Estimates of gene flow resulting from landscape resistance inferences frequently inform conservation management decision-making processes. Therefore, results must be robust across approaches and reflect real-world gene flow instead of methodological artefacts. Here, we tested the impact of 32 individual-based genetic distance metrics on the robustness and accuracy of landscape resistance modelling results. We analysed three empirical microsatellite datasets and 36 simulated datasets that varied in landscape resistance and genetic spatial autocorrelation. We used ResistanceGA to generate optimised multi-feature resistance surfaces for each of these datasets using 32 different genetic distance metrics. Results of the empirical dataset demonstrated that the choice of genetic distance metric can have strong impacts on inferred optimised resistance surfaces. Simulations showed accurate parametrisation of resistance surfaces across most genetic distance metrics only when a small number of environmental features was impacting gene flow. Landscape scenarios with many features impacting gene flow led to a generally poor recovery of true resistance surfaces. Simulation results also emphasise that choosing a genetic distance metric should not be based on marginal R2-based model fit. Until more robust methods are available, resistance surfaces can be optimised with different genetic distance metrics and the convergence of results needs to be assessed via pairwise matrix correlations. Based on the results presented here, high correlation coefficients across different genetic distance categories likely indicate accurate inference of true landscape resistance. Most importantly, empirical results should be interpreted with great caution, especially when they appear counter-intuitive in light of the ecology of a species.
CORE (COnnecting REpositories)
CORE (COnnecting REpositories)