Biometrics and Modeling I

Contributed Oral

Spatial Optimization, Statistical Criteria, and Prediction Accuracy of Random Forest Models for Variation in Avian Abundance
Kimberly Serno, Kevin Gutzwiller
Predictive modeling of species’ abundances is important in conservation because the ability to accurately predict them can improve planning and management effectiveness. One way to improve predictive accuracy in multiscale models is through spatial optimization, which involves measuring environmental variables at various spatial extents and identifying the extent at which each variable is most associated with a response variable. In terms of a model’s prediction accuracy, some statistical criteria for implementing the optimization may be more effective than others. We compared the prediction accuracy of random forest models developed from spatial optimization based on Spearman’s correlation coefficient (SC) and out-of-bag prediction error for univariate random forest modeling (URFM). The response variable was the coefficient of variation (CV) in abundance of Brewer’s Sparrow (Spizella breweri), a declining sagebrush obligate. Predictor variables included two drought indices and the proportions of landcover for nineteen vegetation types, all of which were computed for six biologically relevant extents (400-m, 1-km, 4-km, 5-km, 8-km, and 20-km buffers). For each predictor, we identified the extent whose data were most correlated with (SC) or led to the smallest out-of-bag prediction error for (URFM) the response variable (CV range: 0-223%) during 2009-2013. The mean absolute error (and error relative to range) of prediction for the model developed from the SC-based optimization was 26.3% (11.8%), whereas that for the model developed from the URFM-based optimization was 26.0% (11.7%). URFM-based optimization did not yield appreciably better prediction accuracy, suggesting that researchers should not assume that spatial optimizations based on prediction error will necessarily result in higher prediction accuracy than those derived from simple measures of statistical association. Different statistical criteria may yield more disparate results than what we observed, and we recommend that analysts conduct pilot analyses to identify the criterion that is most effective for their data and model type.
Population Abundance and Survival Rate Estimates of Elk in the Cumberland Mountains
Katherine Kurth, Eryn Watson, Dailee Metts, Brad Miller, Rick Gerhold, Dana Morin, Sheng-I Yang, Lisa Muller
Elk (Cervus canadensis) were reintroduced to the North Cumberland Wildlife Management Area (NCWMA) in East Tennessee between 2000 and 2008, following extirpation in the 1800s. Population estimates of NCWMA elk are below initial projections, warranting increased research. The primary objectives of this study are to estimate elk population abundance, annual survival rates, and causes of death. To estimate abundance and assess survival of the NCWMA elk, we are using data obtained from genetically identified fecal samples and GPS collared elk, respectively. We collected fecal samples (n=357) using a clustered sampling design across 65 designated collection areas composed primarily of wildlife openings. Collection took place weekly within the NCWMA from February to May 2019. Fecal samples were analyzed by Wildlife Genetics International (Nelson, British Columbia, Canada) using 15 microsatellites with sex determination. From 157 successfully genotyped fecal samples, 85 individuals (21 males and 64 females) were identified. We will estimate population abundance using data from the genetically identified elk and spatial capture-recapture models within the secr package in R. In addition to non-invasive sampling, we collared 29 elk (8 males and 21 females) in 2019 and 2020. To date, 7 collared elk mortalities comprised of 2 necropsy confirmed meningeal worm (Parelaphostrongylus tenuis) associated disease cases, 1 poaching incident, 1 vehicle collision, 1 legal hunter harvest, and 2 unknown causes of death due to carcass degradation have been documented. We will use a known-fate survival model using the R package RMark to access program MARK to evaluate yearly survival rates and the associations of sex, season, and year with survival. Evaluating population abundance, survival rates, and causes of mortality will aid in identifying data-based management techniques for supporting population growth and sustainability of NCWMA elk.
Population Dynamics of Brown Bears along Brooks River in Katmai National Park, Alaska
David Williams, Leslie Skora
Measuring wildlife population demographics can be costly, challenging, and invasive. These issues create a missional challenge in Katmai National Park in southwest Alaska, USA where the remote location, harsh weather, and limits to staff and funding make it difficult to monitor one of the largest populations of brown bears (Ursus arctos). Concerns over dramatic changes in the number of bears seen using Brooks River have led to questions about whether the population has changed. Fortunately, individual bears congregating on salmon spawning streams can be identified by unique physical and behavioral features, which has allowed creation of a standardized, long-term, non-invasive record of individual brown bears using Brooks River over time. Our objectives were to use the bear monitoring dataset to measure changes in the bear population at Brooks River, specifically changes in survival, abundance, and productivity related to sockeye salmon (Oncorhynchus nerka) escapement. We used bear identification records from 2000–2018 in a mark-recapture framework to estimate age-sex specific survival (cubs (0.718), subadult males (0.840), subadult females (0.909), adult males (0.876), and adult females (0.891)). We found no changes in survival or relationship between survival and salmon escapement. We used the bear monitoring data as a time series of counts in a state-space model to estimate abundance and productivity from 2000–2019. Bear abundance (mean: 106) was positively related to sockeye salmon escapement in previous years. However, only some age-sex groups exhibited this relationship. Abundance of breeding females and their cubs was positively related to salmon escapement from the same year, which may indicate they were avoiding Brooks River during years when escapement was low and competition among bears was high. The relationship between annual escapement and productivity measures were not significant, although there were signs that a change in productivity may have occurred.
Predicting How Climate Change Threatens the Prey Base of Arctic Marine Predators
Katie Florko
The rapid rates of climate change in the Arctic are affecting the ecology of endemic marine mammals, particularly ice-obligate species (e.g., ringed seals that reproduce only on sea ice) that are vulnerable to sea ice loss. Climate-related effects on Arctic ecosystems reach beyond direct declines in sea ice, but research on Arctic marine mammals often focuses on linking sea ice declines to demographic declines. We assess these potential changes by modelling the prey base of a widely distributed Arctic predator (ringed seal; Pusa hispida) in a sentinel area for change (Hudson Bay) under high- and low- greenhouse gas emissions scenarios from 1950 to 2100. While the predictions of the two scenarios bifurcated at ~2025, the qualitative results are similar. Overall, we forecasted a decrease in the biomass and abundance of energy-rich Arctic cod and an increase in smaller forage fish (e.g., capelin and sand lance) with projected declines in the body size of all eight fish species. Spatially, the well-distributed Arctic cod was projected to decrease throughout Hudson Bay, and capelin and sand lance were projected to increase in the southern regions. Such shifts in fish biomass, abundance, and distribution and decline in body size will likely affect ringed seal foraging through diverse mechanisms. Overall, we predict an ecosystem cascade in Hudson Bay from a moderate Arctic to subarctic to a more temperate ecosystem with significant changes in forage fish distribution and size that will have reverberating impacts throughout the ecosystem.
Does IUCN Status Capture the Variation in Population Density Across Species? a Test on Wild Felids
Stefano Anile, Clayton Nielsen, sebastien devillard
The biodiversity crisis is intensifying worldwide mainly due to global climate change, habitat destruction and fragmentation, and overexploitation of wild species. The IUCN red list states that about 25% of animal species were at least Near Threatened in 2020 with the Felidae comprising an alarming 60% of species listed in this category. Population density plays a key role in conservation biology and is often used to address the effectiveness of conservation actions. However, IUCN conservation status does not rely on population-specific parameters. Here, we addressed the relationship between IUCN conservation status and population density estimates corrected by body size (i.e. taking into account the density-mass allometry law) using a comprehensive dataset of 449 population density estimates for 27 felids species worldwide. We hypothesized felid species with the worst IUCN conservation status had lower relative population densities than species with better conservation status. Contrary to our expectations, Vulnerable and Endangered species did not have lower relative population densities than Least Concern or Near Threatened species. We conclude IUCN conservation status does not capture the real demographic state of wild populations of felids. Using relative population density appears to be a promising way to establish conservation priorities in felids when increasing population density is the main target.
Emigration Effects on Estimates of Age- and Sex-Specific Survival of Small Mammals
Matt Weldy, Damon Lesmeister, Clinton Epps
When estimating survival using Cormack–Jolly–Seber (CJS) models and capture–recapture data, emigration is typically assumed to have a negligible effect on estimates such that apparent survival is indistinguishable from true survival. Consequently, for populations or age classes with high dispersal rates, apparent survival estimates are often biased low and temporal patterns in survival might be masked when site fidelity varies temporally. We used 9 years of annual mark-recapture data to estimate age-, sex-, and time-specific apparent survival of Humboldt’s flying squirrels (Glaucomys oregonensis) and Townsend’s chipmunks (Neotamias townsendii). For Humboldt’s flying squirrels, these estimates support a small body of research investigating potential variation of survival among age and sex classes, but age- and sex-specific survival has not been evaluated for Townsend’s chipmunks. We also quantified the effects of age- and sex-specific emigration on confounded estimates of apparent survival. Our estimates of juvenile flying squirrel survival were high relative to other small mammal species and estimates for both species were variable among years. We found survival differed moderately among age and sex classes for Humboldt’s flying squirrels, but little among age and sex classes for Townsend’s chipmunks, and that the degree of emigration associated confounding on apparent survival estimates varied substantially among years. Without correcting for emigration, apparent survival estimates were lower and obscured temporal variation, particularly for male Humboldt’s flying squirrels and female Townsend’s chipmunks. Our results demonstrate how emigration can influence commonly used estimates of apparent survival obtained using CJS models. Unadjusted estimates of apparent survival, confounded the interpretation of differences in survival between age and sex classes and masked potential temporal patterns in survival because the magnitude of adjustment varied among years. We conclude that apparent survival estimators are robust during some time periods; however, when emigration rates vary in time the effects of emigration should be carefully considered.
WildAgg: An R Package for Streamlining Wildlife Aggregation Analyses
William Janousek, Tabitha Graves
The package `wildagg` is an R package designed to estimate, summarize, and visualize wildlife aggregation metrics using location information like GPS collar data. The development of this package began with two research efforts studying the aggregation and density of elk on the National Elk Refuge, WY. We applied lessons learned to create a straightforward implementation for users that have limited knowledge of program R and related analyses. The package has three primary functions. The first is to calculate daily inter-animal distances for a population of collared individuals, second to estimate the dynamic interaction between pairs of animals based on the proportion of time spent per day within some distance buffer, and third to calculate kernel density estimates across temporal scales. All three of these metrics are useful in determining degrees of animal aggregation and provide a variety of avenues to derive potential mechanisms explaining observed aggregation patterns. The framework we present supports the evaluation of temporally varying management actions that influence aggregation broadly and can be easily implemented to answer questions about disease transmission, human-wildlife conflict, and inter-species competition.
The Spatial Scaling and Individuality of Habitat Selection in a Widespread Ungulate
David Heit, Joshua J. Millspaugh, JON MCROBERTS, Remington Moll, Barbara Keller
Spatial scale is a fundamental concept to animal ecology. As spatial heterogeneity leads many ecological processes to be highly scale-dependent, the selection of the spatial scale at which to study animal ecology is highly important. Improper choices of scale can lead researchers to come to one conclusion at one scale where it would be different or even reversed at another scale. Further, the effects of social, environmental, and anthropogenic conditions on animal behavior and population dynamics manifest at different scales. Despite this, studies of animal ecology are often conducted at only one or a few scales, whether expressly chosen by the researcher, or implicitly determined by the scale of the associated data. Although the importance of scale is widely recognized in ecology, empirical tests of how ecological processes respond to scale remain scarce. An additional and often overlooked issue related to the scaling of animal-habitat relationships is how scale of effect is influenced by intrinsic and extrinsic factors that vary across individual animals. Here, we investigated the degree to which several key intrinsic and extrinsic factors could explain the spatial scaling of habitat selection in a widespread ungulate. To test our hypotheses and their associated predictions for scale of effect, we constructed integrated step-selection functions (iSSFs) at a wide range of spatial scales and fit these functions to telemetry data from 664 GPS-collared white-tailed deer in two different ecosystems in Missouri, USA. Our findings connect emerging theory regarding spatial scaling to quantitative analysis of animal-habitat relationships and help inform the management and conservation of wildlife and the resources that support them.
A Semi-Spatial Integrated Population Model to Assess Population Dynamics of a Recovering Species
Lisanne Petracca, Sarah Converse, Beth Gardner, Jeff Manning, Dan Thornton, Lisa Shipley, Sarah Bassing, Jason Ransom, Robert Long, Ben Maletzke
Recolonizing species exhibit unique population dynamics, namely dispersal to and colonization of new areas, that are important to understand from a conservation and management perspective. Integrated population models (IPMs) have proven useful for making inference about population dynamics by integrating multiple data streams, including data relevant to population state and demographic rates. More recently, spatially explicit integrated population models (SIPMs) have leveraged the power of spatial capture-recapture, resulting in a spatially explicit model of population dynamics. SIPMs, however, require information on the spatial observation process to correctly model spatially explicit data. In a recolonizing population of wolves in Washington, USA, we were lacking data on the spatial observation process but wanted to leverage the power of SIPMs to describe the recolonization process, which is critical to recovery. We used GPS collar, camera trap, and count data collected from this population to develop a semi-spatial integrated population model, in which non-spatial data on survival and reproduction are integrated into a semi-spatial model comprising [1] territory size estimated from telemetry data, [2] territory-specific count data, [3] age-specific probabilities of dispersal estimated from telemetry data, [4] least-cost movement paths between territories of origin and potential new wolf territories (estimated from telemetry data and a second-order resource selection function [RSF]), and [5] a Bernoulli process by which a wolf can remain in a potential territory based on an underlying occupancy model (estimated using camera trap data). Our semi-spatial IPM can be used to assess population dynamics with a spatial component and determine how management strategies can affect population dynamics and recovery.

Contributed Oral
Location: Virtual Date: November 2, 2021 Time: 4:00 pm - 5:00 pm