Inferring extinctions from sighting records of variable reliability

An integrated system to assess marine extinctions

More than 20 global marine extinctions and over 700 local extinctions have reportedly occurred during the past 500 years. However, available methods to determine how many of these species can be confidently declared true disappearances tend to be data-demanding, time-consuming, and not applicable to all taxonomic groups or scales of marine extinctions (global [G] and local [L]). We developed an integrated system to assess marine extinctions (ISAME) that can be applied to any taxonomic group at any geographic scale. We applied the ISAME method to 10 case studies to illustrate the possible ways in which the extinction status of marine species can be categorized as unverified, possibly extinct, or extinct. Of the 10 case studies we assessed, the ISAME method concludes that 6 should be categorized as unverified extinctions due to problems with species' identity and lack of reliable evidence supporting their disappearance (periwinkle-Littoraria flammea [G], houting-Coregonus oxyrinchus [G], long-spined urchin-Diadema antillarum [L], smalltooth sawfish-Pristis pectinata [L], and largetooth sawfish-P. pristis [L]). In contrast, ISAME classified the Guadalupe storm-petrel (Oceanodroma macrodactyla [G]) and the lost shark (Carcharhinus obsolerus [G]) as possibly extinct because the available evidence indicates that their extinction is plausible-while the largetooth sawfish [L] and Steller's sea cow (Hydrodamalis gigas [G]) were confirmed to be extinct. Determining whether a marine population or species is actually extinct or still extant is needed to guide conservation efforts and prevent further biodiversity losses.

Quantifying the relative performance of two undetected-extinction models

Extinctions of undiscovered species (undetected extinctions) constitute a portion of biodiversity loss that is often ignored. We compared the performance of 2 models of undetected extinctions - Tedesco and SEUX - when estimating undetected extinctions with both simulated and real-world data. We generated simulated data by considering a birth-death process in which less abundant species were more likely to go extinct. When detection rates were higher for common species, the 2 models underestimated the true number of undetected extinctions by up to 88.7%, and when detection rates were independent of abundance, the 2 models performed better; the SEUX model had an average bias of +3.1% and the Tedesco model had an average bias of -62.3%. We applied the models to 8 real-world data sets (e.g., Australian amphibians, Australian birds, North American bivalves) and found that true extinctions may be from 15% to 180% higher than observed values. For 6 of the 8 data sets, the SEUX model yielded absolute estimates that were 5.7-66.8% lower than those of the Tedesco model. We mainly attributed this difference to the SEUX model's assumption that there are no undetected extant species currently. We assessed the accuracy of the models' estimates with a logistic regression to test whether detection and extinction rates were uncorrelated across species. Rates were correlated for 3 of the 8 data sets; species discovered later had a higher probability of being extinct, suggesting that extinction numbers could be even higher for these groups. Despite caveats associated with the models, the evidence from both show biodiversity loss in these groups may be more severe than what has been documented.

Identifying species threatened with local extinction in tropical reef fisheries using historical reconstruction of species occurrence

Identifying the species that are at risk of local extinction in highly diverse ecosystems is a big challenge for conservation science. Assessments of species status are costly and difficult to implement in developing countries with diverse ecosystems due to a lack of species-specific surveys, species-specific data, and other resources. Numerous techniques are devised to determine the threat status of species based on the availability of data and budgetary limits. On this basis, we developed a framework that compared occurrence data of historically exploited reef species in Kenya from existing disparate data sources. Occurrence data from archaeological remains (750-1500CE) was compared with occurrence data of these species catch assessments, and underwater surveys (1991-2014CE). This comparison indicated that only 67 species were exploited over a 750 year period, 750-1500CE, whereas 185 species were landed between 1995 and 2014CE. The first step of our framework identified 23 reef species as threatened with local extinction. The second step of the framework further evaluated the possibility of local extinction with Bayesian extinction analyses using occurrence data from naturalists’ species list with the existing occurrence data sources. The Bayesian extinction analysis reduced the number of reef species threatened with local extinction from 23 to 15. We compared our findings with three methods used for assessing extinction risk. Commonly used extinction risk methods varied in their ability to identify reef species that we identified as threatened with local extinction by our comparative and Bayesian method. For example, 12 of the 15 threatened species that we identified using our framework were listed as either least concern, unevaluated, or data deficient in the International Union for the Conservation of Nature red list. Piscivores and macro-invertivores were the only functional groups found to be locally extinct. Comparing occurrence data from disparate sources revealed a large number of historically exploited reef species that are possibly locally extinct. Our framework addressed biases such as uncertainty in priors, sightings and survey effort, when estimating the probability of local extinction. Our inexpensive method showed the value and potential for disparate data to fill knowledge gaps that exist in species extinction assessments.

Inferring about the extinction of a species using certain and uncertain sightings

The sighting record of threatened species is often used to infer the possibility of extinction. Most of these sightings have uncertain validity. Solow and Beet(2014) developed two models using a Bayesian approach which allowed for uncertainty in the sighting record by formally incorporating both certain and uncertain sightings, but in different ways. Interestingly, the two methods give completely different conclusions concerning the extinction of the Ivory-billed Woodpecker. We further examined these two methods to provide a mathematical explanation, and to explore in more depth, as to why the results differed from one another. It was found that the first model was more sensitive to the last uncertain sighting, while the second was more sensitive to the last certain sighting. The difficulties in choosing the appropriate model are discussed.

Estimating the extinction date of the thylacine with mixed certainty data

The thylacine (Thylacinus cynocephalus), one of Australia's most characteristic megafauna, was the largest marsupial carnivore until hunting, and potentially disease, drove it to extinction in 1936. Although thylacines were restricted to Tasmania for 2 millennia prior to their extinction, recent so-called plausible sightings on the Cape York Peninsula in northern Queensland have emerged, leading some to speculate the species may have persisted undetected. We compiled a data set that included physical evidence, expert-validated sightings, and unconfirmed sightings up to the present day and implemented a range of extinction models (focusing on a Bayesian approach that incorporates all 3 types of data by modeling valid and invalid sightings as independent processes) to evaluate the likelihood of the thylacine's persistence. Although the last captive individual died in September 1936, our results suggested that the most likely extinction date would be 1940. Our other extinction models estimated the thylacine's extinction date between 1936 and 1943, and the most optimistic scenario indicated that the species did not persist beyond 1956. The search for the thylacine, much like similar efforts to rediscover other recently extinct charismatic taxa, is likely to be fruitless, especially given that persistence on Tasmania would have been no guarantee the species could reappear in regions that had been unoccupied for millennia. The search for the thylacine may become a rallying point for conservation and wildlife biology and could indirectly help fund and support critical research in understudied areas such as Cape York. However, our results suggest that attempts to rediscover the thylacine will be unsuccessful and that the continued survival of the thylacine is entirely implausible based on most current mathematical theories of extinction.

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.

Extinct or still out there? Disentangling influences on extinction and rediscovery helps to clarify the fate of species on the edge

Each year, two or three species that had been considered to be extinct are rediscovered. Uncertainty about whether or not a species is extinct is common, because rare and highly threatened species are difficult to detect. Biological traits such as body size and range size are expected to be associated with extinction. However, these traits, together with the intensity of search effort, might influence the probability of detection and extinction differently. This makes statistical analysis of extinction and rediscovery challenging. Here, we use a variant of survival analysis known as cure rate modelling to differentiate factors that influence rediscovery from those that influence extinction. We analyse a global data set of 99 mammals that have been categorized as extinct or possibly extinct. We estimate the probability that each of these mammals is still extant and thus estimate the proportion of missing (presumed extinct) mammals that are incorrectly assigned extinction. We find that body mass and population density are predictors of extinction, and body mass and search effort predict rediscovery. In mammals, extinction rate increases with body mass and population density, and these traits act synergistically to greatly elevate extinction rate in large species that also occurred in formerly dense populations. However, when they remain extant, larger-bodied missing species are rediscovered sooner than smaller species. Greater search effort increases the probability of rediscovery in larger species of missing mammals, but has a minimal effect on small species, which take longer to be rediscovered, if extant. By separating the effects of species characteristics on extinction and detection, and using models with the assumption that a proportion of missing species will never be rediscovered, our new approach provides estimates of extinction probability in species with few observation records and scant ecological information.

Inferring extinction in North American and Hawaiian birds in the presence of sighting uncertainty

For most species the timing of extinction events is uncertain, occurring sometime after the last sighting. However, the sightings themselves may also be uncertain. Recently a number of methods have been developed that incorporate sighting uncertainty in the inference of extinction based on a series of sightings. Here we estimate the timing of extinction for 41 of 52 North American and Hawaiian bird taxa and populations, the results of which suggest all became extinct before 2009. By acknowledging sighting uncertainty it results in two opposite effects, one pushing the timing of extinction away from the last sighting and the other drawing the timing of extinction nearer to it. However, for 14 assessed taxa and populations the upper 95% bounds lie beyond the end of the observation period and therefore suggest the possibility of continued persistence. This has important implications for conservation decision-makers and potentially reduces the likelihood of Romeo’s Error.

On the prior distribution of extinction time

Bayesian inference about the extinction of a species based on a record of its sightings requires the specification of a prior distribution for extinction time. Here, I critically review some specifications in the context of a specific model of the sighting record. The practical implication of the choice of prior distribution is illustrated through an application to the sighting record of the Caribbean monk seal.

Eradication of Invading Insect Populations: From Concepts to Applications

Eradication is the deliberate elimination of a species from an area. Given that international quarantine measures can never be 100% effective, surveillance for newly arrived populations of nonnative species coupled with their eradication represents an important strategy for excluding potentially damaging insect species. Historically, eradication efforts have not always been successful and have sometimes been met with public opposition. But new developments in our understanding of the dynamics of low-density populations, the availability of highly effective treatment tactics, and bioeconomic analyses of eradication strategies offer new opportunities for developing more effective surveillance and eradication programs. A key component that connects these new developments is the harnessing of Allee effects, which naturally promote localized species extinction. Here we review these developments and suggest how research might enhance eradication strategies.

Assessing uncertainty in sighting records: an example of the Barbary lion

As species become rare and approach extinction, purported sightings can be controversial, especially when scarce management resources are at stake. We consider the probability that each individual sighting of a series is valid. Obtaining these probabilities requires a strict framework to ensure that they are as accurately representative as possible. We used a process, which has proven to provide accurate estimates from a group of experts, to obtain probabilities for the validation of 32 sightings of the Barbary lion. We consider the scenario where experts are simply asked whether a sighting was valid, as well as asking them to score the sighting based on distinguishablity, observer competence, and verifiability. We find that asking experts to provide scores for these three aspects resulted in each sighting being considered more individually, meaning that this new questioning method provides very different estimated probabilities that a sighting is valid, which greatly affects the outcome from an extinction model. We consider linear opinion pooling and logarithm opinion pooling to combine the three scores, and also to combine opinions on each sighting. We find the two methods produce similar outcomes, allowing the user to focus on chosen features of each method, such as satisfying the marginalisation property or being externally Bayesian.

On uncertain sightings and inference about extinction

The extinction of many species can only be inferred from the record of sightings of individuals. Solow et al. (2012, Uncertain sightings and the extinction of the Ivory-billed Woodpecker. Conservation Biology 26:180-184) describe a Bayesian approach to such inference and apply it to a sighting record of the Ivory-billed Woodpecker (Campephilus principalis). A feature of this sighting record is that all uncertain sightings occurred after the most recent certain sighting. However, this appears to be an artifact. We extended this earlier work in 2 ways. First, we allowed for overlap in time between certain and uncertain sightings. Second, we considered 2 plausible statistical models of a sighting record. In one of these models, certain and uncertain sightings that are valid arise from the same process whereas in the other they arise from independent processes. We applied both models to the case of the Ivory-billed Woodpecker. The result from the first model did not favor extinction, whereas the result for the second model did. This underscores the importance, in applying tests for extinction, of understanding what could be called the natural history of the sighting record.

Quantifying extinction probabilities from sighting records: inference and uncertainties

Methods are needed to estimate the probability that a population is extinct, whether to underpin decisions regarding the continuation of a invasive species eradication program, or to decide whether further searches for a rare and endangered species could be warranted. Current models for inferring extinction probability based on sighting data typically assume a constant or declining sighting rate. We develop methods to analyse these models in a Bayesian framework to estimate detection and survival probabilities of a population conditional on sighting data. We note, however, that the assumption of a constant or declining sighting rate may be hard to justify, especially for incursions of invasive species with potentially positive population growth rates. We therefore explored introducing additional process complexity via density-dependent survival and detection probabilities, with population density no longer constrained to be constant or decreasing. These models were applied to sparse carcass discoveries associated with the recent incursion of the European red fox (Vulpes vulpes) into Tasmania, Australia. While a simple model provided apparently precise estimates of parameters and extinction probability, estimates arising from the more complex model were much more uncertain, with the sparse data unable to clearly resolve the underlying population processes. The outcome of this analysis was a much higher possibility of population persistence. We conclude that if it is safe to assume detection and survival parameters are constant, then existing models can be readily applied to sighting data to estimate extinction probability. If not, methods reliant on these simple assumptions are likely overstating their accuracy, and their use to underpin decision-making potentially fraught. Instead, researchers will need to more carefully specify priors about possible population processes.