Introduction
Wildlife management requires knowledge of population parameters such as density, sex ratio, and productivity. Population densities are often dependent on the quality of occupied habitat as animals rarely use all available habitats equally (Fretwell 1969, Maier et al. 2005, Bjørneraas et al. 2012). Thus, habitat selection of animals is an important part of management and conservation of many species (Allen and Singh 2016). When landscapes are heterogeneous and animals preferentially use certain habitats, the study of habitat selection is necessary in order to understand how habitat is linked to population abundance (Royle et al. 2018). In particular, by assuming a homogeneous landscape or ignoring the effect of landscape heterogeneity on populations, estimates of density can be biased (Royle et al. 2013a).
The preference of animals for certain habitats, i.e. the habitat (or resource) selection can be viewed as a process with three orders (Johnson 1980). The first order is the spatial distribution of the species, the topic often studied in landscape ecology. Second and the third order habitat selection is more of interest in population ecology. Second order habitat selection describes how individuals are distributed within the species’ range in relation to environmental features. Third order selection describes within home range selection of habitats by individuals.
Commonly, habitat selection has been studied invasively using telemetry e.g. by attaching GPS or VHF collars on animals (Morris et al. 2016, Bose et al. 2018). Live capturing a large number of individuals, especially of large species, can be not only harmful for animals but also expensive and impractical. Therefore, even though information on space usage by telemetry can be detailed, it often represents only a few individuals that may not be representative of the population. Telemetry data thus represents individual-level rather than population-level habitat selection. The use of non-invasive genetic methods provides the possibility to study animal space use without physically marking and recapturing them, for instance by collecting feces (Granroth-Wilding et al. 2017, Hagemann et al. 2018) or hair (Sun et al. 2017, O’Meara et al. 2018) left in specific trap or collection devices in the environment (Waits and Paetkau 2005). Spatial information of the individuals is recorded from the capture locations and individual identification is obtained by extracting DNA from the sample and genotyping the samples. Even though sampling in the field and analyzing non-invasive DNA in the laboratory can be laborious and expensive, the resulting genotype data can provide valuable population-level information of space use.
Spatially explicit records of individuals can be analyzed using spatial capture-recapture (SCR), a spatial extension of long-established capture-recapture methods (Efford 2004, Royle et al. 2014). Apart from inferring density, SCR can also be simultaneously used to examine spatial processes of the populations e.g. habitat selection (Royle et al. 2013a) and landscape connectivity (Royle et al. 2013b, Sutherland et al. 2015, Fuller et al. 2016). SCR connects population-level information to the spatial structure of the landscape by accounting for spatial location of the sampling sites and for spatial distribution of the individual encounters. Because SCR includes space explicitly, it allows inclusion of habitat covariates into the models of both density and detection probability. The relationship between the habitat and the density distribution of populations, i.e. second order habitat selection, is modeled by SCR as a density of activity centers as a function of habitat covariates. To study the habitat use of individuals within their home ranges, i.e. third order habitat selection, the habitat structure around the sampling locations or traps can be incorporated to SCR analyses to model how those covariates affect encounter probabilities (Royle et al. 2018). SCR can estimate the effect of certain habitat types on density and encounter probability, even if individuals are not directly encountered in that habitat, by predicting the locations of individual home range centers in the vicinity of the sample units. One main application of SCR has been research on large carnivores (Proffitt et al. 2015, Sun et al. 2017, Stetz et al. 2019, Welfelt et al. 2019), but non-invasive DNA sampling with SCR has also been used to study the relationship of mule deer (Odocoileus hemionus) population density and habitat structure (Brazeal et al. 2017).
White-tailed deer (Odocoileus virginianus) inhabit a large variety of terrestrial habitats and feed on various vegetation types (Halls 1984). In Finland, it has become one of the most important game species after its remarkable and continuing growth in abundance since 1934, when the species was introduced from North America. It is listed as a potentially or locally harmful species in Finland’s National Strategy on Invasive Alien Species (2012). One of the biggest impacts of white-tailed deer are deer-vehicle collisions, but in the areas of the densest population the species can also cause damage to agriculture and forestry for instance by eating vegetable crops and tree seedlings. For the management of this species, it is important not only to estimate abundance, but also to understand what habitat types the white-tailed deer prefers in Finland and how that affects the species densities. To this end, we conducted a SCR based study using fecal DNA in southwestern Finland. Our aim was to understand how white-tailed deer use available habitat types in a short time period (about two to three weeks) just prior to the start of the hunting season. The study was repeated in the same study area in two consecutive years. Additionally, we evaluated the importance of habitat covariates when inferring this species’ density by comparing the “standard” SCR model where density is assumed to be homogeneously distributed over space to inhomogeneous SCR models considering a heterogeneous landscape.