Predicting genetic biodiversity in salamanders using geographic, climatic, and life history traits

consistently predicted to be of significantly higher importance than other predictors in the model. Instead, importance was split across numerous predictors that were found to be unstable between models. This instability supports previous indications that many of the random forest models are likely prone to overfitting. Despite the lack of a strong set of stand- out predictors across models, one pattern does emerge that is applicable to the species in our dataset. Of the top ten most important predictors in each model, approximately 85% are measurements of standard deviation (vs. measurements of mean values or life history traits) (S5 Data). This is supported by further examination of our one model that was able to predict significantly better than random, the majority rules consensus with a correlation coefficient cutoff of 0.90, in which the top five most important predictors are measurements of standard deviation. Significance testing indicates that species identified as containing hidden genetic lineages often have ranges characterized by a larger variance in annual and seasonal precipi- tation, isothermality, and net primary productivity than species not identified as harboring hidden genetic lineages (Fig 5). Fig 5. Comparison of hidden vs not hidden trait values for the top five most important predictors of the best consensus model (majority rules– correlation cutoff 0.90). A, Columns 1–2 of the table identifys the specific model and predictors (i.e., traits). Columns 3–4 show the median trait values for

each group (i.e., hidden vs not hidden). Columns 5–6 show the results of Kruskal-Wallis significance tests, which determine if the difference in median trait values for each group is statistically significant. B, Corresponding boxplots of the median trait values for the top five most important predictors show a significant difference in the range of values between hidden and non-hidden genetic lineages. https://doi.org/10.1371/journal.pone.0310932.g005 PLOS ONE Predicting hidden lineages in salamanders PLOS ONE | https://doi.org/10.1371/journal.pone.0310932 October 18, 2024 11 / 20