Detection probability of invasive ship rats: biological causation and management implications

Abstract

Detection of cryptic animal species in the wild often relies on the use of passive detection devices, which require an animal to physically interact with the device in order to be detected. However, these devices imperfectly detect animals; some individuals may be present, but not detected. Estimating the proportion of individuals that are successfully detected using the sampling methodology, or the detection probability, therefore becomes essential where robust inference on population parameters across time and space is required. Despite this crucial importance, detection probability is usually considered to be a nuisance parameter; something which must be accounted for, but is not of interest in itself. In this thesis, I advance the argument that detection probability is not merely a nuisance parameter, but is itself a worthy topic of study, with biological and environmental casual mechanisms, and practical applications. In the context of invasive species management in particular, probability of detection is a ratelimiting parameter which essentially describes the effectiveness of management tools and techniques. This thesis provides a detailed analysis of detection probability of the invasive ship rat (Rattus rattus), one of the world’s worst invasive species. I used camera traps and electronic identity loggers to collect a large dataset of ship rat encounters and interactions with three types of detection device in northern New Zealand forest. In addition, home range areas and centres were estimated, using location data from VHF radio-collared rats. I used these data to estimate the component parameters of detection probability; namely the probability that a rat would encounter a device, and the probability that a rat would interact with a device upon encounter. I found that encounter probability could be described by a predictive model, with distance from home range centre as the primary predictor variable, and individual rat identity and device location as highly significant random effects moderating the response. Probability of interaction varied significantly among the different device types studied, and was also highly variable among individual rats and individual device locations. Finally, my estimates of encounter and interaction probabilities were used to inform a simulation study which predicted detection outcomes of a variety of management scenarios. The key finding of my simulations was that factors which influence the rate of encounter were more important determinants of population-level detection outcomes than factors influencing device interaction rates. Therefore, research and management effort aimed at increasing the likelihood that rats will encounter devices will generally be more impactful than effort to increase interaction probability.