Predicting date-of-death
We predicted timing of death of trees as a secondary metric to assess the importance of drought stress in driving beetle outbreaks by including metrics related to tree size, vigor, and drought stress. Predictions of date-of-death were conducted at the tree-level. More specifically, tests of date-of-death used annually resolved ∆13C data averaged across all the trees by site, and by time of death, whereas tree size (DBH) and growth rate (BAI) data were used on the individual tree-level. ∆13C values for each site and timing category were applied to all trees within the site (e.g., all early-dying trees at the ASH site were given the same ∆13C value for a given year). Similar to the temporal resolution of ∆13C, annual BAI values matching the timeframe of ∆13C (1960-death) were used. A linear mixed effects model was used to predict date-of-death per tree. Due to the variability among plots, date-of-death was calculated as the number of years from the year of peak mortality by site (positive values indicate years of survival post peak mortality, and negative values, years prior to the peak mortality that trees died, Table 1). The model was fitted with all hypothesized main effects (BAI, DBH, ∆13C, and their interactions) and interpreted in terms of significance at p < 0.05. Random effects were used to account for the nested nature of years repeated for trees within sites. All analyses were done using the “nlme” and “visreg” packages in R (R Core Team 2016; Breheny and Burchett 2017; Pinheiro et al. 2017).