Abstract
Introduced mammalian predators threaten native biodiversity and primary industries
throughout New Zealand. One introduced mammalian predator, the Brushtail possum
(Trichosurus vulpecula), threatens native species by creating competition for resources,
changing forest composition, and preying on native bird species. To eradicate possums from
New Zealand it is essential to further develop control methods. Predator Free 2050 is an
initiative to eradicate seven invasive mammalian pests from New Zealand, including the
Brushtail possum (Trichosurus vulpecula). Trapping is one key method for possum control,
however creates high labor costs and maintenance of the trap network.
Two research questions addressed in this research are (1) which environmental drivers are
associated with possum trap success and (2) can spatial and temporal patterns of trap success
help to inform predictive models to increase the effectiveness of possum trapping regimes. The
study takes place in West Harbour, Dunedin, Otepoti, within the control area of the Halo
project, a delivery partner of Predator Free 2050. Spatial and temporal patterns of possum trap
success from 2016 – 2020 in West Harbour are assessed to inform a predictive model of trap
success across West Harbour. Leveraging high-spatial resolution satellite imagery and existing
fine-scale vegetation mapping, a series of environmental predictor variables are identified in
relation to trap success.
Average trap success was highest in months from February – May. Overall
podocarp/broadleaved forest was associated with the highest average trap success, whilst
Manuka dominant forest and shrub land was associated with the lowest. Gorse/broom trapping
success was highest in spring, broadleaved forest in summer, and exotic coniferous forest and
podocarp/broadleaved forest in autumn. The most successful landcover type also had the
highest summer normalized difference vegetation index (NDVI) values, indicating that traps
set in landcovers with a higher degree of living biomass have higher trap success. Statistically
significant predictor variables in the models include percentage broadleaved/podocarp, density
of traps at time of trapping, summer NDVI and landcover heterogeneity. Using stepwise
regression, a parsimonious model of trap success was fitted. Model validation showed that this
model was able to predict trap success (0-1) with high accuracy with an average error of 0.075
- 0.080. From the results of this research it is clear that summer NDVI, percentage broadleaved
podocarp and landcover heterogeneity should be explored further as promising predictors
variables when seeking to predict trap success in possum control regimes.
It is recommended that to ensure a trap regime yields increased success, statistically significant
predictor variables be incorporated in to future predictive models. It is also recommended that
a low-density network of traps is maintained for the West Harbour study area, with placement
of traps in areas identified with the highest predicted trap success.