*3.1. Selection of Variables*

The results of the variable selection process revealed the importance of 4 and 5 variables out of the initial 98 variables for *Eucalyptus dunni* and for *E. grandis* model projection, respectively, as being non-collinear (n = 27; *VIF* < 10); and highlighted by the variable selection procedure run by the AUCRF R package. The selected variables were classified by importance (1 being the most important, and 5 the least important), considering the mean decrease Gini importance coefficient estimate with the AUCRF R package (Table 2). The selected variables with greater predictive power for *E. dunnii* included: the depth of the A horizon, the highest and lowest temperatures of April and May, respectively, and the average temperature of the driest month. In the case of *E. grandis*, the most important variables were the percentage of clay, the depth of the A horizon, the isothermality, the percentage of silt, and the orientation.

The response curves analysis shows that the probability of occurrence for *E. dunnii* decreases as the temperature of the driest quarter increases and increases in soils where the A horizon is deeper (Figure 2A). The effect of the temperature of the driest trimester remains constant as it rises above 15 ◦C, while as the depth of the A horizon increases above 4 cm a constant increase in the probability of occurrence is observed. For *E. grandis*, the probability of occurrence is associated positively with

the thickness of the A horizon and the aspect, but negatively with the percentage of clay in the A horizon (Figure 2B). The probability of presence is highest when the proportion of clay is near 0% and remains constant when it exceeds 25%. Similarly, to *E. dunnii*, an A horizon thickness greater than 4 cm determines an important increase in the probability of occurrence. The orientation has a quadratic effect, with a higher probability of presence in the extremes of the range; the probability is highest for aspects over 300◦ (north-west). On the other hand, the isothermality shows a polynomic relationship with the probability of occurrence, the probability being highest with levels close to 47.5%. The silt content in the A horizon has an effect very similar to that of the clay content.

**Table 2.** Ranking of the importance of independent variables for prediction of the distribution of *Eucalyptus dunnii* and *E. grandis*.


Note: Bio 9—Mean Temperature of Driest Quarter.

**Figure 2.** Response curves showing the average probability value of the ensemble model for each explanatory variable, for *Eucalyptus dunnii* (**A**) and *E. grandis* (**B**).
