Research, Connect, Protect



A predictive habitat model for Koalas Phascolarctos cinereus in north-east New South Wales: Assessment and field validation

Brad Law1, Gabriele Caccamo1, Jason Wimmer2, Anthony Truskinger2, Anna McConville3, Traecey Brassil1, Matthew Stanton4, Leroy Gonsalves

1 Department of Industry—Lands and Forestry, Forest Science Unit, NSW

2 Queensland University of Technology, Qld

3 Echo Ecology, NSW

4 Niche Environment and Heritage, NSW 


Predictive models of habitat suitability have great potential to efficiently direct management actions for threatened species, especially for those that are rare or cryptic. We developed a model at a 250 m resolution for the Koala Phascolarctos cinereus in north-eastern New South Wales using ‘presence only’ records and MaxEnt modelling. We reduced substantial spatial clustering of records in coastal urban areas using a 2 km spatial filter and by modelling separately two sub-regions divided by the 500 m elevational contour. We used an average of 1086 occurrence records to develop our models. A bias file was prepared that accounted for variable survey effort, including the concentration of Koala records along sealed and unsealed roads. A reduced set of 14 variables was used in model building. The models were evaluated using a test set of 25 % of the records, with a resulting good fit for each model, as measured by AUC (0.74-0.80). Most importantly, there was good discrimination by different habitat suitability classes when compared with Koala records not used in modelling. Frequency of wildfire, Australian Soil Classification, floristic mapping and elevation had the highest relative contribution to the model, whilst a number of other variables made minor contributions.

The combined MaxEnt model was ground-truthed at 65 sites using SongMeters to acoustically record the presence of Koalas in the mating season and via quantitative sampling of browse tree size and availability. Records of Koala bellows (n=276 bellows) were analysed in an occupancy modelling framework, while a site habitat quality index was constructed based on browse tree basal area and diversity. Koala bellows were recorded on 29 % of ground-truth sites compared to Koala pellets that were recorded on 17 % of sites (13 of 2,600 trees searched). Field validation of the continuous model output demonstrated a linear increase in estimated Koala occupancy with higher model output values. Similarly, the site habitat quality index was correlated positively with the model output. However, the model output provided a better fit to estimated Koala occupancy than the site-based habitat quality index, probably because many variables were considered simultaneously by the model rather than just browse species. We suggest that this provides strong evidence for using the MaxEnt model to guide management decisions for Koalas in forested habitat.