Best-ranked growth-mortality models.

Variables include average growth (RW), mean sensitivity (Sens), growth trend (Trend), and abrupt growth changes (AbruptIncreases), with the number of years over which variables were averaged indicated after variable type. The best single-variable models in different growth categories, along with a model containing recent average growth as the only predictor variable (log(RW3)), are shown for comparison. ΔAIC is the difference in AIC between the best-ranked model and the model shown in each table row, with smaller values indicating more parsimonious model fit. ROC is a threshold independent measure of model discrimination, where 0.5 suggests no discrimination and values above 0.8 suggest excellent discrimination. Correct classification rates (CCR) are based on a bootstrapped internal validation with 1000 iterations in which 60% of the data was used for model fitting and 40% was used for model validation. Trees were classified as living if model output was greater than the empirically defined threshold [70]. ROCboot is an average of the ROC statistics generated in the model-fitting portion of the bootstrapping routine. The kappa statistic measures the proportional improvement of the model classification over a random assignment of tree status [99], and was also estimated by taking an average of kappa statistics generated in the bootstrapping routine.