A Bayesian state-space formulation of dynamic occupancy models

Species occurrence and its dynamic components, extinction and colonization probabilities, are focal quantities in biogeography and metapopulation biology, and for species conservation assessments. It has been increasingly appreciated that these parameters must be estimated separately from detection probability to avoid the biases induced by non-detection error. Hence, there is now considerable theoretical and practical interest in dynamic occupancy models that contain explicit representations of metapopulation dynamics such as extinction, colonization, and turnover as well as growth rates. We describe a hierarchical parameterization of these models that is analogous to the state-space formulation of models in time series, where the model is represented by two components, one for the partially observable occupancy process and another for the observations conditional on that process. This parameterization naturally allows estimation of all parameters of the conventional approach to occupancy models, but in addition, yields great flexibility and extensibility, e.g., to modeling heterogeneity or latent structure in model parameters. We also highlight the important distinction between population and finite sample inference; the latter yields much more precise estimates for the particular sample at hand. Finite sample estimates can easily be obtained using the state-space representation of the model but are difficult to obtain under the conventional approach of likelihood-based estimation. We use R and WinBUGS to apply the model to two examples. In a standard analysis for the European Crossbill in a large Swiss monitoring program, we fit a model with year-specific parameters. Estimates of the dynamic parameters varied greatly among years, highlighting the irruptive population dynamics of that species. In the second example, we analyze route occupancy of Cerulean Warblers in the North American Breeding Bird Survey (BBS) using a model allowing for site-specific heterogeneity in model parameters. The results indicate relatively low turnover and a stable distribution of Cerulean Warblers which is in contrast to analyses of counts of individuals from the same survey that indicate important declines. This discrepancy illustrates the inertia in occupancy relative to actual abundance. Furthermore, the model reveals a declining patch survival probability, and increasing turnover, toward the edge of the range of the species, which is consistent with metapopulation perspectives on the genesis of range edges. Given detection/non-detection data, dynamic occupancy models as described here have considerable potential for the study of distributions and range dynamics.     

Chilean Blue Whales as a Case Study to Illustrate Methods to Estimate Abundance and Evaluate Conservation Status of Rare Species

Abstract: Often abundance of rare species cannot be estimated with conventional design‐based methods, so we illustrate with a population of blue whales (Balaenoptera musculus) a spatial model‐based method to estimate abundance. We analyzed data from line‐transect surveys of blue whales off the coast of Chile, where the population was hunted to low levels. Field protocols allowed deviation from planned track lines to collect identification photographs and tissue samples for genetic analyses, which resulted in an ad hoc sampling design with increased effort in areas of higher densities. Thus, we used spatial modeling methods to estimate abundance. Spatial models are increasingly being used to analyze data from surveys of marine, aquatic, and terrestrial species, but estimation of uncertainty from such models is often problematic. We developed a new, broadly applicable variance estimator that showed there were likely 303 whales (95% CI 176–625) in the study area. The survey did not span the whales' entire range, so this is a minimum estimate. We estimated current minimum abundance relative to pre‐exploitation abundance (i.e., status) with a population dynamics model that incorporated our minimum abundance estimate, likely population growth rates from a meta‐analysis of rates of increase in large baleen whales, and two alternative assumptions about historic catches. From this model, we estimated that the population was at a minimum of 9.5% (95% CI 4.9–18.0%) of pre‐exploitation levels in 1998 under one catch assumption and 7.2% (CI 3.7–13.7%) of pre‐exploitation levels under the other. Thus, although Chilean blue whales are probably still at a small fraction of pre‐exploitation abundance, even these minimum abundance estimates demonstrate that their status is better than that of Antarctic blue whales, which are still <1% of pre‐exploitation population size. We anticipate our methods will be broadly applicable in aquatic and terrestrial surveys for rarely encountered species, especially when the surveys are intended to maximize encounter rates and estimate abundance.     

Simultaneous-count models to estimate abundance from counts of unmarked individuals with imperfect detection

We developed a method to estimate population abundance from simultaneous counts of unmarked individuals over multiple sites. We considered that at each sampling occasion, individuals in a population could be detected at 1 of the survey sites or remain undetected and used either multinomial or binomial simultaneous-count models to estimate abundance, the latter being equivalent to an N-mixture model with one site. We tested model performance with simulations over a range of detection probabilities, population sizes, growth rates, number of years, sampling occasions, and sites. We then applied our method to 3 critically endangered vulture species in Cambodia to demonstrate the real-world applicability of the model and to provide the first abundance estimates for these species in Cambodia. Our new approach works best when existing methods are expected to perform poorly (i.e., few sites and large variation in abundance among sites) and if individuals may move among sites between sampling occasions. The approach performed better when there were >8 sampling occasions and net probability of detection was high (>0.5). We believe our approach will be useful in particular for simultaneous surveys at aggregation sites, such as roosts. The method complements existing approaches for estimating abundance of unmarked individuals and is the first method designed specifically for simultaneous counts.     

Loss of Large Predatory Sharks from the Mediterranean Sea

Abstract:  Evidence for severe declines in large predatory fishes is increasing around the world. Because of its long history of intense fishing, the Mediterranean Sea offers a unique perspective on fish population declines over historical timescales. We used a diverse set of records dating back to the early 19th and mid 20th century to reconstruct long-term population trends of large predatory sharks in the northwestern Mediterranean Sea. We compiled 9 time series of abundance indices from commercial and recreational fishery landings, scientific surveys, and sighting records. Generalized linear models were used to extract instantaneous rates of change from each data set, and a meta-analysis was conducted to compare population trends. Only 5 of the 20 species we considered had sufficient records for analysis. Hammerhead ( Sphyrna spp.), blue ( Prionace glauca ), mackerel ( Isurus oxyrinchus and Lamna nasus ), and thresher sharks ( Alopias vulpinus ) declined between 96 and 99.99% relative to their former abundance. According to World Conservation Union (IUCN) criteria, these species would be considered critically endangered. So far, the lack of quantitative population assessments has impeded shark conservation in the Mediterranean Sea. Our study fills this critical information gap, suggesting that current levels of exploitation put large sharks at risk of extinction in the Mediterranean Sea. Possible ecosystem effects of these losses involve a disruption of top-down control and a release of midlevel consumers.     

Linking Indices for Biodiversity Monitoring to Extinction Risk Theory

Biodiversity indices often combine data from different species when used in monitoring programs. Heuristic properties can suggest preferred indices, but we lack objective ways to discriminate between indices with similar heuristics. Biodiversity indices can be evaluated by determining how well they reflect management objectives that a monitoring program aims to support. For example, the Convention on Biological Diversity requires reporting about extinction rates, so simple indices that reflect extinction risk would be valuable. We developed 3 biodiversity indices that are based on simple models of population viability that relate extinction risk to abundance. We based the first index on the geometric mean abundance of species and the second on a more general power mean. In a third index, we integrated the geometric mean abundance and trend. These indices require the same data as previous indices, but they also relate directly to extinction risk. Field data for butterflies and woodland plants and experimental studies of protozoan communities show that the indices correlate with local extinction rates. Applying the index based on the geometric mean to global data on changes in avian abundance suggested that the average extinction probability of birds has increased approximately 1% from 1970 to 2009. Conectando Índices para el Monitoreo de la Biodiversidad con la Teoría de Riesgo de Extinción     

Accuracy of Short‐Term Demographic Data in Projecting Long‐Term Fate of Populations

Short‐term surveys are useful in conservation of species if they can be used to reliably predict the long‐term fate of populations. However, statistical evaluations of reliability are rare. We studied how well short‐term demographic data (1999–2002) of tartar catchfly (Silene tatarica), a perennial riparian plant, projected the fate and growth of 23 populations of this species up to the year 2010. Surveyed populations occurred along a river with natural flood dynamics and along a regulated river. Riparian plant populations are affected by flooding, which maintains unvegetated shores, while forest succession proceeds in areas with little flooding. Flooding is less severe along the regulated river, and vegetation overgrowth reduces abundance of tartar catchfly on unvegetated shores. We built matrix models to calculate population growth rates and estimated times to population extinction in natural and in regulated rivers, 13 and 10 populations, respectively. Models predicted population survival well (model predictions matched observed survival in 91% of populations) and accurately predicted abundance increases and decreases in 65% of populations. The observed and projected population growth rates differed significantly in all but 3 populations. In most cases, the model overestimated population growth. Model predictions did not improve when data from more years were used (1999–2006). In the regulated river, the poorest model predictions occurred in areas where cover of other plant species changed the fastest. Although vegetation cover increased in most populations, it decreased in 4 populations along the natural river. Our results highlight the need to combine disturbance and succession dynamics in demographic models and the importance of habitat management for species survival along regulated rivers. Precisión de Datos Demográficos de Corto Plazo en la Proyección del Destino de Poblaciones a Largo Plazo     

Effects of Social Structure and Prey Dynamics on Extinction Risk in Gray Wolves

Extinction models based on diffusion theory generally fail to incorporate two important aspects of population biology—social structure and prey dynamics. We include these aspects in an individual-based extinction model for small, isolated populations of the gray wolf (Canis lupus). Our model predicts mean times to extinction significantly longer than those predicted by more general (diffusion) models. According to our model, an isolated population of 50 wolves has a 95% chance of surviving just 9 years and only a 30% chance of surviving beyond 100 years. Reflecting the influence of social structure, a wolf population initially comprising 50 individuals is expected to persist only a few years longer, on average (71 years), than is a population initially comprising just a single reproductive pair (62 years). In contrast, substantially greater average prey abundance leads to dramatically longer expected persistence times. Autocorrelated prey dynamics result in a more complex distribution of extinction times than predicted by many extinction models. We contend that demographic stochasticity may pose the greatest threat to small, isolated wolf populations, although environmental stochasticity and genetic effects may compound this threat. Our work highlights the importance of considering social structure and resource dynamics in the development of population viability analyses.     

Extinction and Colonization of Birds on Habitat Islands

Abstract: We used point-count and transect surveys to estimate the distribution and abundance of eight scrub-breeding bird species in 34 habitat fragments and the urban matrix in southern California. We then calculated local extinction and colonization rates by comparing our data with surveys conducted in 1987. We classified factors that influence extinction and colonization rates into two types: (1) extrinsic factors, which are characteristics of the habitat fragments such as area, age, and isolation and (2) intrinsic factors, which are characteristics of the species that inhabit fragments, such as body size and population density. Over the past decade, at least one species went locally extinct in over 50% of the fragments, and local extinctions were almost twice as common as colonizations. Fragment size and, to a lesser extent, fragment age were the most important extrinsic factors determining extinction and colonization. Density indices of scrub birds were the most important intrinsic factors determining extinction rates, predicting the number of sites occupied, the probability of local extinction, relative area requirements, and time to local extinction.     

Predicting [corrected] extinction risk in spite of predator-prey oscillations.

Most population viability analyses (PVA) assume that the effects of species interactions are subsumed by population-level parameters. We examine how robust five commonly used PVA models are to violations of this assumption. We develop a stochastic, stage-structured predator-prey model and simulate prey population vital rates and abundance. We then use simulated data to parameterize and estimate risk for three demographic models (static projection matrix, stochastic projection matrix, stochastic vital rate matrix) and two time series models (diffusion approximation [DA], corrupted diffusion approximation [CDA]). Model bias is measured as the absolute deviation between estimated and observed quasi-extinction risk. Our results highlight three generalities about the application of single-species models to multi-species conservation problems. First, our collective model results suggest that most single-species PVA models overestimate extinction risk when species interactions cause periodic variation in abundance. Second, the DA model produces the most (conservatively) biased risk forecasts. Finally, the CDA model is the most robust PVA to population cycles caused by species interactions. CDA models produce virtually unbiased and relatively precise risk estimates even when populations cycle strongly. High performance of simple time series models like the CDA owes to their ability to effectively partition stochastic and deterministic sources of variation in population abundance.     

Evaluation of Local Ecological Knowledge as a Method for Collecting Extensive Data on Animal Abundance

Abstract:  The use of local ecological knowledge (LEK) has been advocated for biodiversity monitoring and management. To date, however, it has been underused in studying wild populations of animals and, particularly, in obtaining quantitative abundance estimates. We evaluated LEK as a tool for collecting extensive data on local animal abundance and population trends. We interviewed shepherds in southeastern Spain, asking them to estimate the local abundance of the terrestrial tortoise Testudo graeca . We quantified reliability of abundance estimates derived from interviews by comparing them with those obtained from standard field-sampling protocols (distance sampling). We also explored the complementarity of these 2 approaches. LEK provided high-quality and low-cost information about both distribution and abundance of T. graeca . Interviews with shepherds yielded abundance estimates in a much wider range than linear transects, which only detected the species in the upper two-thirds of its abundance range. Abundance estimates from both methodologies showed a close relationship. Analysis of confidence intervals indicated local knowledge could be used to estimate mean local abundances and to detect mean population trends. A cost analysis determined that the information derived from LEK was 100 times cheaper than that obtained through linear-transect surveys. Our results should further the use of LEK as a standard tool for sampling the quantitative abundance of a great variety of taxa, particularly when population densities are low and traditional sampling methods are expensive or difficult to implement.     

Neighborhood and habitat effects on vital rates: expansion of the Barred Owl in the Oregon coast ranges

In this paper, we modify dynamic occupancy models developed for detection-nondetection data to allow for the dependence of local vital rates on neighborhood occupancy, where neighborhood is defined very flexibly. Such dependence of occupancy dynamics on the status of a relevant neighborhood is pervasive, yet frequently ignored. Our framework permits joint inference about the importance of neighborhood effects and habitat covariates in determining colonization and extinction rates. Our specific motivation is the recent expansion of the Barred Owl (Strix varia) in western Oregon, USA, over the period 1990-2010. Because the focal period was one of dramatic range expansion and local population increase, the use of models that incorporate regional occupancy (sources of colonists) as determinants of dynamic rate parameters is especially appropriate. We began our analysis of 21 years of Barred Owl presence/nondetection data in the Tyee Density Study Area (TDSA) by testing a suite of six models that varied only in the covariates included in the modeling of detection probability. We then tested whether models that used regional occupancy as a covariate for colonization and extinction outperformed models with constant or year-specific colonization or extinction rates. Finally we tested whether habitat covariates improved the AIC of our models, focusing on which habitat covariates performed best, and whether the signs of habitat effects are consistent with a priori hypotheses. We conclude that all covariates used to model detection probability lead to improved AIC, that regional occupancy influences colonization and extinction rates, and that habitat plays an important role in determining extinction and colonization rates. As occupancy increases from low levels toward equilibrium, colonization increases and extinction decreases, presumably because there are more and more dispersing juveniles. While both rates are affected, colonization increases more than extinction decreases. Colonization is higher and extinction is lower in survey polygons with more riparian forest. The effects of riparian forest on extinction rates are greater than on colonization rates. Model results have implications for management of the invading Barred Owl, both through habitat alteration and removal.     

A framework for adapting survey design through time for wildlife population assessment

Sampling strategies for monitoring the status and trends in wildlife populations are often determined before the first survey is undertaken. However, there may be little information about the distribution of the population and so the sample design may be inefficient. Through time, as data are collected, more information about the distribution of animals in the survey region is obtained but it can be difficult to incorporate this information in the survey design. This paper introduces a framework for monitoring motile wildlife populations within which the design of future surveys can be adapted using data from past surveys whilst ensuring consistency in design-based estimates of status and trends through time. In each survey, part of the sample is selected from the previous survey sample using simple random sampling. The rest is selected with inclusion probability proportional to predicted abundance. Abundance is predicted using a model constructed from previous survey data and covariates for the whole survey region. Unbiased design-based estimators of status and trends and their variances are derived from two-phase sampling theory. Simulations over the short and long-term indicate that in general more precise estimates of status and trends are obtained using this mixed strategy than a strategy in which all of the sample is retained or all selected with probability proportional to predicted abundance. Furthermore the mixed strategy is robust to poor predictions of abundance. Estimates of status are more precise than those obtained from a rotating panel design.     

Effects of uncertainty and variability on population declines and IUCN Red List classifications

The International Union for Conservation of Nature (IUCN) Red List Categories and Criteria is a quantitative framework for classifying species according to extinction risk. Population models may be used to estimate extinction risk or population declines. Uncertainty and variability arise in threat classifications through measurement and process error in empirical data and uncertainty in the models used to estimate extinction risk and population declines. Furthermore, species traits are known to affect extinction risk. We investigated the effects of measurement and process error, model type, population growth rate, and age at first reproduction on the reliability of risk classifications based on projected population declines on IUCN Red List classifications. We used an age‐structured population model to simulate true population trajectories with different growth rates, reproductive ages and levels of variation, and subjected them to measurement error. We evaluated the ability of scalar and matrix models parameterized with these simulated time series to accurately capture the IUCN Red List classification generated with true population declines. Under all levels of measurement error tested and low process error, classifications were reasonably accurate; scalar and matrix models yielded roughly the same rate of misclassifications, but the distribution of errors differed; matrix models led to greater overestimation of extinction risk than underestimations; process error tended to contribute to misclassifications to a greater extent than measurement error; and more misclassifications occurred for fast, rather than slow, life histories. These results indicate that classifications of highly threatened taxa (i.e., taxa with low growth rates) under criterion A are more likely to be reliable than for less threatened taxa when assessed with population models. Greater scrutiny needs to be placed on data used to parameterize population models for species with high growth rates, particularly when available evidence indicates a potential transition to higher risk categories.     

Conserving the abundance of nonthreatened species

Human modification of the environment is driving declines in population size and distributional extent of much of the world's biota. These declines extend to many of the most abundant and widespread species, for which proportionally small declines can result in the loss of vast numbers of individuals, biomass, and interactions. These losses could have major localized effects on ecological and cultural processes and services without elevating a species’ global extinction risk. Although most conservation effort is directed at species threatened with extinction in the very near term, the value of retaining abundance regardless of global extinction risk is justifiable based on many biodiversity or ecosystem service metrics, including cultural services, at scales from local to global. The challenges of identifying conservation priorities for widespread and abundant species include quantifying the effects of species’ abundance on services and understanding how these effects are realized as populations decline. Negative effects of population declines may be disconnected from the threat processes driving declines because of species movements and environment flows (e.g., hydrology). Conservation prioritization for these species shares greater similarity with invasive species risk assessments than extinction risk assessments because of the importance of local context and per capita effects of abundance on other species. Because conservation priorities usually focus on preventing the extinction of threatened species, the rationale and objectives for incorporating declines of nonthreatened species must be clearly articulated, going beyond extinction risk to encompass the range of likely harmful effects (e.g., secondary extinctions, loss of ecosystem services) if declines persist or are not reversed. Research should focus on characterizing the effects of local declines in species that are not threatened globally across a range of ecosystem services and quantifying the spatial distribution of these effects through the distribution of abundance. The case for conserving abundance in nonthreatened species can be made most powerfully when the costs of losing this abundance are better understood.     

A comparison of models for predicting population persistence

We consider a range of models that may be used to predict the future persistence of populations, particularly those based on discrete-state Markov processes. While the mathematical theory of such processes is very well-developed, they may be difficult to work with when attempting to estimate parameters or expected times to extinction. Hence, we focus on diffusion and other approximations to these models, presenting new and recent developments in parameter estimation for density dependent processes, and the calculation of extinction times for processes subject to catastrophes. We illustrate these and other methods using data from simulated and real time series. We give particular attention to a procedure, due to Ross et al. [Ross, J.V., Taimre, T., Pollett, P.K. On parameter estimation in population models, Theor. Popul. Biol., in press], for estimating the parameters of the stochastic SIS logistic model, and demonstrate ways in which these parameters may be used to estimate expected extinction times. Although the stochastic SIS logistic model is strictly density dependent and allows only for birth and death events, it nonetheless may be used to predict extinction times with some accuracy even for populations that are only weakly density dependent, or that are subject to catastrophes.     

Toward accurate and precise estimates of lion density

Reliable estimates of animal density are fundamental to understanding ecological processes and population dynamics. Furthermore, their accuracy is vital to conservation because wildlife authorities rely on estimates to make decisions. However, it is notoriously difficult to accurately estimate density for wide‐ranging carnivores that occur at low densities. In recent years, significant progress has been made in density estimation of Asian carnivores, but the methods have not been widely adapted to African carnivores, such as lions (Panthera leo). Although abundance indices for lions may produce poor inferences, they continue to be used to estimate density and inform management and policy. We used sighting data from a 3‐month survey and adapted a Bayesian spatially explicit capture‐recapture (SECR) model to estimate spatial lion density in the Maasai Mara National Reserve and surrounding conservancies in Kenya. Our unstructured spatial capture‐recapture sampling design incorporated search effort to explicitly estimate detection probability and density on a fine spatial scale, making our approach robust in the context of varying detection probabilities. Overall posterior mean lion density was estimated to be 17.08 (posterior SD 1.310) lions >1 year old/100 km², and the sex ratio was estimated at 2.2 females to 1 male. Our modeling framework and narrow posterior SD demonstrate that SECR methods can produce statistically rigorous and precise estimates of population parameters, and we argue that they should be favored over less reliable abundance indices. Furthermore, our approach is flexible enough to incorporate different data types, which enables robust population estimates over relatively short survey periods in a variety of systems. Trend analyses are essential to guide conservation decisions but are frequently based on surveys of differing reliability. We therefore call for a unified framework to assess lion numbers in key populations to improve management and policy decisions.     

Detecting Population Declines over Large Areas with Presence-Absence, Time-to-Encounter, and Count Survey Methods

Abstract:  Ecologists often discount presence-absence surveys as a poor way to gain insight into population dynamics, in part because these surveys are not amenable to many standard statistical tests. Still, presence-absence surveys are sometimes the only feasible alternative for monitoring large areas when funds are limited, especially for sparse or difficult-to-detect species. I undertook a detailed simulation study to compare the power of presence-absence, count, and time-to-encounter surveys to detect regional declines in a population. I used a modeling approach that simulates both population numbers and the monitoring process, accounting for observation and other measurement errors. In gauging the efficacy of presence-absence surveys versus other approaches, I varied the number of survey sites, the spatial variation in encounter rate, the mean encounter rate, and the type of population loss. My results showed that presence-absence data can be as or more powerful than count data in many cases. Quantitative guidelines for choosing between presence-absence surveys and count surveys depend on the biological and logistical constraints governing a conservation monitoring situation. Generally, presence-absence surveys work best when there is little variability in abundance among the survey sites, the organism is rare, and the species is difficult to detect so that the time spent getting to each survey site is less than or equal to the time spent surveying each site. Count surveys work best otherwise. I present a case study with count data on the Northern Flicker ( Colaptes auratus ) from the North American Breeding Bird Survey to illustrate how the method might be used with field-survey data. The case study demonstrates that a count survey would be the most cost-effective design but would entail reduction in the number of sites. If this site reduction is not desirable, a presence-absence survey would be the most cost-effective survey.     

Reservoir‐host amplification of disease impact in an endangered amphibian

Emerging wildlife pathogens are an increasing threat to biodiversity. One of the most serious wildlife diseases is chytridiomycosis, caused by the fungal pathogen, Batrachochytrium dendrobatidis (Bd), which has been documented in over 500 amphibian species. Amphibians vary greatly in their susceptibility to Bd; some species tolerate infection, whereas others experience rapid mortality. Reservoir hosts—species that carry infection while maintaining high abundance but are rarely killed by disease—can increase extinction risk in highly susceptible, sympatric species. However, whether reservoir hosts amplify Bd in declining amphibian species has not been examined. We investigated the role of reservoir hosts in the decline of the threatened northern corroboree frog (Pseudophryne pengilleyi) in an amphibian community in southeastern Australia. In the laboratory, we characterized the response of a potential reservoir host, the (nondeclining) common eastern froglet (Crinia signifera), to Bd infection. In the field, we conducted frog abundance surveys and Bd sampling for both P. pengilleyi and C. signifera. We built multinomial logistic regression models to test whether Crinia signifera and environmental factors were associated with P. pengilleyi decline. C. signifera was a reservoir host for Bd. In the laboratory, many individuals maintained intense infections (>1000 zoospore equivalents) over 12 weeks without mortality, and 79% of individuals sampled in the wild also carried infections. The presence of C. signifera at a site was strongly associated with increased Bd prevalence in sympatric P. pengilleyi. Consistent with disease amplification by a reservoir host, P. pengilleyi declined at sites with high C. signifera abundance. Our results suggest that when reservoir hosts are present, population declines of susceptible species may continue long after the initial emergence of Bd, highlighting an urgent need to assess extinction risk in remnant populations of other declined amphibian species.     

Leveraging social media and deep learning to detect rare megafauna in video surveys

Deep learning has become a key tool for the automated monitoring of animal populations with video surveys. However, obtaining large numbers of images to train such models is a major challenge for rare and elusive species because field video surveys provide few sightings. We designed a method that takes advantage of videos accumulated on social media for training deep-learning models to detect rare megafauna species in the field. We trained convolutional neural networks (CNNs) with social media images and tested them on images collected from field surveys. We applied our method to aerial video surveys of dugongs (Dugong dugon) in New Caledonia (southwestern Pacific). CNNs trained with 1303 social media images yielded 25% false positives and 38% false negatives when tested on independent field video surveys. Incorporating a small number of images from New Caledonia (equivalent to 12% of social media images) in the training data set resulted in a nearly 50% decrease in false negatives. Our results highlight how and the extent to which images collected on social media can offer a solid basis for training deep-learning models for rare megafauna detection and that the incorporation of a few images from the study site further boosts detection accuracy. Our method provides a new generation of deep-learning models that can be used to rapidly and accurately process field video surveys for the monitoring of rare megafauna.     

Will Observation Error and Biases Ruin the Use of Simple Extinction Models?

Abstract: Estimating the risk of extinction for populations of endangered species is an important component of conservation biology. These estimates must be made from data that contain both environmental noise in the year-to-year transitions in population size (so-called "process error"), random errors in sampling, and possible biases in sampling ( both forms of observation errors). To determine how much faith to place in estimated extinction rates, it is important to know how sensitive they are to observation error. We used three simple, commonly employed models of population dynamics to generate simulated population time series. We then combined random observation error or systematic biases with those data, fit models to the time series data, and observed how close the extinction dynamics of the fitted models compared with the dynamics of the underlying models. We found that systematic biases in sampling rarely affected estimates of extinction risk. We also found that even moderate levels of random observation error do not significantly affect extinction estimates except over a small range of process errors, corresponding to the region where extinction risk is most uncertain. With more substantial sampling error, estimates of extinction risk degraded rapidly. Field census techniques for a variety of taxa often involve observation errors within ±32% of actual population sizes. For typical time series used in conservation, therefore, we often may not need to be overly concerned about observation errors as an extra source of imperfection in our estimated extinction rates.