Abstract
Detecting animals in the wild is fundamental to understanding the threats they may face, and managing those threats in situ. Detections are required to estimate the size and trajectories of populations, metrics that are often the basis for management decisions. Though simple in concept, it can be difficult to select or develop methods that can consistently detect a species throughout its range, that are economically viable, and can be deployed in remote locations. The challenges are greater for cryptic species, that may be rarely available to sampling, or that are present in habitats where access is difficult. The lizards of Aotearoa New Zealand (NZ) meet both criteria and are also highly threatened, so are often present at low abundances. This combination of rarity and crypsis can make collecting information on the distribution and trends of most species inefficient. In a conservation economy that does not have sufficient resources to manage or monitor all, or even most, threatened species, inefficient sampling is tenable for only a small proportion. Given that resourcing is unlikely to improve, a necessary alternative is to increase the efficiency of sampling, so that limited resources are stretched further. At minimum, it is important to understand the limitations of different sampling approaches, so that data can be interpreted correctly.
This thesis was conceived to evaluate some of the sampling methods used to detect lizards in NZ. First, in Chapter 2, I tested the efficiency of commonly used sampling devices, pitfall and funnel traps. These two devices are used frequently when trapping for terrestrial and fossorial lizards. I compare captures in these devices with the detections made by paired trail cameras to generate a trapping efficiency value. On average, it took 20 trap encounters to result in a single capture. At low population densities, poor trapping rates may preclude detection and limit the precision of estimates that require recaptures. A reflection of this problem, and the focus of Chapter 3, is a monitoring effort for a Lakes skink population, which after eight years has failed to generate sufficiently precise estimates of abundance to understand this population’s trend. I report that a simple change to monitoring, extending the effort period, provided a more precise estimate. However, this approach is not guaranteed to succeed in future or with other populations, particularly as pitfall traps had a declining probability of catching skinks within annual effort. I also validated photo-ID mark-recapture for this population and suggest that survival rates could serve as an alternative to abundance estimates when justifying management interventions. In Chapter 4, I describe a negative impact of trapping when using funnel traps to catch a large-bodied species of skink. Twenty-five percent of captured animals had damage to their snouts, presumably from ramming into the trap walls, a behaviour that may have been exacerbated by high in-trap temperatures. In traps with paired cameras, I also noted several occasions when mice entered and exited funnel traps. Negative impacts, such as the reported snout damage or the risk of in-trap predation may drive learnt trap shyness, contributing to difficulty recapturing animals.
Given the noted limitations of trapping, any strategies to boost efficiency would be of value. In Chapters 5 and 6, I tested two such methods, which are sampling guided by species distribution modelling and by expert opinion. I built correlative MaxEnt models for six lizard species and validated each using withheld data, and data collected independently of model building. Cross-validation indicated that the top models for each species generated reasonably accurate predictions; however, for common species, predictive accuracy decreased notably when validating with independent data, indicating that predictions poorly reflect contemporary distributions of the lizards involved. I suggest that poor quality occurrence data, a dearth of understanding regarding species ecology, and the low resolution of predictor data likely restricts the relevance of predictive modelling for NZ lizards. In Chapter 6, I tested the value of using expert opinion to select areas for sampling, a method which is widespread if not commonly reported. Directing effort in this way may serve to reduce time spent sampling in unproductive habitats, but is inherently biased and often not reproducible. I interviewed lizard experts and asked them to predict trap and site occupancy based on annotated trap photos. The accuracy of site occupancy predictions scaled with expertise level and was better than random, though was driven largely by correct predictions of absence. Trap occupancy was predicted poorly, with a large number of false positive responses. Taken together, results indicate that lizard practitioners choosing which sites to sample may improve efficiency, but that choosing where to sample at the scale of a trap is less valuable. It is likely better to instead apply systematic or consistent sampling designs that can be more easily compared among sites and species.
The apparent inefficiency of trapping for NZ lizards, and the lack of strategies to markedly improve detection rates is indicative that future research should focus on developing novel methods of detections, that, unlike current methods, are specifically designed for lizards. I show that camera traps, which detected lizards far more often than paired pitfall or funnel traps, could serve as such an alternative. Aside from novel approaches, increased quantification of detectability for different sampling methods is required to allow practitioners to plan effort that can meet objectives. Investment in developing best practice guidelines for sampling and a data repository that is fit for purpose is also required to maximise the utility of existing data and current sampling regimes.