Quantifying extinction probabilities from sighting records: inference and uncertainties

Inferring species extinction from sighting data

Understanding whether a species still persists, or the timing of its extinction is challenging, however, such knowledge is fundamental for effective species management.For the vast majority of species our understanding of their existence is based solely on sighting data that can range from museum specimens and clear photographs, through vocalisations, to markings and oral accounts.Here we review the methods that have been developed to infer the extinction of species from a sighting record, providing an understanding of their assumptions and applications. We have also produced an RShiny package which can be used to implement some of the methods presented in the article.While there are a number of potential areas that could be further developed, the methods reviewed provide a useful tool for inferring species extinction.

Analytical Bayesian approach for the design of surveillance and control programs to assess pest-eradication success

Large invasive species eradication programs are undertaken to protect native biodiversity and agriculture. Programs are typically followed by a series of surveys to assess the likelihood of eradication success and, when residual pests are detected, small-scale control or 'mop-ups' are implemented to eliminate these infestations. Further surveys follow to confirm absence with 'freedom' declared when a target probability of absence is reached. Such biosecurity programs comprise many interacting processes - stochastic biological processes including growth, and response and control interventions - and are an important component of post-border biosecurity. Statistical frameworks formulated to contribute to the design and efficiency of these surveillance and control programs are few and, those available, rely on the simulation of the component processes. In this paper we formulate an analytical Bayesian framework for a general biosecurity program with multiple components to assess pest-eradication success. Our model incorporates stochastic growth and detection processes, and several pest control mechanisms. Survey results and economic considerations are also taken into account to support a range of biosecurity management decisions. Using a case study we demonstrate that solutions match published simulation results and extend the available analysis. Principally, we show how analytical solutions can offer a powerful tool to support the design of effective and cost-efficient biosecurity systems, and we establish some general principles that guide and contribute to robust design.

Invasive species eradication: How do we declare success?

Deciding whether or not eradication of an invasive species has been successful is one of the main dilemmas facing managers of eradication programmes. When the species is no longer being detected, a decision must be made about when to stop the eradication programme and declare success. In practice, this decision is usually based on ad hoc rules, which may be inefficient. Since surveillance undertaken to confirm species absence is imperfect, any declaration of eradication success must consider the risk and the consequences of being wrong. If surveillance is insufficient, then eradication may be falsely declared (a Type I error), whereas continuation of surveillance when eradication has already occurred wastes resources (a Type II error). We review the various methods that have been developed for quantifying these errors and incorporating them into the decision-making process. We conclude with an overview of future developments likely to improve the practice of determining invasive species eradication success.

The effectiveness of citizen surveillance for detecting exotic vertebrates

Citizen observations of the natural world are increasing in detail, growing in volume and increasingly being shared on web-based platforms for the purpose of sharing information and/or the crowd-sourcing of species identification. From a biosecurity perspective, such citizen data streams are important as they are responsible for the majority of post-border reports and most detections of exotic pest species of concern. The sharing of sightings amongst what are effectively communities of practice is a key driver of having the sighting of an exotic pest species recognized and reported. Whilst it is clear that the eyes, ears, cameras, and microphones of citizens are a major component of biosecurity surveillance, it is unclear what level of surveillance this provides in the prospective sense. As an example, what confidence does citizen science provide about “proof of absence” for exotic pests of concern? The taxonomy of surveillance used within the field of biosecurity would classify such citizen activities as contributing to “general surveillance,” for which non-detections are typically not recorded and methods of quantitative analysis are still under development. We argue that while not recorded, there is considerable information about citizens activities that routinely underpins peoples mental inference about the level of surveillance provided by citizen activities. Furthermore, we show that it is possible to make such inference from general surveillance transparent by describing and characterizing the activities that potentially generate sightings in a way that is amenable to quantitative analysis. In the context of evaluating surveillance provided by citizens for incursions of exotic vertebrates, we provide examples of citizen observations providing early warning and hence preventing the establishment of species from a range of animal groups. Historically, analysis of the power of general surveillance has been restricted to being conceptual, based on qualitative arguments. We provide this, but also provide a quantitative model framework and provide examples of how different forms of general surveillance data may be analyzed, particularly in supporting inference of eradication/extinction.

Analytical Bayesian models to quantify pest eradication success or species absence using zero-sighting records

It is not possible to establish the absence of a population with certainty using imperfect zero-sighting records, but absence can be inferred. In this paper we use Bayesian methods to formulate analytical inferred distributions and statistics. When such formulations are available, they offer a highly efficient and powerful means of analysis. Our purpose is to provide accessible and versatile formulations to support an assessment of population absence for management decisions, using data from a series of regular and targeted surveys with zero-sightings. The stochastic processes considered here are prior population size, growth and imperfect detection, which are combined into a single distribution with sufficient flexibility to accommodate alternative distributions for each of the driving processes. Analytical solutions formulated include the inferred mean and variance for population size or number of infested survey-units, the probability of absence, the probability of a series of negative surveys conditional on presence, and the probability a population is first detected in a given survey, although we also formulate other statistics and provide explicit thresholds designed to support management decisions. Our formulation and results are straightforward to apply and provide insight into the nonlinear interactions and general characteristics of such systems. Although motivated by an assessment of population absence following a pest eradication program, results are also relevant to the status of threatened species, to 'proof-of-freedom' requirements for trade, and for inferring population size when a population is first detected.

Detecting rare carnivores using scats: Implications for monitoring a fox incursion into Tasmania

The ability to detect the incursion of an invasive species or destroy the last individuals during an eradication program are some of the most difficult aspects of invasive species management. The presence of foxes in Tasmania is a contentious issue with recent structured monitoring efforts, involving collection of carnivore scats and testing for fox DNA, failing to detect any evidence of foxes. Understanding the likelihood that monitoring efforts would detect fox presence, given at least one is present, is therefore critical for understanding the role of scat monitoring for informing the response to an incursion. We undertook trials to estimate the probability of fox scat detection through monitoring by scat‐detector dogs and person searches and used this information to critically evaluate the power of scat monitoring efforts for detecting foxes in the Tasmanian landscape. The probability of detecting a single scat present in a 1‐km² survey unit was highest for scat‐detector dogs searches (0.053) compared with person searches (x¯≅0.015) for each 10 km of search effort. Simulation of the power of recent scat monitoring efforts undertaken in Tasmania from 2011 to 2015 suggested that single foxes would have to be present in at least 20 different locations or fox breeding groups present in at least six different locations, in order to be detected with a high level of confidence (>0.80). We have shown that highly structured detection trials can provide managers with the quantitative tools needed to make judgments about the power of large‐scale scat monitoring programs. Results suggest that a fox population, if present in Tasmania, could remain undetected by a large‐scale, structured scat monitoring program. Therefore, it is likely that other forms of surveillance, in conjunction with scat monitoring, will be necessary to demonstrate that foxes are absent from Tasmania with high confidence.

Are extinction opinions extinct?

Extinction models vary in the information they require, the simplest considering the rate of certain sightings only. More complicated methods include uncertain sightings and allow for variation in the reliability of uncertain sightings. Generally extinction models require expert opinion, either as a prior belief that a species is extinct, or to establish the quality of a sighting record, or both. Is this subjectivity necessary? We present two models to explore whether the individual quality of sightings, judged by experts, is strongly informative of the probability of extinction: the ‘quality breakpoint method’ and the ‘quality as variance method’. For the first method we use the Barbary lion as an exemplar. For the second method we use the Barbary lion, Alaotra grebe, Jamaican petrel and Pohnpei starling as exemplars. The ‘quality breakpoint method’ uses certain and uncertain sighting records, and the quality of uncertain records, to establish whether a change point in the rate of sightings can be established using a simultaneous Bayesian optimisation with a non-informative prior. For the Barbary lion, there is a change in subjective quality of sightings around 1930. Unexpectedly sighting quality increases after this date. This suggests that including quality scores from experts can lead to irregular effects and may not offer reliable results. As an alternative, we use quality as a measure of variance around the sightings, not a change in quality. This leads to predictions with larger standard deviations, however the results remain consistent across any prior belief of extinction. Nonetheless, replacing actual quality scores with random quality scores showed little difference, inferring that the quality scores from experts are superfluous. Therefore, we deem the expensive process of obtaining pooled expert estimates as unnecessary, and even when used we recommend that sighting data should have minimal input from experts in terms of assessing the sighting quality at a fine scale. Rather, sightings should be classed as certain or uncertain, using a framework that is as independent of human bias as possible.

Declaration of local chemical eradication of the Argentine ant: Bayesian estimation with a multinomial-mixture model

Determining the success of eradication of an invasive species requires a way to decide when its risk of reoccurrence has become acceptably low. In Japan, the area populated by the Argentine ant, Linepithema humile (Mayr), is expanding, and eradication via chemical treatment is ongoing at various locations. One such program in Tokyo was apparently successful, because the ant population decreased to undetectable levels within a short time. However, construction of a population model for management purposes was difficult because the probability of detecting ants decreases rapidly as the population collapses. To predict the time when the ant was eradicated, we developed a multinomial-mixture model for chemical eradication based on monthly trapping data and the history of pesticide applications. We decided when to declare that eradication had been successful by considering both ‘eradication’ times, which we associated with eradication probabilities of 95% and 99%, and an optimal stopping time based on a ‘minimum expected economic cost’ that considered the possibility that surveys were stopped too soon. By applying these criteria, we retroactively declared that Argentine ants had been eradicated 38–42 months after the start of treatments (16–17 months after the last sighting).

Making inference from wildlife collision data: inferring predator absence from prey strikes

Wildlife collision data are ubiquitous, though challenging for making ecological inference due to typically irreducible uncertainty relating to the sampling process. We illustrate a new approach that is useful for generating inference from predator data arising from wildlife collisions. By simply conditioning on a second prey species sampled via the same collision process, and by using a biologically realistic numerical response functions, we can produce a coherent numerical response relationship between predator and prey. This relationship can then be used to make inference on the population size of the predator species, including the probability of extinction. The statistical conditioning enables us to account for unmeasured variation in factors influencing the runway strike incidence for individual airports and to enable valid comparisons. A practical application of the approach for testing hypotheses about the distribution and abundance of a predator species is illustrated using the hypothesized red fox incursion into Tasmania, Australia. We estimate that conditional on the numerical response between fox and lagomorph runway strikes on mainland Australia, the predictive probability of observing no runway strikes of foxes in Tasmania after observing 15 lagomorph strikes is 0.001. We conclude there is enough evidence to safely reject the null hypothesis that there is a widespread red fox population in Tasmania at a population density consistent with prey availability. The method is novel and has potential wider application.

Understanding the biological invasion risk posed by the global wildlife trade: propagule pressure drives the introduction and establishment of Nearctic turtles

Biological invasions are a key component of human-induced global change. The continuing increase in global wildlife trade has raised concerns about the parallel increase in the number of new invasive species. However, the factors that link the wildlife trade to the biological invasion process are still poorly understood. Moreover, there are analytical challenges in researching the role of global wildlife trade in biological invasions, particularly issues related to the under-reporting of introduced and established populations in areas with reduced sampling effort. In this work, we use high-quality data on the international trade in Nearctic turtles (1999-2009) coupled with a statistical modelling framework, which explicitly accounts for detection, to investigate the factors that influence the introduction (release, or escape into the wild) of globally traded Nearctic turtles and the establishment success (self-sustaining exotic populations) of slider turtles (Trachemys scripta), the most frequently traded turtle species. We found that the introduction of a species was influenced by the total number of turtles exported to a jurisdiction and the age at maturity of the species, while the establishment success of slider turtles was best associated with the propagule number (number of release events), and the number of native turtles in the jurisdiction of introduction. These results indicate both a direct and indirect association between the wildlife trade and the introduction of turtles and establishment success of slider turtles, respectively. Our results highlight the existence of gaps in the number of globally recorded introduction events and established populations of slider turtles, although the expected bias is low. We emphasize the importance of researching independently the factors that affect the different stages of the invasion pathway. Critically, we observe that the number of traded individuals might not always be an adequate proxy for propagule pressure and establishment success.

Inferring the distribution and demography of an invasive species from sighting data: the red fox incursion into Tasmania

A recent study has inferred that the red fox (Vulpes vulpes) is now widespread in Tasmania as of 2010, based on the extraction of fox DNA from predator scats. Heuristically, this inference appears at first glance to be at odds with the lack of recent confirmed discoveries of either road-killed foxes--the last of which occurred in 2006, or hunter killed foxes--the most recent in 2001. This paper demonstrates a method to codify this heuristic analysis and produce inferences consistent with assumptions and data. It does this by formalising the analysis in a transparent and repeatable manner to make inference on the past, present and future distribution of an invasive species. It utilizes Approximate Bayesian Computation to make inferences. Importantly, the method is able to inform management of invasive species within realistic time frames, and can be applied widely. We illustrate the technique using the Tasmanian fox data. Based on the pattern of carcass discoveries of foxes in Tasmania, we infer that the population of foxes in Tasmania is most likely extinct, or restricted in distribution and demographically weak as of 2013. It is possible, though unlikely, that that population is widespread and/or demographically robust. This inference is largely at odds with the inference from the predator scat survey data. Our results suggest the chances of successfully eradicating the introduced red fox population in Tasmania may be significantly higher than previously thought.

Degradation and detection of fox (Vulpes vulpes) scats in Tasmania: evidence from field trials

Context The Tasmanian Government is attempting to eradicate foxes from Tasmania and carnivore-scat surveys using humans and dogs combined with DNA testing are the main methods of detection. Understanding the rate that scats degrade is a key component for estimating the power of monitoring for detecting cryptic predators and will contribute to a broader understanding of the use of scat monitoring for informing eradication programs. Aims To estimate the degradation rate of fox scats and derive an estimate of the abundance of scats available to observers monitoring for fox presence. Methods In total, 486 fresh fox scats were placed at nine sites within three bioregions in Tasmania and left to degrade for up to 126 days. Scats were observed periodically by both humans and dogs to determine when they became unrecognisable and/or undetectable. Key results Recognition of scats by humans declined faster in summer than in winter and did not vary systematically among bioregions. Median survival times of scats were 19 days in summer and 26 days in winter. Recognition of scats by dogs was higher in summer than in winter, with dogs recognising scats past the time they became unrecognisable to humans. Using estimates of scat degradation derived from human observers, the equilibrium abundance of detectable scats within a fox home range was estimated to be 179–243 scats. However, the abundance of detectable scats on linear features subject to monitoring was estimated to be 10–15 scats. Conclusions Using our estimate of the abundance of scats on linear features, the current distribution of fox scats detected in Tasmania may not be as anomalous as has been suggested by others. However, fox detection from scats will be highly dependent on deposition patterns and distribution of scats on linear features and this should be critically reassessed in Tasmania. Implications Fox scats are not expected to exhibit systematic regional differences in degradation rates that might have an impact on monitoring strategies. Estimates of the abundance of scats detectable by observers are critical for assessing the effectiveness of scat-monitoring programs. We advocate that a rigorous assessment of future scat-monitoring programs in Tasmania be undertaken to determine their power to detect foxes.