Among conservation biologists there is little argument that habitat loss and fragmentation are the greatest threats to future biodiversity. We know from elementary population ecology that, if isolated, even large populations will ultimately be driven to extinction by demographic stochasticity, and that this will occur more rapidly for small populations in small fragments. Theory and observations also show that post-fragmentation species losses can occur slowly, sometimes requiring more than a century for the total ‘extinction debt’ to be realized (Tilman et al., 1994; Vellend et al., 2006). For long-lived animals, such as ambystomatid salamanders, isolated populations doomed to demographic irrelevance and ultimate extinction may persist for decades or longer (Greenwald, 2010). I think of the animals occupying these isolated habitats with little chance of long-term survival as ‘zombie’ populations. As imperfect as our population models are, they represent our best chance for understanding the threats to populations at risk and possibly recovering them from zombie status.
Population and metapopulation viability analyses are valuable tools for illustrating, in an understandable way, the risk of extinctions, and for exploring probabilities of local and regional persistence under different scenarios (Morris & Doak, 2002; Marsh, 2008). I see the greatest value of these modeling exercises in their ability to help land managers to evaluate alternative management actions and appreciate their relative costs and likely benefits. As Greenwald (2010) shows, actions to promote connectivity can enhance population viability. Restoring connectivity to fragmented landscapes is often extremely costly (think wildlife overpasses), and avoiding fragmentation completely may be even more controversial (think not building the freeway at all). So, models evaluating the importance of habitat connectivity must be convincing to those people on the hook for making big decisions. The greatest barrier to constructing convincing models is obtaining parameter estimates in which we have confidence, which traditionally means labor- and time-intensive mark–recapture studies. Fortunately, the recent work of Greenwald (2010) in addition to others indicates that molecular tools may be surprisingly robust to the challenge of more rapidly and painlessly estimating at least some demographic parameters needed for landscape-scale population modeling.
As someone who has invested years in the field marking and recapturing animals, I am optimistic at the promise these methods potentially hold (Trenham et al., 2000; Trenham, Koenig & Shaffer, 2001). The ability of genetic assignment tests to accurately reveal residents and immigrants in samples could change the study of dispersal. My caution in getting too excited about the current abilities of these methods stems from my understanding of several issues that can substantially bias the results. The author addressed the issue of unsampled ‘ghost’ populations. This is a real issue for animals like ambystomatid salamanders where large fractions of the adult population commonly skip breeding and are completely undetectable, thus samples from a single year are unlikely to accurately represent the standing genetic variation present in a population. Despite these issues, one recent study is very encouraging. A detailed study with marked skinks, occupying and dispersing among patches of rocky habitat, showed that assignment tests usually recognized immigrants with high confidence. And further, using skink tissues collected over 3 months, the assignment tests yielded estimates of interpatch dispersal rates essentially identical to those estimated over 7 years of mark–recapture efforts (Berry, Tocher & Sarre et al., 2004). At this point, I think there is still need for more cross validation of these methods with mark–recapture studies in other taxa and landscapes, especially if we are to use these data predicting metapopulation viability.
If these methods produce robust predictions they have the potential to greatly assist conservation planners. For example, the California tiger salamander Ambystoma californiense is the species about which I know the most, and one whose management would benefit greatly from the types of genetic data and the coupled modeling approach explored by Greenwald (2010). The most extreme situation exists in a 15 km by 5 km region of Sonoma County, California, USA, which is home to the entirety of a federally endangered Distinct Population Segment of A. californiense. The entire area is severely fragmented by a dense road network, development and agriculture. There are roughly 50–100 ephemeral breeding pools in which breeding has been observed during the last 10 years, and many of these are situated on small regional preserves – the largest encompasses just 70 ha. Because my empirically based estimate is that >125 ha is required to encompass the normal migratory movements of∼95% of salamanders around a single pond, I have serious concerns about long-term population viability on any of these preserves (Trenham & Shaffer, 2005). Although annual larval sampling indicates that the species remains widely distributed, there is essentially no recent information on breeding population sizes nor has there been any study of movement. Because California tiger salamanders can live 10 years or more, without additional information it is unclear whether populations are viable or entering the realm of the zombies. If genetic data could provide reliable estimates of population size or recent dispersal among ponds, or ideally contribute to a realistic metapopulation viability analysis, I can imagine many ways that available resources might be more effectively focused to improve the prospects for this highly endangered population segment.
Although approaches that make use of molecular genetic data are potentially very useful, for some purposes there is still no substitute for capture–recapture studies; I know of no other means for estimating survival. The modeling approach used by Greenwald (2010) relied heavily on generalized survival probabilities estimated at sites far from the study sites. This was a necessary simplification, but it limits my confidence in the specific predictions of the model. There is a great need for additional work to estimate survival in differing landscapes. Most capture–recapture studies are conducted on protected lands where the species of interest is abundant and conditions are generally benign. In contrast, the focal areas in this study, and those where managers often target conservation efforts, are often far less ideal. There was no real discussion in the paper regarding the habitat quality, level of fragmentation, and the degree to which these sets of ponds were thought to be isolated from other possible source populations. If mortality in these landscapes is substantially higher than the assumed values, model populations and metapopulations would most likely decline to extinction rapidly, even with the high dispersal values.
Finally some words about amphibians and metapopulations. My personal observations are that just because many amphibians use naturally patchy breeding ponds, the metapopulation label cannot be accurately applied to pond-breeding amphibians as a group. My own studies, and an extensive literature review indicated that in unfragmented habitat there is often too much dispersal and not enough extinction for classical metapopulation dynamics to be observed (Marsh & Trenham, 2001; Trenham et al., 2001). However, in marginal and fragmented habitats many amphibian populations do exhibit dynamics consistent with classical or non-equilibrium (i.e. declining) metapopulations (Marsh & Trenham, 2001). I see real value in the metapopulation modeling approach advocated by Greenwald's (2010) work, especially as a tool for exploring general guidelines for land management. If we hope to make confident predictions about the probability of extinction/persistence for a regional population of salamanders or other animals, survival, dispersal and other potentially important parameters should be estimated within the system of interest. With improving genetic methods to model dispersal, and powerful computational tools to estimate demographic parameters from marked populations, our ability to apply Greenwald's (2010) vision of landscape-scale population modeling is nearly within reach. Ultimately we need a tool like this to help convince decision makers that maintaining and promoting habitat connectivity remains our most important approach in the pursuit to minimize the number of zombie populations.