Using computer simulations to assess sampling effects on spatial genetic structure in forest tree species

Quantifying spatial genetic structure is key to inform forest management and restoration strategies. Reliable evaluations of genetic structure require sound sampling schemes because inappropriate sampling may over- and under-estimate spatial patterns of genetic structure. Sampling bias has been investigated through computer simulations mostly for animal species with continuous distributions. For tree species that have different life history traits, results from such studies may not apply. Here, I used spatially explicit landscape genetic simulations to assess the effects of spatial sampling scheme (random, systematic, and cluster), sampling intensity (35, 50, 65, and 80%), and the number of microsatellite loci (8, 14, and 20) on inferences of genetic structure under isolation by distance (IBD) in two forest tree species with varying dispersal distances and patchy distributions. Results showed that random sampling with 20 loci was the best performing sampling scheme, irrespective of sampling intensity and the strength of IBD. In contrast, the cluster and systematic sampling were sensitive to sample size. For the three sampling schemes, the number of loci had a large effect because with 8 loci there was an increasing chance of underestimating IBD. Increasing the number of samples over the number of loci, did not improve the performance of sampling schemes. Hence, researchers should put more effort on increasing the number of loci over increasing sample size. Results also showed that sampling error rates varied between species, and sampling bias appeared stronger for the species with a more aggregated spatial distribution.