2.4 Importance of variables
To analyze the relative importance of environmental variables in explaining sea snakes’ distributions, we first classified the variables into ten groups based on their identity. For example, the group “Temperature (Tem)” included seven variables that represent the median of the mean, maximum and minimum values of the sea surface and benthic temperature (Table 1).
We used two measures to assess the relative importance of environmental variables: the percentage of contribution (PC) and the permutation of importance (PI) estimated by MaxEnt. PC measures the gain of a model built with a single specific variable divided by the total gain of a model built with all the variables. PI is obtained from randomizing the values of each variable so that this variable is not informative and then measures the resulting drop in the area under the curve (Phillips 2011). To identify the consistency or variability in our results, we performed this procedure at three levels: “lineage”, when we analyzed the variables for all species independently of subfamily; “subfamily”, where we divided the results into the Hydrophiinae and Laticaudinae subfamilies; and “genus”, where we repeated the same analysis for each of the six genera. We performed this analysis at both resolutions (5 and 10 arc-minutes).
We also separated the identity variables into two groups (high relative importance and low relative importance) based on their relative importance using the natural Jenks thresholds calculated with thegetJenksBreaks function of the “BAMMtools” package (Rabosky et al. 2014).
To determine if the results depended on spatial resolution, we evaluated the consistency of the importance of each variable between the two spatial resolutions for the two MaxEnt metrics (PC and PI) through a non-parametric Mann-Whitney-Wilcoxon test through thewilxicon.test function in “stats” package (R Core Team, 2013). Overall, our approach allowed us to identify the most important environmental variables driving the distributions of sea snakes and to assess the consistency of these results across different metrics of importance, taxonomic levels and spatial resolutions.