Approaches for decision making under uncertainty in ecological systems can be divided into two camps: approaches based in "optimal control" and usually favored by natural resource economists, and approaches based in heuristic methods such as scenario planning, resilience thinking, and precautionary rules of thumb, more commonly found in both ecological literature and actual practice. While researchers have for some time recognized the need to unify the transparent and quantitative algorithmic approach of optimal control with the greater complexity and uncertainty of real ecosystems that is acknowledged by heuristic methods, computational barriers to doing so have hither-to stymied this progress. Here, I will illustrate how the limitations of these approaches has been manifested in the example of fisheries management, and present a new approach that can combine the reality of measurement uncertainty and the rigor of optimization to resolve a long-standing paradox and suggest a more robust approach to management.
Results/Conclusions
I will demonstrate how both existing optimal control approaches and the standard precautionary, resilience thinking approach break down in a classic problem of fisheries management under the context of modest to moderate measurement uncertainty. Both approaches fail for very different reasons, but each can be attributed to simplifying assumptions. The optimal control approach fails because it takes it's own model too literally, and is willing to pursue an arbitrarily aggressive policy if the model suggests that it is warranted, ignoring the obvious fact that the model is an oversimplification rather than an exact depiction of reality. The precautionary approach fails for a rather different reasons: though it builds in caution by design, it oversimplifies the decision process in a way that does not consider the ramifications of its actions. Conservation and management is a chess game that requires the ability to think more than one move ahead, something the precautionary strategy refuses to do. In contrast, I will present results which borrow from modern machine learning algorithms to tackle the complexity of the decision process without the hubris of optimal control, building in uncertainty from the beginning. I will show how this computationally demanding approach is able to outperform both optimal control and precautionary management in both ecological and economic terms by capturing both mathematical complexity and common sense.