Variable importance in random forest models of stomatal conductance for (a) grasses and (b) woody plants.

<p>For both grasses and woody plants shallow (0–20 cm) soil moisture and soil type (<i>i</i>.<i>e</i>., clay or sand) were the most and least important variables describing g_s, respectively. Variable importance is the difference in prediction error before and after a predictor variable is randomly permutated. Large variable importance values indicate that specifying the variables incorrectly increases prediction error. See text for further variable descriptions.