Populations come in varying levels of spatial structuring and degrees of spatial mixing (Thomas & Kunin, 1999). The degree and nature of such mixing among individuals in a population can have profound effects on population dynamics and persistence (e.g. Lloyd, 1967; Kareiva & Wennergren, 1995; Law, Murrell & Dieckmann, 2003; Hastings & Botsford, 2006). Due to habitat loss and fragmentation, many endangered populations now seem to be structured in space as metapopulations (Hanski & Caggiotti, 2004), but whether or not they actually function as such needs to be carefully addressed (Clinchy, Haydon & Smith, 2002).
In this issue, Van Oort, McLellan & Serrouya (2011) present a study of the movements of hundreds of mountain caribou throughout a period of 23 years (some animals were re-collared and followed for up to 11 years!). This ecotype of caribou has experienced a strong population decline and is currently confined to mountainous areas surrounded by unsuitable habitat. To date, most of the telemetry work on this species has been done on adults and subadults, but Van Oort et al. (2011) were able to use telemetry on 26 calves belonging to 10 different subpopulations. Tracking of calves and yearlings is crucial because we expect that most of the connectivity among subpopulations will depend on dispersal at these age classes (Greenwood, 1980). However, Van Oort et al. (2011) did not observe any dispersal events from either calves or yearlings. Furthermore, they registered only two between-subpopulation dispersal events out of 521 opportunities for adult/subadult females. These results put serious doubts on any hopes that the remaining subpopulations of Mountain Caribou would benefit from rescue effects.
However, it seems to me that further studies need to be done before we formally close the case for between-subpopulation dispersal in mountain caribou. Dealing with samples of populations implies that we are left with some uncertainty about whatever we measure. This uncertainty is best handled by Bayesian analysis (Clark, 2005). If we assume that every dispersal opportunity is identical and independent, we can use a Binomial distribution to model the number of successful events. Furthermore, using a beta distribution to represent our prior beliefs about migration probabilities allows us to represent our updated knowledge about such probabilities (after observing how many successful migrations we got out of the total number of opportunities) with another beta distribution. For example, Van Oort et al. (2011) observed no dispersal out of 17 possible events in yearlings. We can assume previous ignorance about dispersal probabilities by setting our prior distribution as beta(1,1) which is uniform between 0 and 1. The posterior for migration probabilities is then beta(1+observed dispersal events, 1+total opportunities−total observed dispersal events), which for this case is beta(1,18). This posterior has an expected value of 0.0526 with 95% of the distribution contained between 0.0014 and 0.1853.
This uncertainty around dispersal probability can be incorporated in population models using a beta-binomial distribution for between subpopulation movements. For yearlings, this translates to 95% quantiles of 0–20 dispersal events (with a mean of 5.26) out of a population of 100 yearlings, or 0–94 (with a mean of 26.33) for a population of 500. Less uncertain values are obtained for females older than 3 years where only two dispersal events out of 521 opportunities were observed. This yields an estimated probability of dispersal of 0.00574 (0.00119–0.01377), which implies 0–3 dispersal events for a population of 100 females or 0–9 for a population of 500. These quick calculations can be of course easily challenged because there might be no reason to have a uniform prior on dispersal probability. In fact, we might expect low dispersal probabilities to begin with and the fact that no dispersal events were observed for calves and yearlings will result in much narrower credible intervals. However, as far as I am aware, no metapopulation models for mountain caribou have been analysed so that we can judge whether such low dispersal probabilities are relevant for population dynamics especially for these long-lived animals. In a simulation study of cougar spatial populations dynamics Beier (1993) found important increases in persistence if as few as one to four animals per decade could immigrate into a small population. Would similar results apply for gregarious animals such as caribou?
In any case, and leaving the above digression aside, the study of Van Oort et al. (2011) as well as that of Faille et al. (2010), highlight the fact that site fidelity in this species can be an important component when addressing extinction risk as individuals might stick to their ranges even if local conditions are worse than elsewhere within their movement capabilities. Individuals that live in herds might have little reason to depart a population, and the short-term fitness advantages of staying at home might be at odds with longer-term population persistence (Matthiopoulos, Harwood & Thomas, 2005).
Dispersal involves the three distinct stages of departure, transience and settlement (Bowler & Benton, 2005; Clobert et al., 2009). Success at each stage would depend on individual condition (nutritional, reproductive state), phenotype (sex, endurance) and experience (memory). For most endangered species, we still know little about how animals decide to leave their territory or abandon a group, and how they explore and choose where to establish new territories or home ranges. It is becoming increasingly apparent that we need to refine our understanding of the mechanistic links between behaviour, individual condition and movement in order to predict population dynamics (Morales et al., 2010). Studies like the one of (Van Oort et al., 2011) in this issue, highlight the fact that what looks like a metapopulation might not actually function as one, and that we need solid data on life history and behaviour in order to make informed conservation decisions.