Integrated modeling of bilateral photo‐identification data in mark–recapture analyses

Abstract

Abstract. When natural marks provide sufficient resolution to identify individual animals, noninvasive sampling using cameras has a number of distinct advantages relative to ''traditional'' mark-recapture methods. However, analyses from photo-identification records often pose additional challenges. For example, it is often unclear how to link left-and rightside photos to the same individual, and previous studies have primarily used data from just one side for statistical inference. Here we describe how a recently developed statistical method can be adapted for integrated mark-recapture analyses using bilateral photo-identification records. The approach works by assuming that the true encounter history for each animal is a latent (unobserved) realization from a multinomial distribution. Based on the type of photo encounter (e.g., right, left, or both sides), the recorded (observed) encounter histories can only arise from certain combinations of these latent histories. In this manner, the approach properly accounts for uncertainty about the true number of distinct animals observed in the study. Using a Markov chain Monte Carlo sampling procedure, we conduct a small simulation study to show that this approach has reasonable properties and outperforms other methods. We further illustrate our approach by estimating population size from bobcat photoidentification records. Although motivated by bilateral photo-identification records, we note that the proposed methodology can be used to combine and jointly analyze other types of mark-recapture data (e.g., photo and DNA records).

multimark: an R package for analysis of capture–recapture data consisting of multiple “noninvasive” marks

“…2010; Bonner and Holmberg 2013; McClintock et al. 2013, 2014). Given a feasible set of latent encounter histories ( Y ), fitting capture–recapture models such as Eqs.…”

“…(2013). While the mathematical and computational details are generally of little interest to ecologists, performs these operations in the background and requires only simple data formatting and model specification formulas familiar to most R users.…”

“…An example of the “never” data type is provided with and includes 23 left‐sided $({\tilde{\mathbf{Y}}}_1)$ and 23 right‐sided $({\tilde{\mathbf{Y}}}_2)$ encounter histories for bobcats ( Lynx rufus ) collected from remote single‐camera stations in southern California over T  = 8 sampling periods between July 2006 and January 2007 (McClintock et al. 2013; Alonso et al. 2015).…”

“…There is also a feature enabling designation of individual encounter histories as known with certainty despite no simultaneous type 1 and type 2 detections (i.e., y i , t ≠4 ∀  t ), a situation that can arise from a previous physical capture or concurrent telemetry study (e.g., McClintock et al. 2013). …”

“…2000; Durban and Elston 2005; King and Brooks 2008; Royle 2008; McClintock et al. 2013). These software needs were the motivation for , an R (R Core Team 2013) package for Bayesian analysis of capture–recapture data consisting of multiple noninvasive marks.…”