Biometrics & Modeling II

Contributed Oral

Habitat Use by Band-Tailed Pigeons Across Multiple Spatio-Temporal Scales
Beth Ross, Daniel Collins, Matthew Boggie, Christopher Coxen, Scott Carleton
While understanding the spatial scale at which wildlife responds to environmental conditions and management is important for future conservation work, traditional statistical approaches to quantify resource use across multiple spatial scales are often complex and may result in multi-collinearity in models or intractable results. Alternatively, Bayesian variable selection selects the most relevant spatial or temporal scale for different covariates in one model. We used variable selection to identify relevant spatio-temporal scales of habitat use of Band-tailed Pigeons during the breeding season in New Mexico, USA. We used the number of locations of Argos-tagged Band-tailed Pigeons in a 500-m grid cell as a response variable. We then calculated environmental covariates at different spatial scales (within a 500-m grid cell and within a 1-km and 2-km buffer) and climatic covariates at different temporal scales (fall and winter preceding the breeding season [Oct – April], during the breeding season [April – August], and the full “Pigeon year” [Oct – August]). We then used Bayesian Latent Indicator Scale Selection to determine the appropriate spatial (environmental) and temporal (climate) scale for inference. Our results indicated that Band-tailed Pigeons selected sites at a broad spatial scale (2 km) for elevation and mixed forest cover, while the 1-km scale was more relevant for topographic wetness index and topographic position index, and the 500 m scale for aspect. Band-tailed Pigeons increased use in response to mixed forest cover, preferred high elevation sites, and used sites with moderate topographic wetness index. The most relevant temporal scale for habitat selection for precipitation and PDSI was the breeding season, with Band-tailed Pigeons using sites with greater precipitation and lower PDSI values. In addition to providing insights into habitat use by Band-tailed Pigeons, our model illustrates how variable selection can be used to appropriately choose spatial and temporal scales for inference.
Integrating Counts from Helicopter and Airplane Surveys to Estimate Densities of Waterfowl
Christian Roy, Amelia Cox, Scott Gilliland, Eric Reed
Aerial surveys conducted from fixed-wing aircraft typically underestimate waterfowl densities because of availability and detection biases, and provides only genus level estimates for species which observers are unable to identify. In contrast, surveys conducted from helicopters typically result in higher detection due to the platforms’ reduced speed and maneuverability, and their ability to hover along with the use of image stabilized binoculars allows observers to identify the species of most observations. Fixed-wing aircraft, however, have much  greater range and are more economical to run than helicopters, hence large-scale waterfowl surveys usually rely heavily on fixed-wings.  Where possible, ground counts are used to correct abundance estimates from fixed-wing surveys for errors resulting from detection and identification error; however, ground counts are impracticable in remote road less areas.  We developed a new modeling approach that combines the strengths of fixed-wing and helicopters to improve accuracy of abundance estimates for sea ducks and scaup within the boreal forest. We applied this new approach, which uses a hierarchical Bayesian model to combine helicopter and fixed-wing aircraft data in a single model, on waterfowl survey data collected during a 2009 experimental sea duck and scaup survey in central Labrador, Newfoundland and Labrador, Canada. Our model provided estimates of waterfowl abundance at the species level while adjusting for detection probability in both platforms. This allowed us to derive an availability correction factor for the fixed-wing aircraft portion of the survey and species specific estimates of abundance for Surf, Black and White-wing Scoters as well as Lesser and Greater Scaup.  We describe how our model could be extended to other waterfowl monitoring surveys.
An Integrated Population Modeling Approach to Estimate Population Demography and Total Allowable Harvest of Polar Bears
Andrew Derocher, David Koons, Alastair Franke, Markus Dyck, Kylee Dunham
Unlike other polar bear populations in Nunavut (e.g., Western Hudson Bay population), detailed capture-mark-recapture data for the Davis Strait population cannot be collected annually because of prohibitive costs. Capture-recovery data are collected every year, but releases of newly marked individuals only occur in years when capture-recapture studies are conducted, both of which help inform age structure and offspring survival. However, harvest data are collected annually as part of regular harvest monitoring and the age structure of the harvest. This is an ideal situation for integrating available data to estimate demographic parameters, harvest rates, age and sex structure, abundance, and realized population growth rates on an annual basis while accounting for imperfect detection and statistical uncertainty. Integrated population models (IPM) combine population count data and demographic data to make inferences about population dynamics for wildlife management and conservation. Integrated models have several advantages compared to typical approaches that analyze data sets individually: 1) IPMs permit estimation of a greater number of demographic parameters; 2) parameter estimates have greater precision, which increases statistical power, and inference; 3) uncertainty inherent to the modeling process can accounted for explicitly, and; 4) IPMs explicitly link variation in population size and demographic rates via a demographic model. To provide better estimates of demographic parameters and inform sustainable total allowable harvest levels, we use an integrated population model to describe the Davis Strait polar bear subpopulation population dynamics. Using multiple data sets collected from 2005 through 2018, we use capture-mark-recapture data, capture recovery data, harvest data, age structure data, habitat data (annual date of ice retreat and advance), and prey data (annual abundance of harp seal pups) to implement a birth-death balance equation that describes population dynamics.
Estimating Density of Wildlife with Camera Traps: Spatially Explicit Capture-Recapture and Time in Front of the Camera
Camille Warbington, Mark Boyce
In recent years, camera traps have become ubiquitous for estimating density of wildlife populations.  Researchers frequently use spatially explicit capture-recapture (SECR) models of density, whether in a maximum likelihood (ML) or Bayesian framework.  However, not all wildlife populations “fit” SECR assumptions and model requirements.  Time in front of the camera (TIFC) is another method for density estimation that does not have the same limitations and assumptions as SECR.  While SECR density estimates are widely published, we are unaware of any that compare SECR results to TIFC estimates.  Sitatunga, a wetland specialist African antelope, has ecological requirements and behaviour that do not align with SECR assumptions.  We present results on density estimates from ML and Bayesian SECR models, as well as TIFC methods.  Our results show that sitatunga density estimates from TIFC are comparable to SECR estimates, without the same model limitations.  We suggest TIFC as an alternative to SECR when the species in question violates assumptions or is otherwise unsuited to density estimation with SECR.
Choosing an Appropriate Artificial Intelligence Platform and Workflow for Processing Camera Trap Data
Juliana Velez, John Fieberg
Camera traps have transformed how ecologists study the distribution of wildlife species, their activity patterns, and interactions among members of the same ecological community. Although they provide a cost-effective method for monitoring multiple species over large spatial and temporal scales, the time required to process the data can be a limiting factor. Thus, there has been considerable attention given to developing artificial intelligence (AI) tools, such as machine learning (ML), to efficiently detect objects and classify species in camera trap photos. Using ML involves training a model to use particular features in an image to identify and assign species labels to the animals present. To reduce the technical challenges associated with training ML models, several research communities have recently developed systems that incorporate ML in easy-to-use interfaces with the aim of making these tools accessible to a wider audience of camera trap users. To help new users choose an appropriate system, we review key characteristics of popular AI systems, including their software and programming requirements, data management tools, and AI features. We distinguish between fully- and semi-automated workflows and discuss their potential use for projects of varying duration and depending on data processing requirements (e.g., near-real time detection, annotation of additional features such as behaviour, sex, age, etc.). The data used to train ML models will vary by system, and  the performance of AI systems will depend heavily on the similarity between training and new (“test”) data (e.g., in terms of species present and background features in the photos). Thus, we also provide R code and data from our own work to demonstrate how users can evaluate model performance using common performance metrics (e.g., precision, recall and F1 score). By sharing these examples via an open-source gitbook, we hope to encourage ecologists to use AI to process their camera trap data.
Modeling Effects of Severe Winter Weather on Survival and Band Recoveries of American Black Duck
Robert Emmet, Beth Gardner, Andy Royle, Patrick Devers
Migratory birds experience a variety of pressures throughout their annual cycle. Spatial and temporal variation in risks and resources can result in differences in demographic rates; failing to account for these differences can lead to biased demographic rate estimates and an incorrect understanding of drivers of population dynamics. Band recovery models used to estimate survival often account for spatial or temporal variation in survival and recovery probabilities, but rarely account for both. We developed a novel approach to estimating spatially explicit seasonal survival and recovery probabilities from band recovery data using efficient Bayesian methods. We used this band recovery model to investigate potential relationships between severe winter weather and non-hunting-season survival and recovery probabilities in American black ducks, a harvested species of international conservation concern. We found that severe winter weather was not a significant predictor of non-hunting-season survival, but that prolonged or extreme cold temperatures were associated with an increase in recovery probability, indicating a possible carryover effect of severe winter weather. Our novel band recovery modeling framework, built using the R-INLA software package, offers major computational advantages over existing spatially explicit band recovery models and allows researchers to use fine-scale spatial information from banding locations to test ecological hypotheses concerning drivers of survival and harvest vulnerability.
Diffusion Modeling Reveals Effects of Multiple Release Sites and Human Activity on a Recolonizing Apex Predator
Joe Eisaguirre, Perry Williams, Xinyi Lu, Michelle Kissling, William Beatty, George Esslinger, Jamie Womble, Mevin Hooten
Reintroducing predators is a promising conservation tool to help remedy human-caused ecosystem changes.  However, the growth and spread of a reintroduced population is a spatiotemporal process that is driven by a suite of factors, such as habitat change, human activity, and prey availability.  Sea otters (Enhydra lutris) are apex predators of nearshore marine ecosystems that had declined nearly to extinction across much of their range by the early 20th century.  In Southeast Alaska, which is comprised of a diverse matrix of nearshore habitat and managed areas, reintroduction of 413 individuals in the late 1960s initiated the growth and spread of a population that now exceeds 25,000.  Periodic aerial surveys in the region provide a time series of spatially-explicit data to investigate factors influencing this successful and ongoing recovery.  We integrated an ecological diffusion model that accounted for spatially-variable motility and density-dependent population growth, as well as multiple population epicenters, into a Bayesian hierarchical framework to help understand the factors influencing the success of this recovery.  Our results indicated that sea otters exhibited higher residence time as well as greater equilibrium abundance in Glacier Bay, a protected area, and in areas where there is limited or no commercial fishing.  Asymptotic spread rates suggested sea otters colonized Southeast Alaska at rates of 1-8 km/yr with lower rates occurring in areas correlated with higher residence time, which primarily included areas near shore and closed to commercial fishing.  Further, we found that the intrinsic growth rate of sea otters may be higher than previous estimates suggested.  This study shows how predator recolonization can occur from multiple population epicenters.  Additionally, our results suggest spatial heterogeneity in the physical environment as well as human activity and management can influence recolonization processes, both in terms of movement (or motility) and density dependence.
Time-Dependent Memory and Individual Variation in a Peripheral Population of Arctic Carnivores
Peter Thompson, Mark Lewis, Mark Edwards, Andrew Derocher
Animal movement modelling provides unique insight about the way animals perceive their landscape and how this perception may influence their space use patterns. When coupled with data describing an animal’s environment, ecologists can fit statistical models to animal movement data to describe how spatial memory informs animal movement. We performed such an analysis on a population of grizzly bears (Ursus arctos) in the Canadian Arctic using an animal movement model that includes time-dependent spatial memory patterns. Grizzly bear populations in the Arctic lie on the periphery of the species’ range, and as a result endure harsh environmental conditions. The spatial and temporal structure of their environment makes the population an interesting case study for memory-informed movement. The model we fit tests four alternate hypotheses (some incorporating memory; some not) against each other, and we found a high degree of individual variation in the grizzly bear population in the use of memory. 11 of the 21 bears appeared to use complex, time-dependent spatial memory to inform their movement decisions, doing so in different ways. These results, coupled with existing knowledge on individual variation in the population, highlight the diversity of foraging strategies for Arctic grizzly bears while also displaying the inference that can be drawn from this innovative movement model.
Harvest and Density-Dependent Predation Drive Moose Population Decline in Ontario, Canada
Robby Marrotte, Brent Patterson, Joseph M. Northrup
The relative effect of top-down versus bottom-up forces in regulating and limiting ungulate populations is an important theme in ecology and wildlife management. Untangling these effects is important for basic understanding of trophic dynamics and is critical for effective management. Here, we assessed the interactive effects of intraspecific competition, predation, harvest and disease on moose (Alces alces) within an area close to 1 million km2 over 20 years. We integrated two independent sources of observations within a hierarchical Bayesian population model to compare the relative influence of factors driving moose abundance across 55 replicated populations between 1999-2018 in Ontario, Canada. Across populations, moose declined by nearly 20% over the entire period examined. At high density, moose were regulated by intraspecific competition. Predation primarily limited population growth, except at low density, where it was regulating. Harvest was largely additive and potentially contributed to population decline. Our results provide strong evidence for density dependent predation, highlighting that population dynamics are context dependent and vary strongly across gradients in climate, forest type and predator abundance. Consequently, management of moose would be optimized by taking different strategies across populations in accordance with their population trajectories and abundance of predators.

Contributed Oral
Location: Virtual Date: November 3, 2021 Time: 12:00 pm - 1:00 pm