Optimal monitoring and statistical modelling methods for feral cats and other mammalian predators in a pastoral landscape

Introduction. Mammalian predators have a global impact on biodiversity. Mammalian predator species often occur at low abundance and require efficient and non-invasive monitoring techniques, alongside reliable statistical modelling methods, to explain the probability of detection, presence and, when possible, abundance. Camera traps are a dynamic technology currently used around the world to monitor a wide range of species in a variety of ecological research programmes However, as camera traps continue to change and improve, there is a need for more standardisation for both camera settings and deployment methods, depending on the objective of the study. Aims. The aims of this study were to: 1) determine the optimal orientation for camera traps to detect mammalian pest species; 2) determine the optimal statistical method for modelling changes in a feral cat population pre-and post-predator control operation; 3) assess the effectiveness of a Bayesian abundance estimator at providing abundance estimates for a population of hedgehogs pre- and post-predator control; and 4) deploy camera traps on a wide scale to determine the baseline relative abundance and detection rate for feral cats prior to a predator control operation. Materials and methods. 1) I deployed 20 pairs of camera traps (one horizontal and one vertical) for 21 days across a pastoral landscape in Hawke’s Bay, North Island, New Zealand, and compared numbers of detections for target species (feral cats) and mustelids (M. furo, M. erminea, and M. nivalis); 2) I deployed 40 horizontally-oriented cameras on pre-determined grid sites across two pastoral properties in Hawke’s Bay, for two consecutive periods of 21 days, to monitor feral cats pre- and post-predator control. I compared four statistical modelling methods for gauging the success of the control operation: index manipulation method (IMI), capture-mark-recapture (CMR), a generalised linear mixed model method (GLMM), and the spatial presence-absence model (SPA). The IMI method was used as a benchmark method for comparison as used in previous studies; 3) I used the same camera trap data from the previous/an earlier?? chapter to estimate hedgehog abundance pre- and post-control, using the SPA model; and 4) I deployed 68 horizontally-oriented cameras on a wide scale (26,000 ha) across two pastoral areas of coastal Hawke’s Bay (The Cape to City restoration project) for 21 days, to monitor feral cats prior to a predator control operation. I compared the GLMM method from the previous study with an abundance-induced heterogeneity model (RN) for estimating the proportion of cameras detecting cats per site and the relative abundance at each site. I also used the RN model to compare the effect of habitat type (forest, forest margin, mixed, and open) on the abundance and proportion of detections. Results. 1) Horizontal cameras produced a significantly greater number of photos overall (P < 0.001) and more independent encounters with the target species (P = 0.03). Orientation did not influence the number of false triggers (P = 0.53); 2) The IMI and SPA models gave similar, accurate estimates showing a decrease in cat abundance (90% and 88%, respectively) post-predator removal. The GLMM method showed a significant decrease in camera detection rates post-control (90%). The CMR models were unable to give accurate abundance estimates due to the low sample size of reliably identifiable cats; 3) The SPA model produced more precise estimates for a population of hedgehogs (due to a higher number of multiple detections than feral cats) and successfully showed a reduction in abundance post-predator control; 4) Strategically placed cameras had much higher detection rates than previous studies with the GLMM method estimating 5.2% (95 % C.I 2.3-7.9) for Site 1, and 4.3% (2.6-10.3) for Site 2. The RN model estimated detection rates of 5.5% (95% CI 4.1-6.9) for Site 1, and 4.5% (3.1-5.9) for Site 2. The RN model also indicated variation in the relative abundance based on habitat type with significantly higher detection rates in forests and along forest margins compared with mixed scrub and open farmland. Discussion and conclusions. Horizontally-oriented cameras performed well at detecting feral cats and mustelids. While the GLMM method and the SPA model gave accurate results in comparison to the IMI method, they lacked precision. CMR models have had success with large, well marked felids;, however, they do not perform well with a small sample size of identifiable cat images (very few clearly marked individuals). Although requiring two separate measures (pre-monitoring, manipulation, then post-monitoring), the IMI method was simple to calculate using a variance equation and appeared to be accurate. This may be a valuable method for implementation by wildlife managers in the future. The SPA model performed well (more precision) for hedgehogs, for which there were more multiple detections per camera station. This model requires large amounts of data and is more appropriately suited for a species that occurs at higher densities than feral cats. However, a further examination of the size of hedgehog home ranges and the possibility that they simply saturated the detection network with high numbers, must be done to ensure that the model’s requirement for spatially-correlated detection units has been satisfied. Both the GLMM and RN model showed no substantial differences in cat detections for either site prior to a predator control operation. The RN model was able to incorporate heterogeneity at the individual camera station level; thus, it provided more precise estimates at the overall site level