Generalized spatial mark–resight models with an application to grizzly bears

AbstractThe high cost associated with capture–recapture studies presents a major challenge when monitoring and managing wildlife populations. Recently developed spatial mark–resight (SMR) models were proposed as a cost‐effective alternative because they only require a single marking event. However, existingSMRmodels ignore the marking process and make the tenuous assumption that marked and unmarked populations have the same encounter probabilities. This assumption will be violated in most situations because the marking process results in different spatial distributions of marked and unmarked animals.We developed a generalizedSMRmodel that includes sub‐models for the marking and resighting processes, thereby relaxing the assumption that marked and unmarked populations have the same spatial distributions and encounter probabilities.Our simulation study demonstrated that conventionalSMRmodels produce biased density estimates with low credible interval coverage (CIC) when marked and unmarked animals had differing spatial distributions. In contrast, generalizedSMRmodels produced unbiased density estimates with correct CIC in all scenarios.We applied ourSMRmodel to grizzly bear (Ursus arctos) data where the marking process occurred along a transportation route through Banff and Yoho National Parks, Canada. Twenty‐two grizzly bears were trapped, fitted with radiocollars and then detected along with unmarked bears on 214 remote cameras. Closed population density estimates (posterior median ± 1SD) averaged from 2012 to 2014 were much lower for conventionalSMRmodels (7.4 ± 1.0 bears per 1,000 km2) than for generalizedSMRmodels (12.4 ± 1.5). When compared to previousDNA‐based estimates, conventionalSMRestimates erroneously suggested a 51% decline in density. Conversely, generalizedSMRestimates were similar to previous estimates, indicating that the grizzly bear population was relatively stable.Synthesis and applications. Mark–resight studies often cost less than capture–recapture studies, but require that marked and unmarked animals have equal encounter rates. Generalized spatial mark–resight models relax this assumption by including sub‐models for both the marking and resighting processes. They produce unbiased density estimates even when marked and unmarked animals have differing spatial distributions and encounter rates. They thus provide a cost‐effective and widely applicable approach for estimating the density of wildlife populations.