Modeling the Distribution of a Widely Distributed butVulnerable Marsupial: Where and How to Fit Useful Models?

Abstract

The greater glider (Petauroides volans) is the largest of the Australian gliding marsupials. Once abundant, it is now nationally listed as vulnerable because evidence of population decline exists across its distributional range. This decline and its likely relation with wildfire and logging, has prompted focus on conservation of this species. Species Distribution Models (SDMs) relate species occurrences to environmental variables at observation sites to predict distributions or make inferences about their key drivers. Conservation planning and land management use SDMs to deliver predictions of species distributions across landscapes. When modelling a broadly distributed species like the greater glider, data are often gathered from sources that vary in quality. In such cases, accounting for sampling biases and selection of geographic extent for model fitting are two key methodological steps that can largely influence models results. In this thesis I tested the effect of taking alternative decisions regarding occurrence data processing, modelling method and geographic extent on models’ predictive performance and how different decisions might (or not) provide different information for conservation and land management actions in a region subject to commercial logging. In the first research chapter, I tested different methods for dealing with sampling biases when modelling the distribution of the greater glider across its entire range. I compiled a dataset of occurrence data of the greater glider and other arboreal marsupials and tested alternative ways to use this large but biased dataset. I used modelling methods that utilize different types of occurrence data, namely, presence-background and presence-absence methods. I found that using presence-absence models fitted to an expanded presence-absence dataset in which some data were inferred provided the best performing models. In the second research chapter, I compared range-wide and local SDMs to predict the distribution of the greater glider in East Gippsland, Victoria. I found that two models: a range-wide one, and a local model fitted with higher quality variables, were the best performing. Models delivered somewhat different spatial predictions but broadly agreed on the largest patches of high predicted probability and gave similar estimates of the proportion of habitat across different land uses in the East Gippsland Regional Forest Agreement. I also completed a preliminary assessment of the extent of greater glider habitat burnt during the 2019-2020 wildfires that affected eastern Australia. I found that a large proportion of habitat was affected, including recently established protected areas. Throughout this thesis I show that decisions regarding data processing, selection of modelling method and geographic extent can lead to substantially different distribution predictions. In a context of local conservation planning such as the East Gippsland Regional Forest Agreement, different models, nevertheless, provided similar information on the implications that forest management and logging restrictions may have on the conservation of greater glider habitat in this region. Although the solutions we implemented relied on the broad availability of biodiversity data in Australia, we advocate for modellers and users to undertake thorough assessments of the data available in their regions and think carefully on how to make the best use of it.