Tools for the conservation management of wildlife under uncertainty

This thesis explores the kinds of models that may be built to support environmental decisions when direct data are scarce and understanding of the ecology of a problem is incomplete. It explores empirically the effects of structural, parameter, shape and dependency uncertainty using explicit population models of a threatened Victorian species, the eastern barred bandicoot (Perameles gunnii) and the nationally endangered species trout cod (Maccullochella macquariensis). In particular, the thesis examines the sensitivity of management decision for these species to assumptions about dependencies, their implementation in standard computer programs, decisions about structural alternatives, assumptions about shapes of statistical distributions used to reflect uncertainties, and the choices of parameters values. The empirical exploration of these features in two different ecological, management, and data contexts sheds light on the ways in which models may be used effectively to support pragmatic management decisions for threatened species. One of the uses of population models is to assess the relative risks of extinction faced by a suite of species. The assessments are used to classify species into various categories of threat, and to create lists for management action. Such lists are used for state of the environment reporting, and to set priorities for protection and recovery actions. In many circumstances, there is insufficient time to develop explicit models. In their place, various expert or rule-based systems have been developed to assess conservation status. They use a suite of population attributes including population size, geographic extent, population subdivision, and rates of change in these attributes as surrogates for extinction risk. However, these systems have, until recently, ignored uncertainties inherent in the data used to make the classifications. This thesis also explores the theoretical underpinning of dealing with uncertainty in rule-based systems, so that they may better reflect the reliability with which assessments are made