1 Introduction
Robust abundance estimates are fundamental parameters for managing
wildlife populations, and central to understanding extinction risk
(Campbell et al., 2002; Lande, 1993; Shaffer, 1981). Monitoring and
understanding variation in abundance is critical for recovery efforts of
threatened and endangered populations; however, producing accurate
population estimates remains a challenge for many species. This is
particularly true for species that occur at low density or in low
abundance, that are cryptic, or that exhibit elusive behaviours which
make capture difficult (Kéry, Gardner, Stoeckle, Weber, & Royle, 2011;
Pollock, Marsh, Lawler, & Alldredge, 2006). Non-spatial
capture-recapture (CR) analyses have been the standard method used to
estimate abundance of many vertebrate species, however,
spatially-explicit capture-recapture (SCR) models are becoming the new
standard because they are robust to small sample sizes, produce precise
density estimates, and can accommodate low capture probabilities
(Borchers & Efford, 2008; Efford, Borchers, & Byrom, 2009; Ivan,
White, & Shenk, 2013; Royle, Chandler, Sollmann, & Gardner, 2013). By
including spatial information of captured individuals directly into the
analyses, SCR models resolve issues surrounding the effective trapping
area and temporary migration and are robust to assumptions about
geographic closure that are common issues in non-spatial CR studies
(Efford & Fewster, 2013; Royle et al., 2013). Recapturing individuals
at different locations also provides information on individual activity
centers, which are used to estimate animal density within the study area
(Borchers & Efford, 2008; Royle et al., 2013).
SCR models directly depend on adequate number of unique individuals
captured and recaptured at multiple spatial locations (Efford &
Boulanger, 2019; Sun, Fuller, & Royle, 2014). Simulations are
recommended to enable the assessment of sampling design on SCR parameter
estimates, to inform optimal sampling design (Royle et al., 2013). Such
studies have primarily focused on large carnivores, such as black bears
(Ursus americanus ; Clark, 2019; Sollmann et al., 2012; Sun et
al., 2014; Wilton et al., 2014), and a few additional taxa (Kristensen
& Kovach, 2018; Tobler & Powell, 2013), while limited work has been
done on species occurring at low densities over large areas and with
more limited home range sizes. Non-invasive genetic sampling approaches
can be used to alleviate the challenges associated with surveying rare
and elusive species, by constructing capture histories from DNA
collected from feces, hair, or other non-invasively collected samples
(Kristensen & Kovach, 2018; Lampa, Henle, Klenke, Hoehn, & Gruber,
2013; Waits & Paetkau, 2005). Non-invasive methods often result in
higher capture rates and lower expense than traditional
capture-recapture methods (Lampa et al., 2013; Prugh, Ritland, Arthur,
& Krebs, 2005; Waits & Paetkau, 2005), and SCR is increasingly being
used in combination with non-invasive methods (Royle et al., 2013;
Kristensen & Kovach, 2018; Lamb et al., 2018). Knowledge of the target
species’ home range size helps inform the spatial sampling design
(Sollmann et al., 2012; Sun et al., 2014), but empirical studies are
still necessary as detection probabilities may be influenced by other
factors (e.g. variable habitat conditions (Efford & Fewster, 2013).
Efford & Boulanger (2019) presented formulae to determine the precision
of new study designs by computing intermediate variables, such as the
number of detected individuals and expected number of recaptures, which
strongly correlate with precision. However, these formulae require
starting values for density and detection parameters (Efford, 2019b),
which may not be available for less studied species.
Here, we developed a framework to assess results from empirical studies
to inform sampling designs (Fig. 1). The framework consists of (1)
determining the number of unique individuals captured and spatially
recaptured from empirical data; (2) running SCR models under the
assumption of homogeneous distribution to estimate the detection
parameters g0 (detection probability) and \(\sigma\) (spatial
extent of an individual’s use of the landscape) to assess the precision
of the density estimates; (3) running simulations to assess the
influence of the species’ behaviour on density estimates and relative
bias; (4) using empirical data to assess different sampling designs and
evaluate precision and relative bias of the estimates; and (5) making
recommendations on study design based on the resulting precision and
relative bias of the estimates. The framework is implemented in R (R
Core Team, 2019), using maximum likelihood methods.
To collect empirical data, we completed aerial surveys across the ranges
of seven boreal caribou populations in Alberta, Canada. These ranges
varied in size, exhibited differences in estimated caribou population
density, and contained different levels of natural and anthropogenic
disturbances (Fig. 2; see Appendix 1 for details). For each caribou
population we used an aerial transect survey design to conduct
non-invasive genetic sampling, through the collection of caribou fecal
pellets. While we studied boreal caribou, our approach for evaluating
study design is applicable to other species and systems.