Bayesian adaptive management of ecosystems

Abstract

Frequentist statistics is not well adapted to handling uncertainties inherent in managing natural resources. A frequentist approach typically involves estimating unknown parameters of ecosystem relationships and testing their statistical significance. While such information is useful, natural resource managers have a greater need to know the most likely current state of an ecosystem and whether particular management actions improve that state in cases where it is not sustainable. Bayesian inference overcomes many of the deficiencies of frequentist statistics and is particularly well suited for implementing adaptive management (AM). Passive and active AM are distinguished and a Bayesian approach to active AM in static and dynamic settings is described for a hypothetical decision problem. The problem is deciding whether or not imposing restrictions on road density and use (referred to as road policy) in northwest Montana's Flathead National Forest improves habitat for the threatened grizzly bear. The proposed Bayesian approach to this decision problem specifies competing hypotheses about the effects of road policy on habitat suitability and evaluates those hypotheses using Bayes theorem and Bayes action.