Sampling effect in predicting the evolutionary response of populations to climate change.

Genomic data and machine learning approaches have gained interest due to their potential to identify adaptive genetic variation across populations and to assess species vulnerability to climate change. By identifying gene-environment associations for putatively adaptive loci, these approaches project changes to adaptive genetic composition as a function of future climate change (genetic offsets), which are interpreted as measuring the future maladaptation of populations due to climate change. In principle, higher genetic offsets relate to increased population vulnerability and therefore can be used to set priorities for conservation and management. However, it is not clear how sensitive these metrics are to the intensity of population and individual sampling. Here, we use five genomic datasets with varying numbers of SNPs (N = 7006-1,398,773), sampled populations (N = 23-47) and individuals (N = 185-595) to evaluate the estimation sensitivity of genetic offsets to varying degrees of sampling intensity. We found that genetic offsets are sensitive to the number of populations being sampled, especially with less than 10 populations and when genetic structure is high. We also found that the number of individuals sampled per population had small effects on the estimation of genetic offsets, with more robust results when five or more individuals are sampled. Finally, uncertainty associated with the use of different future climate scenarios slightly increased estimation uncertainty in the genetic offsets. Our results suggest that sampling efforts should focus on increasing the number of populations, rather than the number of individuals per populations, and that multiple future climate scenarios should be evaluated to ascertain estimation sensitivity.