A spatial, closed integrated population model to estimate wildlife population size and structure

Abstract

AbstractNoninvasive data collection methods are increasingly used to monitor endangered, cryptic, or otherwise difficult to observe wildlife populations. Spatial capture‐recapture and count‐based statistical methods are often used to model these data to derive population estimates, each with their own advantages. The goal of our study was to integrate these 2 data sources and model structures to maximize the use of single‐season surveys in which capture‐recapture and count‐based data are collected. We formulated a Bayesian spatial, closed integrated population model (cIPM) that leverages scat DNA‐based spatial capture‐recapture and camera‐based spatial count sub‐models to derive sex and age class‐specific population estimates. Our cIPM has the advantage of sharing the spatial scaling (σ) parameter directly between sub‐models and drawing proposed individuals from the same data‐augmented superpopulation for each sub‐model. This allows for separate estimation of sex and age‐class groups simultaneously without any assumptions of population structure or conversions between model output types. We applied our cIPM to a wild population of Roosevelt elk (Cervus canadensis roosevelti, Karuk: íshyuux) during winter 2018–2019 in their winter range within Karuk Ancestral Territory in northern California, USA. We also conducted a simulation study to test our cIPM on a known population consisting of 2 age classes and 2 sexes. The cIPM was able to estimate most population, detection, and σ parameters with good accuracy, including population size. The assignment of sex and age was biased, however, resulting in fewer juvenile males, more juvenile females, and more adult males. Our cIPM applied to the data estimated approximately 310 (95% CI = 235–411) female adults, 90 (95% CI = 57–134) male adults, 75 (95% CI = 43–149) female calves, and 23 (95% CI = 12–47) male calves. These estimates were different from those that would have been made had we relied on common assumptions such as equal calf sex ratios, though this ratio was likely biased towards females based on our simulation results. Our methods are widely applicable and maximize the amount of information available to managers and researchers from any single‐season survey in which separate data sources capturing sex and age‐class information are obtained.