5.6 Data processing and analysis
We investigated differences in P12 and P50 across species and stands (hypotheses 1 and 2) with a two-way ANOVA, where species and stand age were fixed factors and region was a blocking factor. We compared vessel density and lumen area with a two-way ANOVA, where species and stand age were fixed factors. We removed region as a blocking factor because there was no significant region or region interaction effect at p = 0.05. The relationships between xylem anatomy (e.g., vessel density and vessel lumen area) and embolism thresholds (e.g., P12 and P50) were assessed with a least squares linear regression within and across species. All ANOVA analyses were performed at the α = 0.05 level and were followed by a Tukey post-hoc test for significant main effects. Significant interaction terms were assessed by pairwise comparison of least square means.
We analyzed the relationship between embolism thresholds and degree of isohydricity (hypothesis 3) in two ways. First, we used in-situΨL observations and laboratory generated xylem embolism curves to estimate the percent of native embolism across species and stands during the study period. Specifically, we used the minimum ΨL observation (ΨL,min) of a non-transpiring (bagged) leaf for each species in each stand as an approximation of equilibrated Ψx (Williams & Araujo, 2002; Zhang et al., 2013). While this approach is common (Choatet al., 2010; Zhang et al., 2013; Johnson et al.,2016), a gradient often exists between stem xylem and distal tissues such that ΨL and Ψx are not always equal (Simonin et al., 2015; Johnson et al., 2016; Holtzmanet al ., 2021). However, ΨL and Ψxare often correlated and most similar when hydrologic stress forces stomatal closure (Holtzman et al., 2021). Thus, ΨL,min as determined from bagged leaves is likely a close approximation of Ψx. but may over-estimate true extent of embolism propagation if equilibration between ΨL and Ψx was not achieved during bagging. Nevertheless, estimating native embolism in this manner yielded similar values to those reported in the literature for L. tulipifera (Johnson et al., 2016), A. saccharum (Wheeleret al., 2013), and Quercus species (Sperry & Sullivan 1992; Taneda & Sperry, 2008; Peguero-Pina et al., 2018). We assessed differences in estimated native embolisms across species and stands with a two-way ANOVA, where species and stand age were fixed factors and region was a blocking factor. The relationship between native embolism and degree of isohydricity was then assessed with a least square linear regression between mean estimated native embolism and interquartile range of ΨL of each species in each stand. We excluded IN 35yo data from this analysis because leaf bagging was not possible in this site (see section 5.3).
Second, we investigated hypothesis 3 in the context of a hydraulic safety margin (Ψsafety) (MPa). Safety margins from P12 (Ψsafety, P12) (MPa) and P50 (Ψsafety, P50) (MPa) were calculated as (Domec & Gartner, 2001; Delzon & Cochard, 2014):
\(\Psi_{\text{safety}}=\ \Psi_{L,min}-\Psi_{\text{thresh}}\) (3)
where Ψthresh,(MPa) is mean embolism threshold (e.g., P12 or P50) for the same species in the same stand. A negative Ψsafety suggests a high level of xylem embolism, while a positive Ψsafety suggests a window of safety from critical levels of xylem damage (Johnson et al ., 2016). We then performed a least square linear regression between Ψsafety and the ΨL interquartile range across species and stands. Ψsafety should characterize the difference between the largest xylem water tensions experienced by the plant and the level of water stress leading to a threshold of hydraulic failure (Domec & Gartner, 2001; Delzon & Cochard, 2014). Therefore, we considered whether these analyses were sensitive to hydrologic conditions during the study period, since the observed ΨL, min may underestimate ΨL during extreme exposure to drought (Bhaskar & Ackerly, 2006). We used parametric bootstrapping to quantify a range of slopes of the relationship between ΨL interquartile range and Ψsafety and ΨL interquartile range and estimated native embolism. Specifically, for each unique site-species combination, we created normal distributions of each variable using the observed mean and standard deviation of each metric for each site-species. We then drew 100 estimates of ΨL from the lowest 10% quantile of 50,000 data points drawn from the normal ΨL distribution, and 100 estimates of Ψ­thresh from the middle 60% of 50,000 data points drawn from the normal Ψ­thresh distribution. We experimented with a range of thresholds for the ΨLquantile, ultimately selecting 10% as it produced estimates of ΨL that were at least occasionally lower than the observed minimum ΨL for each site-species. However, most of the simulated ΨL within this quantile were greater than the observed minimum ΨL, such that this is a relatively conservative approach that underestimates the minimum ΨL more than it overestimates it. In future work, other probability distributions, including extreme value distributions (Martínez-Vilatla et al., 2021) could be used instead. Together, these simulated data gave us 100 estimates of Ψsafetythat accommodated uncertainty in both ΨL and Ψthresh and allowed us to estimate 100 unique slopes of the relationship between Ψsafety and the ΨL interquartile range. Again, we excluded data from the IN 35yo site in this analysis.