Including Phylogenetic Conservatism of Shortgrass Prairie Restoration Desired Species Does Not Improve Species Germination Predictions

Y. Chen, D. Schwilk, R. Cox, and M.G. Johnson, bioXriv (2022).


PREMISE We investigated whether phylogenetic conservatism can improve the performance of seed germination prediction models. Previous studies in tallgrass prairie and alpine meadow revealed that seed morphological traits demonstrate phylogenetic conservatism. We hypothesized that phylogenetic conservatism in seed traits could help predict the seed germination, under the assumption that seed traits contain phylogenetic signals.

METHODS We measured seed germination percentage and seed morphological traits (seed mass, seed height, and seed surface area) on 34 native species from shortgrass prairie in North America. We supplemented these data with similar data from the literature on 11 more species. We calculated the robustness of phylogenetic signal of each trait to the number of species sampled. We also compressed the phylogenetic distance matrix to a two-dimensional space, and applied Akaike information criterion to evaluate the effects of phylogeny on seed germination prediction models.

KEY RESULTS We found weak but significant phylogenetic signal in seed mass and seed height in the full data set. These phylogenetic signals were not able to improve seed germination prediction model performance among shortgrass prairie species. Our robustness tests of phylogenetic signals using random sub-sampling showed that the detection rate of phylogenetic signals in seed mass was increased along with the expansion of species pool, and nearly 100% at 40 species. However, the detection rate of phylogenetic signals in seed height was constantly low, around 20%.

CONCLUSIONS When the phylogenetic signals are weak, the phylogenetic position does not improve germination prediction model performance. Therefore, phylogenetic signals detected during a single species pool calculation may not accurately reflect the phylogenetic conservatism of the trait in an ecosystem. We suggest tests for robustness of phylogenetic signals by applying random sub-sampling tests.