Fig. 3

Accuracy metrics for the selected random forest models of snag fall within 5 years after fire for snags that existed before the fire (pre-fire) and snags that died in or after the fire (post-fire). Panel rows indicate whether the models were constrained to pre-fire or post-fire predictor variables. Within each panel, up to seven models occupy seven rows (see Table 2); the accuracy of each model is represented by the three boxplots that occupy that row. The boxplots represent the distribution of the model accuracy from 15 random forest analyses, each run with a different seed. A–D The models with the best performance in the lowest performing model accuracy metric for each snag population and predictor variable set. r-values in the lower-left corner of each panel indicate the collinearity threshold used to select the predictor variables in that panel (see the “Random forest models” section)