Fitting Occupancy Models Research Articles

Efficient statistical inference methods for assessing changes in species’ populations using citizen science data

Abstract The global decline of biodiversity, driven by habitat degradation and climate breakdown, is a significant concern. Accurate measures of change are crucial to provide reliable evidence of species’ population changes. Meanwhile citizen science data have witnessed a remarkable expansion in both quantity and sources and serve as the foundation for assessing species’ status. The growing data reservoir presents opportunities for novel and improved inference but often comes with computational costs: computational efficiency is paramount, especially as regular analysis updates are necessary. Building upon recent research, we present illustrations of computationally efficient methods for fitting new models, applied to three major citizen science data sets for butterflies. We extend a method for modelling abundance changes of seasonal organisms, firstly to accommodate multiple years of count data efficiently, and secondly for application to counts from a snapshot mass-participation survey. We also present a variational inference approach for fitting occupancy models efficiently to opportunistic citizen science data. The continuous growth of citizen science data offers unprecedented opportunities to enhance our understanding of how species respond to anthropogenic pressures. Efficient techniques in fitting new models are vital for accurately assessing species’ status, supporting policy-making, setting measurable targets, and enabling effective conservation efforts.

Insights into surveying pangolins using ground and arboreal camera traps

Abstract Arboreal camera traps are becoming more commonly used for monitoring wildlife. Pangolins (Order: Pholidota) are a threatened group of mammals that are challenging to monitor across their range. In this study, we assessed the use of arboreal and ground camera traps for monitoring the three pangolin species native to West Africa in the Ziama Man and Biosphere Reserve, Guinea. We fit occupancy models to our data to examine the effect of factors related to camera height and tree height on detection probability. In addition, we evaluated the utility of deploying multiple cameras within the same tree. Our study showed that arboreal camera traps can successfully detect both arboreal pangolin species, with the highest detection in mid-canopy for white-bellied pangolin and mid-to high-canopy for black-bellied pangolins. In addition, our results suggest at least 4–6 cameras deployed on each tree to maximize the opportunity of detecting these species. We did not detect giant pangolins. Further studies are needed to continue improving detection of all three pangolins for monitoring and adaptive management of these heavily harvested and traded species.

Considerations for fitting occupancy models to data from eBird and similar volunteer-collected data

Abstract An occupancy model makes use of data that are structured as sets of repeated visits to each of many sites, in order to estimate the actual probability of occupancy (i.e. proportion of occupied sites) after correcting for imperfect detection using the information contained in the sets of repeated observations. We explore the conditions under which preexisting, volunteer-collected data from the citizen science project eBird can be used for fitting occupancy models. Because the majority of eBird’s data are not collected in the form of repeated observations at individual locations, we explore 2 ways in which the single-visit records could be used in occupancy models. First, we assess the potential for space-for-time substitution: aggregating single-visit records from different locations within a region into pseudo-repeat visits. On average, eBird’s observers did not make their observations at locations that were representative of the habitat in the surrounding area, which would lead to biased estimates of occupancy probabilities when using space-for-time substitution. Thus, the use of space-for-time substitution is not always appropriate. Second, we explored the utility of including data from single-visit records to supplement sets of repeated-visit data. In a simulation study we found that inclusion of single-visit records increased the precision of occupancy estimates, but only when detection probabilities are high. When detection probability was low, the addition of single-visit records exacerbated biases in estimates of occupancy probability. We conclude that subsets of data from eBird, and likely from similar projects, can be used for occupancy modeling either using space-for-time substitution or supplementing repeated-visit data with data from single-visit records. The appropriateness of either alternative will depend on the goals of a study and on the probabilities of detection and occupancy of the species of interest.

A Gibbs sampler for multi-species occupancy models

Multi-species occupancy (MSO) models use detection-nondetection data from species observed at different locations to estimate the probability that a particular species occupies a particular geographical region. The models are particularly useful for estimating the occupancy probabilities associated with rare species since they are seldom observed when undertaking field surveys. In this paper, we develop Gibbs sampling algorithms that can be used to fit various Bayesian MSO models to detection-nondetection data. Bayesian analysis of these models can be undertaken using statistical packages such as JAGS, Stan, and NIMBLE. However, since these packages were not developed specifically to fit occupancy models, one often experiences long run-times when undertaking analysis. However, we find that these packages that were not developed specifically to fit MSO models are less efficient than our special-purpose Gibbs sampling algorithms.

Monitoring mesocarnivores with tracks and technology using multi‐method modeling

AbstractMesocarnivores play important ecological roles and are valued by diverse stakeholders. These species are often the focus of conservation efforts or are managed for sustainable harvest. Management actions require accurate population monitoring, but such monitoring is challenging because mesocarnivores are elusive and persist at low densities. We addressed this challenge by evaluating 2 monitoring methods (scent stations and motion‐sensitive cameras) using multi‐method modeling. We estimated occurrence probabilities and habitat relationships for 8 mesocarnivore species by fitting occupancy models to data collected at 75 sites from October to December 2021 across a 3,200‐km2system in New Hampshire, USA. We assessed the relative estimated precision of the methodological approaches and their costs. We also evaluated tradeoffs in occurrence estimation and uncertainty among study designs by analyzing simulations run across various numbers of study sites and 2 study durations. Cameras cost roughly 10 times more than scent stations but strongly outperformed them in terms of species' detectability and parameter estimate precision. Multi‐method models yielded the most precise estimates of occurrence probability and habitat relationships. Parameter estimates were on average twice as precise for camera and multi‐method models compared to scent stations. Additionally, the estimated precision and direction (positive or negative) of habitat relationships varied with the method employed. Longer camera deployments, additional study sites, and multi‐method approaches nearly always reduced uncertainty, but these reductions were species‐specific and generally most pronounced for more rarely detected species. Overall, our results demonstrate the utility of motion‐sensitive cameras traps for monitoring mesocarnivores while revealing the additional benefits of multi‐method modeling. Our results also provide guidance for designing monitoring programs for mesocarnivores while navigating tradeoffs between study design, cost, and uncertainty. Despite its benefits, multi‐method modeling remains uncommon as a general monitoring approach. We suggest managers consider this approach in light of existing datasets and design monitoring programs that integrate traditional methods with emergent technologies.

Space-for-time is not necessarily a substitution when monitoring the distribution of pelagic fishes in the San Francisco Bay-Delta.

Occupancy models are often used to analyze long‐term monitoring data to better understand how and why species redistribute across dynamic landscapes while accounting for incomplete capture. However, this approach requires replicate detection/non‐detection data at a sample unit and many long‐term monitoring programs lack temporal replicate surveys. In such cases, it has been suggested that surveying subunits within a larger sample unit may be an efficient substitution (i.e., space‐for‐time substitution). Still, the efficacy of fitting occupancy models using a space‐for‐time substitution has not been fully explored and is likely context dependent. Herein, we fit occupancy models to Delta Smelt (Hypomesus transpacificus) and Longfin Smelt (Spirinchus thaleichthys) catch data collected by two different monitoring programs that use the same sampling gear in the San Francisco Bay‐Delta, USA. We demonstrate how our inferences concerning the distribution of these species changes when using a space‐for‐time substitution. Specifically, we found the probability that a sample unit was occupied was much greater when using a space‐for‐time substitution, presumably due to the change in the spatial scale of our inferences. Furthermore, we observed that as the spatial scale of our inferences increased, our ability to detect environmental effects on system dynamics was obscured, which we suspect is related to the tradeoffs associated with spatial grain and extent. Overall, our findings highlight the importance of considering how the unique characteristics of monitoring programs influences inferences, which has broad implications for how to appropriately leverage existing long‐term monitoring data to understand the distribution of species.

Making inference with messy (citizen science) data: when are data accurate enough and how can they be improved?

Measurement or observation error is common in ecological data: as citizen scientists and automated algorithms play larger roles processing growing volumes of data to address problems at large scales, concerns about data quality and strategies for improving it have received greater focus. However, practical guidance pertaining to fundamental data quality questions for data users or managers-how accurate do data need to be and what is the best or most efficient way to improve it?-remains limited. We present a generalizable framework for evaluating data quality and identifying remediation practices, and demonstrate the framework using trail camera images classified using crowdsourcing to determine acceptable rates of misclassification and identify optimal remediation strategies for analysis using occupancy models. We used expert validation to estimate baseline classification accuracy and simulation to determine the sensitivity of two occupancy estimators (standard and false-positive extensions) to different empirical misclassification rates. We used regression techniques to identify important predictors of misclassification and prioritize remediation strategies. More than 93% of images were accurately classified, but simulation results suggested that most species were not identified accurately enough to permit distribution estimation at our predefined threshold for accuracy (<5% absolute bias). A model developed to screen incorrect classifications predicted misclassified images with >97% accuracy: enough to meet our accuracy threshold. Occupancy models that accounted for false-positive error provided even more accurate inference even at high rates of misclassification (30%). As simulation suggested occupancy models were less sensitive to additional false-negative error, screening models or fitting occupancy models accounting for false-positive error emerged as efficient data remediation solutions. Combining simulation-based sensitivity analysis with empirical estimation of baseline error and its variability allows users and managers of potentially error-prone data to identify and fix problematic data more efficiently. It may be particularly helpful for "big data" efforts dependent upon citizen scientists or automated classification algorithms with many downstream users, but given the ubiquity of observation or measurement error, even conventional studies may benefit from focusing more attention upon data quality.

Efficient Bayesian analysis of occupancy models with logit link functions.

Occupancy models (Ecology, 2002; 83: 2248) were developed to infer the probability that a species under investigation occupies a site. Bayesian analysis of these models can be undertaken using statistical packages such as WinBUGS, OpenBUGS, JAGS, and more recently Stan, however, since these packages were not developed specifically to fit occupancy models, one often experiences long run times when undertaking an analysis. Bayesian spatial single‐season occupancy models can also be fit using the R package stocc. The approach assumes that the detection and occupancy regression effects are modeled using probit link functions. The use of the logistic link function, however, is algebraically more tractable and allows one to easily interpret the coefficient effects of an estimated model by using odds ratios, which is not easily done for a probit link function for models that do not include spatial random effects. We develop a Gibbs sampler to obtain posterior samples from the posterior distribution of the parameters of various occupancy models (nonspatial and spatial) when logit link functions are used to model the regression effects of the detection and occupancy processes. We apply our methods to data extracted from the 2nd Southern African Bird Atlas Project to produce a species distribution map of the Cape weaver (Ploceus capensis) and helmeted guineafowl (Numida meleagris) for South Africa. We found that the Gibbs sampling algorithm developed produces posterior samples that are identical to those obtained when using JAGS and Stan and that in certain cases the posterior chains mix much faster than those obtained when using JAGS, stocc, and Stan. Our algorithms are implemented in the R package, Rcppocc. The software is freely available and stored on GitHub (https://github.com/AllanClark/Rcppocc).

Detecting spatial ontogenetic niche shifts in complex dendritic ecological networks

AbstractOntogenetic niche shifts (ONS) are important drivers of population and community dynamics, but they can be difficult to identify for species with prolonged larval or juvenile stages, or for species that inhabit continuous habitats. Most studies of ONS focus on single transitions among discrete habitat patches at local scales. However, for species with long larval or juvenile periods, affinity for particular locations within connected habitat networks may differ among cohorts. The resulting spatial patterns of distribution can result from a combination of landscape‐scale habitat structure, position of a habitat patch within a network, and local habitat characteristics—all of which may interact and change as individuals grow. We estimated such spatial ONS for spring salamanders (Gyrinophilus porphyriticus), which have a larval period that can last 4 years or more. Using mixture models to identify larval cohorts from size frequency data, we fit occupancy models for each age class using two measures of the branching structure of stream networks and three measures of stream network position. Larval salamander cohorts showed different preferences for the position of a site within the stream network, and the strength of these responses depended on the basin‐wide spatial structure of the stream network. The isolation of a site had a stronger effect on occupancy in watersheds with more isolated headwater streams, while the catchment area, which is associated with gradients in stream habitat, had a stronger effect on occupancy in watersheds with more paired headwater streams. Our results show that considering the spatial structure of habitat networks can provide new insights on ONS in long‐lived species.

Penalized likelihood methods improve parameter estimates in occupancy models

Summary Occupancy models are employed in species distribution modelling to account for imperfect detection during field surveys. While this approach is popular in the literature, problems can occur when estimating the model parameters. In particular, the maximum likelihood estimates can exhibit bias and large variance for data sets with small sample sizes, which can result in estimated occupancy probabilities near 0 and 1 (‘boundary estimates’). In this paper, we explore strategies for estimating parameters based on maximizing a penalized likelihood. Penalized likelihood methods augment the usual likelihood with a penalty function that encodes information about what parameter values are undesirable. We introduce penalties for occupancy models that have analogues in ridge regression and Bayesian approaches, and we compare them to a penalty developed for occupancy models in prior work. We examine the bias, variance and mean squared error of parameter estimates obtained from each method on synthetic data. Across all of the synthetic data sets, the penalized estimation methods had lower mean squared error than the maximum likelihood estimates. We also provide an example of the application of these methods to point counts of avian species. Penalized likelihood methods show similar improvements when tested using empirical bird point count data. We discuss considerations for choosing among these methods when modelling occupancy. We conclude that penalized methods may be of practical utility for fitting occupancy models with small sample sizes, and we are releasing R code that implements these methods.

Prevalence of the generalist flea Pulex simulans on black-tailed prairie dogs (Cynomys ludovicianus) in New Mexico, USA: the importance of considering imperfect detection.

If a parasite is not detected during a survey, one of two explanations is possible: the parasite was truly absent or it was present but not detected. We fit occupancy models to account for imperfect detection when combing fleas (Siphonaptera) from black-tailed prairie dogs (Cynomys ludovicianus) during June-August 2012 in the Vermejo Park Ranch, New Mexico, USA. With the use of detection histories from combing events during monthly trapping sessions, we fit occupancy models for two flea species: Oropsylla hirusta (a prairie dog specialist) and Pulex simulans (a generalist). Detection probability was <100% for both species and about 21% lower for P. simulans. Pulex simulans may be especially difficult to detect because it is about half the size of O. hirusta. Monthly occupancy (prevalence) for P. simulans was estimated at 24% (June, 95% confidence interval = 19-30), 39% (July, 32-47), and 56% (August, 49-64) in new prairie dog colonies, and 43% (32-54), 61% (49-71), and 79% (70-87) in old colonies. These results suggest P. simulans can attain high prevalence on prairie dogs, especially in old colonies. If P. simulans is highly prevalent on prairie dogs, it may serve as a "bridge vector" between Cynomys and other mammalian hosts of the plague bacterium Yersinia pestis, and even function as a reservoir of Y. pestis between outbreaks.

Fitting occupancy models with E‐SURGE: hidden Markov modelling of presence–absence data

Summary Occupancy – the proportion of area occupied by a species – is a key notion for addressing important questions in ecology, biogeography and conservation biology. Occupancy models allow estimating and inferring about species occurrence while accounting for false absences (or imperfect species detection). Occupancy models can be formulated as hidden Markov models (HMM) in which the state process captures the Markovian dynamic of the actual but latent states, while the observation process consists of observations that are made from these underlying states. We show how occupancy models can be implemented in program E‐SURGE, which was initially developed to analyse capture–recapture data in the HMM framework. Replacing individuals by sites provides the user with access to several features of E‐SURGE that are not available altogether or just not available in standard occupancy software: i) flexible model specification through a user‐friendly syntax without having to write custom code, ii) decomposition of the observation and state processes in several steps to provide flexible parameterisation, iii) up‐to‐date diagnostics of model identifiability, and iv) advanced numerical algorithms to produce fast and reliable results (including site random effects). To illustrate E‐SURGE features, we provide implementation and analysis details for several occupancy models. We also provide simulated and real‐world examples as well as further specifications and information in a companion wiki platform http://occupancyinesurge.wikidot.com/.

Accounting for false positives improves estimates of occupancy from key informant interviews

AbstractAimMuch research in conservation biogeography is fundamentally dependent on obtaining reliable data on species distributions across space and time. Such data are now increasingly being generated using various types of public surveys. These data are often integrated with occupancy models to evaluate distributional patterns, range dynamics and conservation status of multiple species at broad spatio‐temporal scales. Occupancy models have traditionally corrected for imperfect detection due to false negatives while implicitly assuming that false positives do not occur. However, public survey data are also prone to false‐positive errors, which when unaccounted for can cause bias in occupancy estimates. We test whether false positives in a dataset collected from public surveys lead to overestimation of species site occupancy and whether estimators that simultaneously account for false‐positive and false‐negative errors improve occupancy estimates.LocationWestern Ghats, India.MethodsWe fit occupancy models that simultaneously account for false positives and negatives to data collected from a large‐scale key informant interview survey for 30 species of large vertebrates. We tested their performance against standard occupancy models that account only for false negatives.ResultsStandard occupancy models that correct only for false negatives tended to overestimate species occupancy due to false‐positive errors. Occupancy models that simultaneously accounted for false positives and negatives had greater support [lower Akaike's information criterion (AIC)] and, consistent with predictions, generated systematically lower occupancy estimates than standard models. Furthermore, accounting for false positives improved the accuracy of occupancy estimates despite the added complexity to the statistical estimator.Main conclusionsIntegrating large‐scale public surveys with occupancy modelling approaches is a powerful tool for informing conservation and management. However, in many if not most cases, it will be important to explicitly account for false positives to ensure the reliability of occupancy estimates obtained from public survey datasets such as key informant interviews, volunteer surveys, citizen science programmes, historical archives and acoustic surveys.