Climate analysis
In each garden, most traits were significantly correlated with population provenance climate (Table 3), supporting our third hypothesis of strong associations between phenotype and climate. A single axis (PC1) explained 95.8% of the variation in provenance climate and was influenced primarily by four temperature-related climate variables (degree-days above 5 °C, degree-days below 18 °C, degree-days above 18 °C, and summer heat-to-moisture index (Wang et al. 2012)), plus elevation. Populations sourced from areas with higher temperatures, lower precipitation, lower elevation, and longer growing seasons had higher PC1 scores. Positive correlations with PC1 indicate that trait values are higher in those hotter provenance populations, while negative correlations with PC1 mean that trait values are higher in the colder provenance populations (Fig. 3). Phenology traits showed the strongest correlations with provenance climate (r = -0.75 to -0.77 for bud flush in Yuma and Agua Fria, and r = 0.40 to 0.47 for bud set in all gardens, Table 3). Here, the negative correlation for bud flush indicates that the populations from hotter source climates (with higher PC1 scores) had earlier spring flush dates, at least when growing in the two hotter gardens. The positive correlation for bud set in all gardens indicates that populations sourced from colder areas (lower PC1 scores) had later bud flush dates in the spring. Although lower in magnitude, the correlation between climate and SLA was also consistent in direction. In both the hot and the cold garden, SLA was higher (meaning leaves were thinner) in populations from hot climate origins.
In contrast with the consistent direction of correlations between home climate and trait value for phenology and SLA, growth traits were more likely to show garden-dependent relationships between population origin and population performance (Table 3, Fig. 3). In general, tree height and basal diameter acted as indicators that overall tree performance is consistent with local adaptation, with hot, southern populations growing larger in the hottest Arizona garden, and northern, cold populations growing larger in the coldest Utah garden.
We also found that traits more strongly correlated with climate had stronger evidence for selection driving their divergence. The full model including the three gardens and all traits showed a significant relationship between the strength of the trait-climate correlation and QST (Fig. 5, P = 0.0012,R2 = 0.57, F(1,13) = 16.88). Neither garden nor trait identity had significant effects on the mean QST value. When gardens were modeled separately, the two warmer gardens showed significant relationships between trait-climate correlations and QST (Yuma: P = 0.059, R2adj = 0.66; Agua Fria:P = 0.028, R2adj = 0.79), while the coldest garden in Canyonlands did not (P = 0.571, R2adj = 0.18). This result shows that traits with the highest differentiation among populations are those where that trait variation is most strongly correlated with the climate, indicating a strong role for selection by climate in driving differentiation. Interestingly, this relationship was masked in the cold garden, where southern populations did not flush out their leaves early in the spring as they do in their home climate, thereby reducing the amount of population differentiation in that trait compared to what we could detect in the warmer gardens (Figs 3-5).