Fig. 3

Multiway importance plot of 25 variables (top ten predictors are labeled) used to predict probability of fire effects on prehistoric artifacts and features after five fires that occurred between 2000 and 2011 in the Jemez Mountains, New Mexico, USA (n = 858). See Table 2 and Additional file 1 for full names and descriptions of predictors. For mean decrease in accuracy (x-axis), more important predictors result in greater decreases when removed. The Gini coefficient represents node impurity. Variables that lead to more homogenous nodes give greater decreases in mean Gini. Therefore, a higher mean decrease in Gini (y-axis) indicates higher variable importance in the model. P-values <0.01 indicate predictors that were selected more frequently as tree nodes than would be expected if predictors were selected randomly