Data analysis
All values are reported as a plant-level mean ± 1 SD. The thermal safety margin (TSM) of each plant was calculated with Tcrit or T50 as the upper threshold. From this value, we subtracted the mean maximum temperature at this site, 32 °C, and then subtracted either the mean observed leaf-air temperature differences measured with the handheld IR sensor (∆Tleaf), the mean observed canopy-air temperature differences measured with UAV thermal imagery (∆Tcan), or the modelled leaf-air temperature differences calculated under common conditions from the traits, leaf width, stomatal conductance, and leaf inclination angle (∆Tmod).
To assess whether there was provenance differentiation in ∆Tleaf we used a linear mixed effects model including fixed effects for ‘provenance’ and random effects for measurement days ‘doy’ as well as individual leaves nested within individual plants ‘unique_id/leaf’. The underlying data was based on individual spot measurements of Tleaf minus measured Tair at that time. Models were fitted to each species separately and pairwise comparisons between provenances were assessed using ‘emmeans’ (Lenth, 2022).
To assess whether there was provenance differentiation in leaf traits (width, angle, conductance, thermal time constant), leaf-air temperature differences (∆Tmod, ∆Tcan) and thermal tolerance (Tcrit and T50), as well as the thermal safety margins (TSM), we performed a two-way unpaired student’s t-test on each species separately. We used plant averages as a single observation so that all tests had balanced designs with n = 6 plants for each group. Homogeneity of variance was tested using the levene test and normality was tested using the Shapiro test.
To determine whether provenance-differentiations in traits were a result of systematic differences in plant height between provenances, we performed two analyses. First, we tested for within-canopy differences in leaf traits and thermal tolerance in the lowland provenance ofC. australe and H. novo-guineensis. To do so we performed paired t-tests on 5 replicate plants per group. The leaf trait, LMA, did not meet assumptions of normality so data was log transformed for analysis. In addition, we examined the relationship between modelled and observed leaf temperatures, and plant height using linear regression with either Tleaf, or T­mod as the response variable, and plant height as an explanatory variable. All analyses were performed using R version 4.2.2.