Results
13C chronologies averaged across early- and late-dying trees from all sites were correlated with summer (i.e., June, July and August) monthly maximum temperatures, climatic moisture deficit (CMD), water year precipitation, and Palmer Drought Severity Index, and averaged -0.38, -0.44, 0.26, and 0.36, respectively, (Appendix S1: Table S2). In contrast, ring width, latewood width, and earlywood width chronologies had correlations with the same climate variables that were near zero (Appendix S1: Table S2). As expected, the slopes of ∆13C versus summer CMD were generally negative for early- or late-dying trees at each site. Because ∆13C was most highly correlated to CMD when compared to other climate variables (Appendix S1: Table S2), subsequent tests of drought sensitivity were defined as the slope of ∆13C vs summer CMD.
Models that assessed drought sensitivity revealed no significant difference between timing of death (early or late mortality during the beetle outbreak at each stand) regardless of which approach was used (hypothesis testing or model comparison). The slopes of ∆13C vs CMD, which reflected the drought sensitivity of leaf gas exchange, were very similar between early- and late-dying trees and did not differ significantly (Table 1; Fig. 4; Appendix S1: Table S3). These overall patterns reflected greater variation in slopes of ∆13C vs CMD among sites for late-dying trees, and the extent to which the variation in drought sensitivity for late-dying trees overlapped with the slopes of early-dying trees was consistent among sites (Fig. 5).
The only evidence we found for differential susceptibility of early- vs late-dying trees to drought stress was when tests were performed on years including only the top 66 percentile of CMD, or the warmest and driest years in the dataset. Moreover, these differences in the sensitivity of responses of Δ13C to variation in CMD were only apparent when a three-way interaction of CMD, timing of death, and the arithmetic mean tree diameter (DBH) of the early/late-dying trees was included (see Appendix S1 for full model methods and results). Without accounting for DBH in the model, drought sensitivity did not differ between early/late-dying trees even in the warmest, driest years (Appendix S1: Table S4, Appendix S1: Fig. S3). In an attempt to account for the site-specific variability in drought sensitivity (Fig. 5), models of drought sensitivity that included stand structure showed one potentially significant effect of drought sensitivity, depending on stand conditions (Appendix S1: Fig. S4). Sites with a large range in the DBH of late-dying trees—indicative of stands with an abundance of large host trees pre-disturbance, where some died at the beginning of the outbreak but others died later—showed a difference in drought sensitivity based on timing of death, where late-dying trees were less sensitive to drought compared to early-dying ones (Appendix S1: Fig. S4). However, our sites did not encompass the range in stand structural conditions expected across all P. engelmannii types, and thus this line of inquiry needs further investigation (see Discussion).
Tree diameter (DBH) was the only factor (p < 0.05) found to affect the timing of individual tree death (i.e., years before or after peak death date within a stand; Appendix S1: Fig. S5). As tree size increased, timing of death occurred sooner such that trees with DBH > 57 cm tended to succumb to D. rufipennis early in the outbreak, and trees < 57 cm DBH tended to survive longer (Appendix S1: Fig. S5). Growth rates (BAI), and leaf gas exchange (∆13C), nor any other interactions with these variables were significant in affecting timing of death.