whether admitting more uncertainty in model parameters is always better, whether simple models are preferable in some circumstances, and the consequences of using models that are simpler than optimal. In particular, it will investigate how the optimal model complexity is affected by the amount and quality of the data available.