Effect of subsampling genotyped hair samples on model averaging to estimate black bear population abundance and density

DNA-based capture-mark-recapture techniques are commonly used to monitor wildlife populations. Analyzing all collected samples can be cost prohibitive for studies of highdensity populations; therefore, subsampling is frequently used to offset genetic analysis costs yet obtain reliable population abundance estimates. Because model selection and parameter estimation depend on sample size, choosing an appropriate subsampling procedure is a critical part of study design. Monitoring high-density populations at large scales can be logistically challenging and may require estimating population density at small scales and extrapolating to larger areas. Density estimates must be precise and robust to closure violations common to small-scale studies. I used DNA-based capture data for an American black bear (Ursus americanus) population in Great Smoky Mountains National Park, Tennessee to investigate the effects of subsampling on model selection by incrementally reducing the number of samples selected for DNA analysis, and subsequently comparing model selection results and model-averaged estimates to results based on the full dataset. I also evaluated a spatially explicit mark-recapture method for estimating density for a study area located in contiguous black bear habitat. I assessed population closure and compared density estimates from a conventional abundance conversion method with estimates from the spatially explicit method. My results indicated high subsampling intensities (e.g., 1 sample/site/week) can achieve adequate capture probabilities and reliable population abundance estimates (i.e., CV 20%) given the sample site density and number of sampling periods in this study. However, capture probabilities associated with lower subsampling intensities were vi inadequate for reliable model selection and produced substantially biased estimates of abundance. Population closure was violated in my study and likely caused positively biased density estimates compared with estimates obtained from the spatially explicit method. Based on the full dataset, reliable and cost-effective density estimates needed to monitor populations at larger scales are possible using spatially explicit methods to estimate population density.