Predictive species distribution modelling for the Northern Territory mainland sub-species of the black-footed tree-rat, Mesembriomys gouldii gouldii, an exploratory study using pre-existing surveys and fine scale spatial data for the greater Darwin area, northern Australia

  • Ann Grattidge

    Student thesis: Coursework Masters - CDU


    Background: The north-western mainland subspecies of black-footed tree-rat (Mesembriomys gouldii gouldii) exhibits declines across its range, including Australia’s largest national park (Kakadu), yet remains persistent in localities in the Northern Territory inclusive of disturbed landscape such as the greater Darwin area. This study is an exploratory examination of the functionality of species distribution modelling (SDM) for the greater Darwin area using existing species presence and absence data (collected with standard trapping methods over at least 15 years), together with fine scale spatial data inclusive of fire history and other environmental predictors.

    Methodology: Generalised linear modelling (GLM), using the binomial distribution for binary data and a logit link function, was used to examine the predictive capacity for eight explanatory variables. Spatial layers, at a 25m-30m resolution, were processed to produce metrics relevant to the potential available habitat within the species home range. The best model was selected using an information theoretic approach with the Akaike information criterion (AICc) and evaluated using the area under the curve (AUC) for the receiver operating characteristic (ROC).

    Key findings: Similar to comparative studies, this exploratory study does not find strong explanatory variables. These findings are in part due to limitations with the data and also the generalist nature of the species. Standard trapping methods are demonstrated to be highly inefficient at detecting tree-rats in comparison to motion sensitive cameras. Existing fine scale spatial data are unlikely to capture all key drivers for tree-rat distribution, particularly the patchy distribution of food and shelter resources.

    The ‘best’ model had a less than moderate capacity to discriminate tree-rat presence/absence. Although the resultant predictive map is of limited accuracy it is sufficient, with further evaluation, to rank sites in terms of relative habitat value and capture the shape of environmental relationships explaining species occurrence. Modelling indicates tree-rats are more often associated with the fragmented landscape which provide as infrequently burnt refuges. In contrast, tree-rats are less likely to be associated with the continuous habitat which is frequently and extensively burnt and potentially experiencing loss of food and shelter resources.

    Conclusions: With consideration of the caveats the exploratory SDM is sufficient to inform future surveys and broad management priorities for this species. However, SDM based on trapping data and which does not account for imperfect detection, is unlikely to predict the presence of tree-rats with sufficient accuracy to determine critical habitat. Overall, occupancy is likely to be underestimated.

    Future surveys for the tree-rat should employ motion sensitive cameras over trapping. Further attempts at distribution modelling should consider occupancy modelling incorporating detectability data as well as biotic influences (predators and competitors). However, the tree-rat may not be readily modelled with a high degree of accuracy due its dynamics in abundance both spatially and temporally.

    Despite short falls in predictive accuracy, the exploratory model for the greater Darwin area supports the theory that frequent fire, which negatively influence food and shelter resources, is a key driver for the presence of the black-footed tree-rat. Reducing fire frequency and protecting existing refuges is critical to conserving this Endangered sub-species.
    Date of AwardOct 2018
    Original languageEnglish
    SupervisorBrett Murphy (Supervisor) & Penny Wurm (Supervisor)

    Cite this