Understanding how environmental variables affect spatial or temporal variation in species abundance is one of the main goals of ecological research. Indeed, accurately estimating presence or abundance of a species is usually the most important information required to evaluate the conservation status of a site or to assess the efficacy of management actions (Heink and Kowarik 2010). Analyzing count data without accounting for detection probability can lead to biased abundance and trend estimates (Royle and Nichols 2003; Kéry et al. 2005). To reduce the risk of bias, many monitoring programs now go beyond the use of observed counts as a proxy for true population size (Royle et al. 2004, 2005). Recently developed analytical approaches now enable the estimation of demographic parameters from unmarked individuals (Royle 2004; Dail and Madsen 2011). Such models use count data collected at a number of visits in a given season from a suite of sites, in order to follow temporal variations in population size. These methods show promise in ecology, wildlife management and conservation biology, especially when a limited number of individuals are captured at several sampling sites.
In this study, we examine the value of dynamic N-mixture models for understanding the population dynamics of the northern flying squirrel (Glaucomys sabrinus), which is of particular interest in North American forest management. The species has been considered an ecological indicator of mature and uncut forests, as well as of boreal forest ecosystem health (Smith 2007, 2012; Holloway and Smith 2011). According to recent studies, occupancy and abundance of northern flying squirrel populations are mostly explained by two key attributes of landscape composition: food and cavity availability. First, food resources may constitute a limiting factor for populations of G. sabrinus throughout its range (Ransome and Sullivan 2004; Lehmkuhl et al. 2006; Smith 2007). Conifer trees are known to provide a source of food through seeds and mycorrhizal fungi (Holloway and Malcolm 2006), the most common elements in the diet of G. sabrinus (Pyare and Longland 2002). As a result, abundance of this species is often related to the availability of conifer trees (Cotton and Parker 2000; Lehmkuhl et al. 2004; Holloway and Malcolm 2006). Second, tree cavities in the form of dens or nest sites are often found in large-diameter trees or snags of old forests (Holloway and Malcolm 2007a; Smith 2007; Pyare et al. 2010). These cavities constitute the most reliable predictors of microhabitat use and population density of northern flying squirrels in a wide range of habitat types (Holloway and Smith 2011; Smith 2012). However, recent studies using capture–mark–recapture (Lehmkuhl et al. 2006) and occupancy models (Trudeau et al. 2011) accounting for imperfect detectability suggest that highest northern flying squirrel population densities are not always linked to older stands, especially in mixed-wood forests.
Given this lack of consensus in the literature on the importance of mature stands and associated cavities, our main objectives were first, to evaluate the effect of cavity availability on population dynamics of northern flying squirrels through a before-after control-impact (BACI) design consisting of experimental supplementation of cavities between two sampling seasons, and second, to test the application of a dynamic N-mixture model in a BACI design. We hypothesized that (1) initial squirrel abundance increases with conifer basal area (indirect measure of food availability – surrogate of seeds and mycorrhizal fungi) and snag basal area (indirect measure of natural cavity availability) and (2) recruitment rate and apparent survival increase with the addition of artificial cavities, particularly where natural tree cavities and food availability are low (interactive effects of nest box addition x snag basal area, and nest box addition x conifer basal area). Finally, to assess the robustness of our results, we compared the estimates obtained from the dynamic N-mixture models against single season N-mixture models, classic capture–mark–recapture models for closed populations, generalized linear mixed models on unadjusted counts, and Cormack–Jolly–Seber (CJS) models.