The use of genetic data in population viability analysis has promise as a broadly applicable approach to addressing conservation questions (Greenwald, 2010). However, the commentaries on my article in this issue (Jehle, 2010; Koons, 2010; Trenham, 2010) bring up several important points that warrant reiteration for researchers interested in conducting similar analyses. Toward that end, I outline here three issues that should be considered before applying this methodology, and include for each the ‘ideal world’ scenario. Researchers must decide on a case-by-case basis how well a particular dataset or study system conforms (or fails to conform) to the ideal, as this will determine the accuracy and broader applicability of model results.

(1) Appropriate conceptual framework. Since the development of the metapopulation framework, it has been used extensively, especially in amphibian studies. However, as Trenham (2010) points out, the underlying assumptions are not always addressed or tested (Marsh & Trenham, 2001; Smith & Green, 2005). For example, of 53 studies reviewed by Smith & Green (2005), only 10 tested whether any single population was large enough to ensure long-term survival, and only three tested whether local dynamics were asynchronous. Regardless of the organism of interest, this is a critical first step to consider if the goal is to use RAMAS/Metapop (Akçakaya, 1998) or another modeling method that assumes metapopulation structure. This framework is not appropriate for organisms in large, continuous populations (i.e. no probability of local extinction) or for those in completely isolated populations (i.e. no probability of recolonization); in these cases alternative population models must be developed that do not assume metapopulation structure. Indeed, Koons (2010) suggests development of situation-specific population models, which allows for greater flexibility in defining the underlying assumptions. Clearly, the goal should be as close a match as possible between population models and the ‘real world’ scenario, and careful consideration should be given to population structure and demographic parameters before model development. All models will provide an answer; the accuracy and relevance of that answer depend entirely on the quality of model development and parameterization.

(2) Availability of relevant demographic data. Ideally, data on population parameters such as mortality and fecundity would be available from long-term capture–mark–recapture (CMR) studies of the populations of interest. These data would serve three important purposes. First, they would act as a comparison or replacement for genetic estimates of effective population size, which (as noted by the commentaries) may influence model results in some systems (Koons, 2010) and may vary greatly from census population sizes (Jehle, 2010). Second, dispersal events documented by CMR could validate assignment test results (Berry, Tocher & Sarre, 2004; Koons, 2010; Trenham, 2010). Finally, I reiterate an excellent point made by Trenham (2010): demographic parameters likely vary depending on landscape context. For example, mortality might be underestimated by using values derived from populations in pristine landscapes, and this could clearly influence model results. As he noted, this was a necessary simplification in my study, but ideally demographic data would come from the populations of interest or from populations in similar landscapes.

The assessment of demographic parameters along fragmentation gradients, while logistically challenging, would be an extremely valuable direction for future research. This would allow for accurate model parameterization across a range of landscapes, as well as further comparisons of models based on dispersal–distance curves versus genetic assignment tests (e.g. is there a certain critical level of habitat heterogeneity above which assignment tests outperform dispersal–distance functions?).

(3) Thorough and representative genetic sampling. It is important to consider the spatial and temporal extent of sampling when designing a population genetic study. An ideal study system would consist of all potential source populations within a biologically meaningful dispersal distance. If this is not feasible, awareness of the quantity and location of these unsampled ‘ghost’ populations would aid interpretation of assignment test results. If there are numerous ghost populations, a less stringent assignment test approach might be most appropriate (e.g. the ‘high dispersal’ scenario; Greenwald, 2010). The ideal temporal extent of sampling will depend on the organism of interest; multi-year sampling would be especially valuable in cases with potential temporal variation in population genetic structure (e.g. different breeding cohorts) or in dispersal across years (Jehle, 2010; Trenham, 2010).

Another important consideration is balancing the number of sampled individuals and the number of loci. Several simulation studies have explored the power of assignment tests under various combinations of number of alleles, loci, individuals, populations and levels of population differentiation. For moderately differentiated populations (FST=0.1) one study showed that 100% correct assignment could be obtained using a Bayesian assignment method with 10 microsatellite loci from 30 to 50 individuals from each of 10 populations (Cornuet et al., 1999). Adding additional loci appears to increase power much more than adding polymorphism (alleles/locus; Bernatchez & Duchesne, 2000) or increasing the number of sampled individuals (Paetkau et al., 2004).

In summary, the approach of combining population genetic data and viability analysis can be a useful tool for conservation research, with the potential to identify source populations, populations at risk of local extinction, and persistence estimates for population networks. It may be especially valuable for organisms in fragmented or degraded habitat, where the assumption underlying dispersal–distance curves (equal probability of dispersal in any direction) is likely to be violated. I will be interested in the success of this technique in other systems, especially for long-term study systems where the ‘ideal’ conditions outlined above can be met or approached (e.g. the European crested newt Triturus cristatus, Jehle, 2010; or the California tiger salamander Ambystoma californiense, Trenham, 2010). However, in the ‘crisis discipline’ (Soulé, 1985) of conservation biology, the need for expedient decisions may preclude this level of thorough information gathering. If time does not permit a long-term CMR study, genetic assignment tests can provide a rapid empirical estimate of population connectivity for use in viability modeling.


  1. Top of page
  2. References