Population viability analysis (PVA) and genetic assignment tests are two important items in the toolbox of contemporary conservation biologists, but appear to have little in common as they are applied to scientific problems of different scale and scope. PVAs use life history and environmental data to predict whether populations are able to survive a certain period of time, for example comparing different management strategies. Assignment tests, on the other hand, are statistical procedures applied to individual multi-locus DNA fingerprints (most often derived from microsatellites), and measure the likelihood with which a genetic profile originates from a specific gene pool. Applied to wild populations, assignment tests can, for example, identify individual migrants, which bear genetic signatures that match better to a distant population rather than to the population where they were sampled.
In a remarkable paper that uses three datasets from temperate urodele amphibians (North American salamanders of the genus Ambystoma), Greenwald (2010) explores assignment tests to quantify the degree of connectivity between demes comprising a metapopulation, an important parameter to predict whether a species is locally able to survive. The general idea is that empirically collected genetic data should provide a more flexible and realistic measure of migration than the commonly applied approach of spatial dispersal modelling (assuming a general dispersal function calibrated with field data, in which short-distance movements are common and long-distance movements are rare). Greenwald (2010) shows that assignment tests, although they only approximate the true figures of dispersal depending on statistical power and theoretical assumptions, lead to a potentially more realistic parameterisation of migration rates in PVA models, and, as an important consequence, to markedly lower estimates of metapopulation survival. Moreover, in the absence of population census data such as derived from capture–mark–recapture, Greenwald (2010) also used genetic data to measure individual abundances in each dataset approximated through effective population sizes (Ne).
With regard to expanding our knowledge about the dynamics of populations in fragmented habitats, Greenwald (2010) usefully bridges different analytical levels that are usually unconnected with each other. This can be illustrated by looking at two further studies conducted on a pond-breeding urodele amphibian, the European crested newt Triturus cristatus. Using a predominately genetic approach, Jehle et al. (2005) revealed that between-deme dispersal rates are usually asymmetric (mostly, large demes serve as sources for small demes, which constitute sinks), and that dispersal events appear to some degree stochastic (not all ponds that could exchange individuals based on their geographic distance actually do so). However, despite detailed gene flow information combined with genetic Ne estimates, the study failed to draw any conclusion about population size fluctuations and metapopulation viability. Griffiths, Sewell & McCrea (2010), on the other hand, investigated connected T. cristatus populations based on known life histories of adults, and predicted that population persistence was largely dependent on adult survival (negatively affected by mild winters) and modelled pond connectivity via juveniles, which were only produced in some pond–year combinations. As recruits were impossible to mark in the field, measuring actual rates of gene flow and migration across all life stages remained however elusive. What Greenwald (2010) did with Ambystoma is precisely what would be desirable for subdivided T. cristatus populations: completing our picture of present and future population dynamics by incorporating powerful tools from demography and genetics across different scales.
Building upon the work of Greenwald (2010), aspects that deserve further investigation relate to the distinctions between census and effective population size, as well as migration per se and gene flow through reproduction of immigrants. In tailed amphibians, the former two differ from each other by a factor of about 2–10 (e.g. Jehle et al. 2005). No empirical data are available on the reproductive success of immigrants in recipient populations, but it is likely that not all are successful in finding mates. In Greenwald's three study systems, the differential sampling regimes (either adults, which can be migrants, or larvae, which can only be descendants of migrants) allow a quantification of either of the two connectivity measures, and it would be possible to discern between them through adjusted PVA models. Another, perhaps more important aspect of consideration is that genetic assignment tests are only able to measure dispersal over a timescale of usually 1–2 generations. Over the longer periods that are typically projected with PVAs (in the case of Greenwald, 100 years), the extent and direction of migration is, however, likely subject to temporal fluctuations: source demes can turn into sinks and vice versa, and the actual migration rate can for example depend on the yearly amount of recruitment (in a good year, an excess of offspring in one pond might result in a temporary wave of immigration in a neighbouring pond, see for example Griffiths et al. 2010). Although a practically tedious endeavour, it would therefore be worthwhile to repeatedly obtain dispersal measures from assignment tests across several seasons or generations. A further question that should be addressed in the future is whether assignment tests as a general rule produce more pessimistic viability estimates for metapopulations than estimates based on spatial dispersal modelling, or whether this result is specific to Greenwald (2010). This needs the incorporation of alternative distance functions based on further field-derived migration data, and/or an extension of the approach to other study species and systems.
Although it has been demonstrated in many occasions, Greenwald (2010) once again brings to our attention that the precision of PVAs crucially depends on the extent to which the model parameters reflect the real world. The probably most often cited quote in ecological modelling, ‘essentially all models are wrong, but some are useful’ (by George Box, a son-in-law of Sir Ronald Fisher) can be easily translated into a more positive ‘the best models are the most useful ones’. Greenwald (2010) builds a remarkable bridge between analysing individual DNA fingerprints and predicting the fate of whole metapopulations, and the approach has a lot of potential to improve existing PVA models whenever sufficient population genetic data are available.