Using informative priors to account for identifiability issues in occupancy models with identification errors

On the use of automated cameras to estimate species richness for large‐ and medium‐sized rainforest mammals

National Parks across America play an important role in protecting natural resources and providing access to recreation for visitors. However, these goals may come into conflict as visitation rates rise. Grand Canyon National Park in Northern Arizona is one of the most highly visited parks in the United States, with over 6 million visitors a year. Backcountry hiking and camping are popular activities in the park, and many highly visited hiking trails and campgrounds overlap with known breeding areas of a threatened species, Mexican Spotted Owl. In this thesis, I explore the intersection of recreation and wildlife conservation at this popular park through the lens of long-term occupancy of a threatened species. My aims are to (1) assess the potential impact of visitor use on long-term occupancy (2001 to 2021) of Mexican Spotted Owls at the Grand Canyon, and (2) evaluate the potential for autonomous recording units (ARUs) to complement current survey protocols. To assess long-term occupancy, I ran a multi-season occupancy model using 20-years of call-back survey data conducted in protected activity centers (PACs), along with measures of visitor use and habitat characteristics. To assess the use of ARUs, I ran a single-season occupancy model using three years of data, which was collected using autonomous recording units in PACs from 2019 to 2021. I found that visitor use in the Grand Canyon had no effect on owl occupancy, which remained stable across PACs over the 20-year study period. Owl occupancy remained high across the 20-year survey period and was strongly informed by habitat characteristics. Specifically, Mexican Spotted Owls occupied PACs with higher proportions of mixed shrubland habitat and Supai formation. Conversely, owl occupancy decreased in PACs with more pinyon-juniper woodland habitat and Redwall Limestone. Assessing the use of ARUs as a complement to current protocol, ARUs were found to be a useful tool for supplementing traditional call-back surveys, particularly at PACs with extremely limited access. In particular, ARUs detected Mexican Spotted Owls with high probability early in the breeding season prior to the official call-back survey period, which allows managers to extend their monitoring period. In highly remote PACs, ARUs were more suitable than call-backs because they could collect more data with less effort. Incorporating this method into Spotted Owl survey protocol may be essential for improving monitoring of under-sampled locations, which is a critical component for assessing long-term trends for this species across its range.

While camera traps can effectively detect semi-aquatic mammal species, they are also often temporally and monetarily inefficient and have a difficult time detecting smaller bodied, elusive mammals. Recent studies have shown that extracting DNA from environmental samples can be a non-invasive, alternative method of detecting elusive species. Environmental DNA (eDNA) has not yet been used to survey American mink (Neogale vison), a cryptic and understudied North American mustelid. To help determine best survey practices for the species, we compared the effectiveness and efficiency of eDNA and camera traps in surveys for American mink. We used both methods to monitor the shoreline of seven bodies of water in northeastern Indiana from March to May 2021. We extracted DNA from filtered environmental water samples and used quantitative real-time PCR to determine the presence of mink at each site. We used Akaike's Information Criterion to rank probability of detection models with and without survey method as a covariate. We detected mink at four of the seven sites and seven of the 21 total survey weeks using camera traps (probability of detection (ρ) = 0.36). We detected mink at five sites and during five survey weeks using eDNA (ρ = 0.25). However, the highest probability of detection was obtained when both methods were combined, and data were pooled (ρ = 0.47). Survey method did not influence model fit, suggesting no difference in detectability between camera traps and eDNA. Environmental DNA was twice as expensive, but only required a little over half (58%) of the time when compared to camera trapping. We recommend ways in which an improved eDNA methodology may be more cost effective for future studies. For this study, a combination of both methods yielded the highest probability for detecting mink presence.

Subterranean habitats represent focal habitats in many conservation strategies; however, these environments are some of the most difficult to sample. New sampling methods, such as environmental DNA (eDNA), show promise to improve stygobiont detection, but sources of sampling bias are poorly understood. Therefore, we determined the factors affecting detection probability using traditional visual surveys and eDNA surveys for both cavefishes and cave crayfishes and demonstrated how detection affects survey efforts for these taxa. We sampled 40 sites (179 visual and 183 eDNA surveys) across the Ozark Highlands ecoregion. We estimated the detection probability of cave crayfishes and cavefishes using both survey methods under varying environmental conditions. The effectiveness of eDNA or visual surveys varied by environmental conditions (i.e., water volume, prevailing substrate, and water velocity) and the target taxa. When sampling in areas with average water velocity, no flow, and coarse substrate, eDNA surveys had a higher detection probability (0.49) than visual surveys (0.35) for cavefishes and visual surveys (0.67) had a higher detection probability than eDNA surveys (0.40) for cave crayfishes. Under the same sampling conditions, 5 visual surveys compared to 10 eDNA surveys would be needed to confidently detect cave crayfishes and 9 visual surveys compared to 4 eDNA surveys for cavefishes. Environmental DNA is a complementary tool to traditional visual surveys; however, the limitations we identified indicate eDNA currently cannot replace visual surveys in subterranean environments. Although sampling designs that account for imperfect sampling are particularly useful, they may not be practical; thus, increasing sampling efforts to offset known detection bias would benefit conservation strategies.

Camera trapping of wildlife has been practiced since the early 20th century (Chapman, 1927), but in the last 20 years or so, camera traps have become readily available and much more affordable. As a result, they are becoming a mainstream tool in conservation and ecology, with uses ranging from simple species inventories (e.g. Silveira, Jacomo & Diniz, 2003), the discovery of new species (e.g. Rovero et al., 2008), through abundance estimation (e.g. Karanth, 1995), conservation assessments (e.g. Kinnaird et al., 2003; Linkie et al., 2006), population dynamics (e.g. Karanth et al., 2006) and forest ecology (e.g. Beck & Terborgh, 2002; Weckel, Giuliano & Silver, 2006; Kitamura et al., 2008). This explosion in camera trap use is reflected in 50% annual growth over the past decade in the number of published papers that either directly address camera trapping methods or use them as a research tool (Fig. 1). In this context, Tobler et al. have produced a useful template for obtaining unbiased species richness estimates from camera trap data. As they point out, a major benefit of this approach is that it can piggy back on studies that are targeted at specific species. Up to now, the single most common use of camera traps has been to estimate the abundance of cat species, relying on individual recognition from coat patterns (e.g. Karanth & Nichols, 1998; Maffei, Cuellar & Noss, 2004; Trolle & Kery, 2005). Methods that extend to a wider range of terrestrial species (including the many species that cannot be individually recognized) greatly increase the value of camera trap studies and techniques, and if used retrospectively could make use of a vast amount of data that has already been collected. However, Tobler et al. raise some questions that deserve further exploration. First, given that targeted studies usually focus on very elusive species, they need to maximize capture rate. This is often done by using baits or lures, or by focusing trap placement in areas frequently used by the target species. For large carnivores, this generally means placement on dirt roads or large game trails. This kind of selective placement has been shown to influence capture rates of some species (Trolle & Kery, 2005; Weckel et al., 2006), but despite the fact that Tobler et al.’s study was primarily aimed at jaguars, the results suggest that this factor need not influence species richness estimates. While this is encouraging, they do suggest that some forms of directed placement may violate the assumptions of the diversity models they recommend. If this is so, it will be important to explore further the survey designs and environmental conditions under which the available species richness estimators break down. While Tobler et al. focus on species richness estimates from camera trap data, they also show an interesting relationship between species body mass and the probability of triggering cameras. From this they conclude that trap rates cannot be used as an index to compare relative abundance across species. They rightly point out that there are factors other than abundance that influence trapping rate, particularly animal movement rates, body size and patterns of habitat use. Given these confounding factors, it would of course be wrong to assume that differences in trapping rates directly reflect differences in abundance,

Disruptions in fire regimes increasingly threaten biodiversity worldwide, yet the long‐term impacts of megafires on mammal assemblages in fire‐prone ecosystems remain poorly understood. The 2020 megafires in the Brazilian Pantanal, the world's largest tropical wetland, burned one‐third of its area, causing unprecedented biodiversity impacts and highlighting the need to understand ecological consequences. We assessed the effects of the 2020 megafires on mammal diversity and relative abundance in the Pantanal Matogrossense National Park using environmental DNA (eDNA) metabarcoding, camera trapping and Bayesian multispecies occupancy models. These methods allowed us to evaluate species, site and assemblage‐level responses to fire, accounting for imperfect detection and landscape heterogeneity. Despite the Pantanal being fire‐prone, we observed significant negative effects on mammals at multiple ecological levels, persisting 2 years post‐fire. Species richness declined from 84 species observed pre‐fire to 44 observed post‐fire. eDNA metabarcoding identified 38 species (23 exclusively), while camera traps recorded 21 species (6 exclusively). A significant negative correlation was found between burned area proportion and both species richness and aggregated relative abundance per site. Impacts were severe on large‐bodied, threatened species, such as the giant armadillo ( Priodontes maximus ), jaguar ( Panthera onca ) and lowland tapir ( Tapirus terrestris ). They experienced significant declines in relative abundance, highlighting the vulnerability of megafauna to severe wildfires. The combination of eDNA and camera trapping proved essential for capturing the full extent of biodiversity impacts in post‐fire landscapes. Synthesis and applications . Our findings highlight the severe and lasting consequences of the 2020 wildfires, observed 2 years after the event, with habitat loss and direct mortality leading to population declines for both forest‐dwelling and open‐habitat species. The complementary use of eDNA metabarcoding and camera trapping enhances biodiversity assessments, though gaps in reference databases remain. These results emphasize the urgent need for adaptive management strategies to mitigate fire impacts, restore habitats and bolster the Pantanal's resilience. Effective governance, stakeholder collaboration and integrated fire management policies are crucial for safeguarding this globally significant wetland.

Assessing species geographic distributions is critical to approximate their ecological niches, understand how global change may reshape their occurrence patterns, and predict their extinction risks. Yet, species records are over-aggregated across taxonomic, geographic, environmental, and anthropogenic dimensions. The under-sampling of remote locations biases the quantification of species geographic distributions and ecological niche for most species. Here, we used nearly one thousand environmental DNA (eDNA) samples across the world’s oceans, including polar regions and tropical remote islands, to determine the extent to which the geographic and ecological niche ranges of marine fishes are underestimated through the lens of global occurrence records based on conventional surveys. Our eDNA surveys revealed that the known geographic ranges for 93% of species and the ecological niche ranges for 7% of species were underestimated, and contributed to filling them. We show that the probability to detect a range filling for a given species is primarily shaped by the GBIF/OBIS sampling effort in a cell, but also by the number of occurrences available for the species. Most gap fillings were achieved by addressing a methodological sampling bias, notably when eDNA facilitated the detection of small fishes in previously sampled locations using conventional methods. Using a machine learning model, we found that a local effort of 10 eDNA samples would detect 24 additional fish species on average and a maximum of 98 species in previously unsampled tropical areas. Yet, a null model revealed that only half of ecological niche range fillings would be due to eDNA surveys, beyond a random allocation of classical sampling effort. Altogether, our results suggest that sampling in remote areas and performing eDNA surveys in over-sampled areas may both increase fish ecological niche ranges toward unexpected values with consequences in biodiversity modeling, management, and conservation.

Estimating habitat and spatial associations for wildlife is common across ecological studies and it is well known that individual traits can drive population dynamics and vice versa. Thus, it is commonly assumed that individual- and population-level data should represent the same underlying processes, but few studies have directly compared contemporaneous data representing these different perspectives. We evaluated the circumstances under which data collected from Lagrangian (individual-level) and Eulerian (population-level) perspectives could yield comparable inference to understand how scalable information is from the individual to the population. We used Global Positioning System (GPS) collar (Lagrangian) and camera trap (Eulerian) data for seven species collected simultaneously in eastern Washington (2018-2020) to compare inferences made from different survey perspectives. We fit the respective data streams to resource selection functions (RSFs) and occupancy models and compared estimated habitat- and space-use patterns for each species. Although previous studies have considered whether individual- and population-level data generated comparable information, ours is the first to make this comparison for multiple species simultaneously and to specifically ask whether inferences from the two perspectives differed depending on the focal species. We found general agreement between the predicted spatial distributions for most paired analyses, although specific habitat relationships differed. We hypothesize the discrepancies arose due to differences in statistical power associated with camera and GPS-collar sampling, as well as spatial mismatches in the data. Our research suggests data collected from individual-based sampling methods can capture coarse population-wide patterns for a diversity of species, but results differ when interpreting specific wildlife-habitat relationships.

The impact of humans on biodiversity, in the form of the spatially extensive occurrence of humans and subsequent habitat degradation, leads to negative interactions between humans and native wildlife. However, knowledge of the spatial and temporal interface between humans and wildlife is necessary to understand the root cause of such negative interactions, yet considerably understudied in the context of human-dominated landscapes in south and south-eastern Asia. We took this opportunity, gaining insights on seasonal spatial interaction and spatio-temporal overlap between sloth bears (Melursus ursinus) and humans, and subsequently predicted the conflict source sites and dispersion (i.e., hotspots) based on the robust geographic profiling (GP) method in the Sanjay Tiger Reserve (STR), a human-dominated landscape of central India. Detection data of sloth bear and human were obtained from camera trap survey conducted for two years (2017–2018) and records of conflict incidents (2009–2019) were collected from forest department. We found that sloth bears can co-occur with humans independently of seasons, based on occupancy models. However, during summer, higher temporal overlap (Δ4 = 0.46) and lower spatial overlap (0.31) were observed between sloth bears and humans. Contrastingly, lower temporal overlap (Δ4 = 0.29) and higher spatial overlap (0.44) were observed between the same two during winter. The activity patterns of sloth bears and humans differed significantly across seasons and within the same species in different seasons. Our findings indicated that significant changes in human activity, especially during summer, increased the likelihood of sloth bear-human interaction and subsequent conflict incidents. The mapping of conflict source and dispersion (with high accuracy) also predicted a greater probability of conflict during summer, compared to winter, and thus showed the successful application of GP models in this field. Also, camera trap data alone were able to predict the occurrence of hotspots, demonstrating the use of camera trap records in the successful prediction of source-dispersion of conflict. This study would be useful for decision-makers to alleviate sloth bear–human conflict based on insights on seasonal variation of spatio-temporal overlap between the two and direct conservation efforts accordingly.

Small macropodoid marsupials are well represented among Australia’s extinct and threatened mammals. Population monitoring is central to understanding how remaining species respond to on‐going landscape change and threatening processes on private land and within managed conservation reserves. Camera trapping and occupancy modelling provide a reliable approach to monitor these often cryptic species. However, understanding the survey effort required to detect population declines of a given magnitude with high statistical certainty is often overlooked. We use empirical data from camera trapping and occupancy modelling to determine an optimal survey design for a regionally endangered population of the black‐striped wallaby. We established a 1200 ha grid of 60 cameras and sampled in three 30‐day intervals over a 1‐year period. We investigated factors that may influence occupancy and detection followed by power analyses with our broad objective being to devise a monitoring program that would be robust enough to detect a 50% decline with at least 80% power. This species was not detected at rainforest sites. Naïve occupancy within eucalypt forest was 0.42. The detection probability (using 5‐day detection occasions) varied with season (from 0.17 ± 0.04 in autumn to 0.34 ± 0.05 in spring). Occupancy was negatively influenced by shrub cover and negatively influenced by elevation. We found no apparent influence of dingo detection on occupancy or detection of the black‐striped wallaby. We explored different configurations of detection sites and number of survey occasions to satisfy our 50:80 criteria when initial occupancy was 0.40. Sampling 60 sites for a minimum of 30 days would satisfy these criteria. Power analyses can inform optimal designs for threatened species monitoring and similar investigations should be conducted for other threatened small macropods to assist their conservation.

There is a growing body of evidence that weasel species are in decline globally. More data on their ecology and distribution are needed to plan and justify any conservation management actions. Camera trapping can be an effective survey method for many species; however, the small size and quick movements of weasels present challenges in detection and little consensus exists on practices for attracting them to improve detection. This study tested different combinations of meat baits and scent and audio lures to assess the most effective methods. Camera traps were set up in clusters of three at 42 sites to test the effectiveness of these combinations, accounting for season, in terms of the time to first detection (TFD), detection probability using occupancy models, and the number and clarity of weasel photos. We also repeated TFD and detection probability analyses for setups that were ≥ 20 m apart in case of overlap of effects. The average TFD across all sites was 43 days. Fall typically had the shortest TFD with beaver bait in fall achieving the best results. After accounting for occupancy, predicted detection probability across a 60-day survey was highest in fall with the best combination being salmon lure and beaver bait. The treatment type did not impact the average number of photos captured, but the clarity of photos was significantly positively related to use of bait and lure, type of lure, and specific combinations of bait and lure.

Intensive resource extraction activity in the oil sands of Canada alters the quantity, structure and distribution of native ecosystems, which in turn creates substantive challenges for conservation, land management and habitat reclamation. Progressive reclamation occurs on active mine sites in the Athabasca oil sands region (AOSR) of Canada, along with concurrent assessments of reclamation effectiveness. Yet, little is known about the ability of reclaimed habitats to mitigate for the short‐ and long‐term impacts associated with open‐pit mines and provide functional habitat for wildlife. We used a robust quasi‐experiment that combined camera trap data from an observational study with an innovative occupancy model to assess the effectiveness of upland reclamation to provide habitat for wildlife in the AOSR. The dynamic occupancy models were applied to 15 years of camera trap data to assess wildlife usage patterns of nine species of wildlife over seven types of habitats. The habitats sampled ranged from mining‐disturbed habitats reclaimed to upland forest ecosites common in the region, habitats disturbed by natural (fire) and human disturbances (clear‐cut logging), with comparisons to mostly intact mature forests. Our results indicate that the nine species of wildlife assessed used habitats in a manner consistent with expectations: some preferred disturbance‐dominated habitats while others used mature forest to a higher degree. We demonstrate that the application of dynamic occupancy models to camera trap data reliably discerned these trends, providing the means to predict wildlife usage patterns in an ever‐changing landscape, including one containing bitumen extraction as a contributor to landscape‐level modifications. Practical implication : Our work illustrates how continued monitoring of wildlife using camera traps contributes to assessments of reclamation effectiveness with respect to wildlife occurrence, distribution and usage patterns in anthropogenically and naturally disturbed landscapes. Evaluating the effectiveness of reclamation is especially important given the expected increase in habitat reclamation on active oil sands mines over the next several decades, coupled with the need to ensure that disturbed habitats regain ecological function able to sustain wildlife in the future.

Analysing DNA that organisms release into the environment (environmental DNA, or eDNA) has enormous potential for assessing rare and cryptic species. At present the method is only reliably used to assess the presence-absence of species in natural environments, as seasonal influences on eDNA in relation to presence, abundance, life stages and seasonal behaviours are poorly understood. A naturally colonised, replicated pond system was used to show how seasonal changes in eDNA were influenced by abundance of adults and larvae of great crested newts (Triturus cristatus). Peaks in eDNA were observed in early June when adult breeding was coming to an end, and between mid-July and mid-August corresponding to a peak in newt larval abundance. Changes in adult body condition associated with reproduction also influenced eDNA concentrations, as did temperature (but not rainfall or UV). eDNA concentration fell rapidly as larvae metamorphosed and left the ponds. eDNA concentration may therefore reflect relative abundance in different ponds, although environmental factors can affect the concentrations observed. Nevertheless, eDNA surveys may still represent an improvement over unadjusted counts which are widely used in population assessments but have unreliable relationships with population size.

Camera trapping is a standard tool in ecological research and wildlife conservation. Study designs, particularly for small-bodied or cryptic wildlife species often attempt to boost low detection probabilities by using non-random camera placement or baited cameras, which may bias data, or incorrectly estimate detection and occupancy. We investigated the ability of non-baited, multi-camera arrays to increase detection probabilities of wildlife. Study design components were evaluated for their influence on wildlife detectability by iteratively parsing an empirical dataset (1) by different sizes of camera arrays deployed (1–10 cameras), and (2) by total season length (1–365 days). Four species from our dataset that represented a range of body sizes and differing degrees of presumed detectability based on life history traits were investigated: white-tailed deer (Odocoileus virginianus), bobcat (Lynx rufus), raccoon (Procyon lotor), and Virginia opossum (Didelphis virginiana). For all species, increasing from a single camera to a multi-camera array significantly improved detection probability across the range of season lengths and number of study sites evaluated. The use of a two camera array increased survey detection an average of 80% (range 40–128%) from the detection probability of a single camera across the four species. Species that were detected infrequently benefited most from a multiple-camera array, where the addition of up to eight cameras produced significant increases in detectability. However, for species detected at high frequencies, single cameras produced a season-long (i.e, the length of time over which cameras are deployed and actively monitored) detectability greater than 0.75. These results highlight the need for researchers to be critical about camera trap study designs based on their intended target species, as detectability for each focal species responded differently to array size and season length. We suggest that researchers a priori identify target species for which inference will be made, and then design camera trapping studies around the most difficult to detect of those species.

Context Use of camera trap data in wildlife research is reliant on accurate classification of animals at the species, sex–age category or individual level. One such example is white-tailed deer (Odocoileus virginianus) camera surveys, which are often conducted to produce demographic estimates used by managers to establish harvest goals for a population. Previous research suggests that misclassification of deer by sex–age category (e.g. adult male, adult female, fawn) is common in these surveys, and represents a source of bias that could misinform important management decisions. Aim To examine whether training material has an effect on classification accuracy of white-tailed deer and explore other observer-based, experiential factors as they relate to classification accuracy. Methods We developed and tested the efficacy of species-specific training material designed to reduce sex–age misclassifications associated with white-tailed deer images. Key results Exposure to training material resulted in the greatest improvement in classification accuracy of deer images compared with any other respondent-based factors we investigated. Other factors, such as professional experience as a wildlife biologist, field experience viewing white-tailed deer and experience viewing deer images from camera traps, were positively associated with classification accuracy of deer images. Conclusions Our findings suggest that training material has the ability to reduce misclassifications, leading to more accurate demographic estimates for white-tailed deer populations. We also found that prior experience using camera traps and familiarity with target species was positively related to classification accuracy. Implications Species-specific training material would provide a valuable resource to wildlife managers tasked with classifying animals at the species, sex–age category or individual level.