Predicting Bobcat Distribution and Density across Central Wisconsin

Abstract

Though knowing abundance is fundamental for species management, carnivores are inherently difficult to census. Their population parameters have been historically inferred indirectly using telemetry information or detection-based indices with little statistical support. Advances in sampling techniques and analysis over the last 20 years have revolutionized carnivore abundance estimates. Noninvasive techniques (e.g. cameras, hair-traps, scat surveys) have expedited sampling procedures and the marking of individuals for population assessment. Acknowledgment of imperfect detection has strengthened analyses designed to determine species distribution, and modeling variation in detection has improved abundance and density estimation. However, surveying for carnivore species remains a burgeoning practice and improving efficiency by increasing sample size or the spatial domain of interest is crucial for advancing carnivore science. The underlying purpose of this project was to estimate bobcat population size over a roughly 15,000 km2 region in central Wisconsin. Given constraints of time, money, and the domain of inference, the greater theme of this project was to test methodological and analytical ways to improve carnivore survey efficiency. Chapter 1 focuses on comparing two noninvasive detection methods (remote cameras and scat-detecting dogs) as means to detect bobcat and estimate population size. We paired these methods at 4 independent sites encompassing roughly 64 km2 each with 16 sub-divided 4 km2 units subject to camera and dog effort. We compared detection dogs and cameras based on detection and “capture” (individually assigned detections) totals, density estimates derived from capture-recapture techniques, associated detection probabilities, and costs. Both techniques were viable for density estimation (~3 bobcat per 100 km2), and estimated that a single detection dog survey day accrued as much data as roughly 4 weeks with a single camera. However, camera surveys cost 60% less, provided roughly 100% greater raw data return, and provided a smaller ratio of cost to cumulative detection probability when stations were active for more than 4 weeks. These results differed from previous comparative studies performed within shorter temporal sampling windows that showed detection dogs were more efficient than cameras. We noted that study objectives and design were the primary consideration for interpreting methodological comparisons and choosing survey methods. The second chapter tested the hypothesis that predicted occurrence can in turn predict population density, and tried to elucidate how variation in individual space use might confound this relationship. We undertook surveys for bobcat at 9 remote camera arrays across central Wisconsin, and used site occupancy methods to develop a population-level (design 1) resource selection probability function. Concurrently, we identified individual animals and estimated density at each site using spatially explicit capture-recapture models. We assessed the relationship between predicted use and estimated density at each site, and compared the influence of primary resource selection covariates on abundance and detection probability using capture-recapture mixed models. We then used spatially explicit models again to assess how landscape and trail-specific attributes influenced individual detection rates at specific camera stations and how landscape attributes influenced density in order to predict population size throughout the study area. Bobcat use was best predicted by the proportion of woody cover and the configuration of wetland patches surrounding each site, and we retained these covariates for subsequent detection, abundance, and density models. Validation using independent data suggested adequate (75%) discriminative power. Site-specific detection probability was most influenced by local trail characteristics, particularly the substrate, the level of human activity, and the width of the trail. We observed a moderate fit between predicted habitat suitability and density using null spatial capture-recapture models. At the array level, landscape characteristics were much more influential on population size than individual probability of detection. The relationship between individual detection probability and camera specific covariates suggested that gravel substrates and paths with high surrounding obstruction (i.e., water features) increased individual detection rates. Increases in woody cover and wetland edge density both increased the distance at which an animal could be detected from its activity center, but decreased an individual’s site specific detection probability. Our top capture-recapture model included density covariates and both trap-specific woody cover and gravel substrates used as detection covariates. When we retroactively used this model to compare predicted use and density at our individual arrays, we found a much stronger fit between the parameters. We concluded that bobcat detection was driven more by population density than individual space use, the relationship between population level use and population size remains imperfect. Woody cover and wetland edge density had differential influences on overall site use and population density, and we suggest that these differences reflect differential influences on individual space use. We suggest individual detection probability reflected functional resource selection response, and that detection probability itself is related to resource selection over a hierarchy of spatial scales from characteristics of the home range to extremely local characteristics. We estimated total bobcat population size within central Wisconsin as roughly 362 (95% CI 250-490) individuals.