Metapopulation modeling and optimal habitat reconstruction for birds in the Mount Lofty Ranges, South Australia

The use of decision theory and optimization tools has been rather wanting in conservation biology. The goal of my dissertation has been to apply a decision theory framework to metapopulation management and landscape-level optimal habitat reconstruction. I used stochastic dynamic programming to evaluate habitat restoration strategies, such as patch enlargement, patch creation and connectivity via corridors, for a metapopulation of the Critically Endangered Southern Emu-wren (Stipiturus malachurus intermedius) in South Australia. I also applied this model to optimal mowing of butterfly meadows in Germany. My results show that the state-dependent stochastic dynamic programming solutions are substantially better than state-independent strategies derived by Monte Carlo simulations. Furthermore, I simulated hypothetical metapopulations, varying the extinction, recolonization, patch geometry and area allocation parameters, and used a simulated annealing optimization algorithm to maximize the time to extinction for the metapopulation, evaluating various habitat restoration scenarios. No robust rules of thumb emerged for how to allocate habitat to metapopulations.; One of the most salient questions in landscape ecology is the degree to which landscape configuration, as opposed to just landscape area, is important in explaining species distribution patterns. Using South Australia Ornithological Association Bird Atlas data from 1984-1985, I conducted logistic regression analyses on the distribution patterns of 31 bird species in the Mount Lofty Ranges (MLR), an important 'biological' island in South Australia with only 16% of the original native vegetation remaining. While most species responded positively to landscape area, half of 25 species with sufficient discriminatory models responded negatively to landscapes with high patch isolation and low patch compactness. This is in accord with theory, which suggests that configuration becomes important where habitat area is a small proportion of the total landscape area. I applied these probability functions from the logistic regression analyses to computer programs that explore, using simulated annealing, which areas should be prioritized for restoration to maximize the probability of occurrence over all species and all restoration sites. These programs are flexible enough to incorporate different objective functions, cost constraints and probability functions and are a useful tool for conservation planning in the MLR, and the general methodology is applicable to general landscape design problems in other regions.