[CI] = 0.46-0.57) and it declined with tree height. We also used the allometric relationship between field-based measurements of soil-to-leaf hydraulic conductance and laboratory-based measurements of leaf hydraulic conductance to compute the average fR leaf for 19 tree samples, which was 0.40 (95% CI = 0.29-0.56). The in situ technique produces a more accurate descriptor of fR leaf because it accounts for dynamic leaf hydraulic conductance. Both approaches demonstrate the outsized role of leaves in controlling tree hydrodynamics. A larger fR leaf may help stems from loss of hydraulic conductance. Thus, the decline in fR leaf with tree height would contribute to greater drought vulnerability in taller trees and potentially to their observed disproportionate drought mortality.

drought response, hydrodynamic modelling, leaf hydraulic conductivity, plant hydraulics, plant water relations, whole-tree hydraulic conductance

Transpiration from terrestrial plants is a fundamental component of Earth's water cycle. Transpired water flows under tension through a hydraulic continuum that extends from soil to leaves within the soil-plant-atmosphere continuum (SPAC) (Sperry et al., 2002;Venturas et al., 2017). When water flow through the SPAC is impeded, such as during droughts, stomata close, slowing transpiration, carbon dioxide uptake and plant growth (Anderegg et al., 2020;Schwalm et al., 2017). Plants invest in roots, stems, and leaves presumably to maximize fitness, including growth, but multiple tradeoffs limit the capacity for water uptake and transport. For example, mechanical strength and water storage tradeoff with hydraulic transport capacity within woody tissues (Pratt et al., 2021;Scholz et al., 2011). Likewise, within whole plants, variation in branching architecture has opposing effects on light interception and SPAC hydraulic efficiency (Smith et al., 2014) and allocation to leaves increases both productivity and susceptibility to hydraulic failure during droughts (Trugman et al., 2019).

Several models predict how plant traits and environmental conditions influence SPAC fluxes including transpiration (e.g., Christoffersen et al., 2016;Sperry et al., 2002). These SPAC models are becoming integral components of Earth system models that predict feedbacks between ecosystems and the atmosphere (Fisher et al., 2018). Central to SPAC modelling is how hydraulic resistance (i.e., the pressure difference per water flux) is partitioned among components along the hydraulic continuum, that is, rhizosphere, roots, stems, and leaves (Sperry et al., 1998). In trees, most of the SPAC pathlength is in roots and stems, whereas leaves have a relatively short pathlength. Yet, leaf hydraulic resistance (R leaf ) can be relatively high, as water passes through small xylem conduits and living cells. Consequently, the fraction of total SPAC hydraulic resistance (R total ) within leaves, or fR leaf (=R leaf /R total ), may be outsized.

To date, the most comprehensive analysis of fR leaf among plant species is that of Sack et al. (2003). They combined data from several studies, including herbs, woody seedlings and saplings, and mature trees and shrubs. A standardized major axis (SMA) fit through the logtransformed values of leaf hydraulic conductance (K leaf ) as a function of total-SPAC hydraulic conductance (K total ) had a slope of 1.21 (95% confidence interval [CI] = 0.99-1.43), an intercept of log(4.2), and r = 0.91 (Sack et al., 2003). Since resistance is the inverse of conductance, this result suggests that leaves consistently contribute about 25% (i.e., 1/4.2) of the total SPAC hydraulic resistance (i.e., fR leaf = ca. 0.25). This scaling relationship suggests a general convergence among plants in hydraulic architecture and function: leaves contribute a disproportionately high amount of hydraulic resistance for their relatively short pathlength in the SPAC, acting as hydraulic bottlenecks.

However, among the plants analysed by Sack et al. (2003), the mature trees and shrubs do not fit the overall SMA as well as the herbs, seedlings, and saplings. Indeed, we fit an SMA through the 11 samples of mature trees and shrubs within the data set of Sack et al. (2003) and found a slope of 1.01 (95% CI = 0.51-1.99), intercept of log(2.22), and r = 0.25. This analysis suggests that fR leaf is generally higher (i.e., ca. 0.45) and more variable (i.e., lower r) in mature trees and shrubs than in herbs, seedlings, and saplings. Similarly, studies that have inferred fR leaf in situ from pressure differences between soil, stems, and leaves on apricot trees, orange trees, loblolly pine trees, and seasonally dry tropical forest trees have found fR leaf of 0.3-0.8 (Alarcón et al., 2003;Brodribb et al., 2002;Domec et al., 2009;Moreshet et al., 1990). fR leaf may also decline with tree size. For example, von Allman et al. (2015) predicted that fR leaf decreased from ca. 0.40 to 0.18 among maple and oak trees with trunk diameters 5-30 cm.

Variation in fR leaf is expected to affect plant performance. As soil and air dry, reduced plant water potential (Ψ) is associated with loss of hydraulic conductance (K) in roots, stems, and leaves, which can be described with Ψ 50 , that is, the Ψ at which an organ loses 50% of K (Venturas et al., 2017). In trees, the loss of stem K (K stem ) beyond critical thresholds is associated with drought mortality (Adams et al., 2017). The 'hydraulic segmentation' hypothesis of Zimmerman (1983) predicts that higher fR leaf buffers stem Ψ (Ψ stem ) during droughts, protecting trees from K stem loss. Similarly, the 'vulnerability segmentation' hypothesis of Tyree et al. (1991) predicts that when leaf Ψ 50 is high relative to stem Ψ 50 , K leaf loss slows transpiration and buffers Ψ stem , protecting trees from K stem loss.

Here, we illustrate these hypotheses using a simple model of water transport that assumes plants regulate transpiration to avoid SPAC hydraulic failure (Sperry et al., 2016). We ran several simulations of a soil dry down while varying hydraulic segmentation (i.e., fR leaf ) and vulnerability segmentation (i.e., relative values of leaf vs. stem Ψ 50 ) (Figure 1). During the dry down, Ψ stem increased with fR leaf (i.e., the slope in Figure 1e), illustrating the hydraulic segmentation hypothesis. Likewise, for any given fR leaf , trees with vulnerability segmentation (leaf Ψ 50 > stem Ψ 50 ) maintained higher Ψ stem (i.e., the difference in elevation between curves in Figure 1e). In nature, both mechanisms are likely to occur to varying degrees among tree species. Both mechanisms cause reduced K total and transpiration during water deficit (Figure 1a,c,g,h), which represents a tradeoff between hydraulic safety and productivity. However, when leaves and stems are equally vulnerable (leaf Ψ 50 = stem Ψ 50 ), modelled K total and transpiration were insensitive to fR leaf (open symbols in Figure 1g,h). In this case, a tradeoff for high fR leaf is unapparent.

F I G U R E 1 Effects of varying fR leaf within the hydraulic transport model of Sperry et al. (2016). The model was parameterised with a single soil layer that dried from 0 to -5 MPa while vapour pressure deficit was 1 kPa. Root and stem Ψ 50 were set to -2 MPa. Leaf Ψ 50 was set to -2.0 (i.e., no vulnerability segmentation (no VS)) and -1.0 MPa (i.e., VS). fR leaf was set to 0.1, 0.5, and 0.9 (i.e., low, mid, and high fR leaf ). The legend in panel (a) is a key to the six combinations of fR leaf and VS for all panels. Everything else was held constant, particularly the initial K total , which here represents the total investment in the plant hydraulic system. Thus, higher fR leaf represents a relative increase in allocation to hydraulic conductance in roots and stems and a relative decrease in allocation to hydraulic conductance in leaves. As soil water potential declined, transpiration (a), stem water potential (b), and K total (c) declined at rates that varied with initial fR leaf and VS. Within plants, fR leaf changed during the dry down as leaves, stems and roots lost hydraulic conductance at varying rates (d). Panels (e-h) show a single point during the dry down, when soil water potential was -1 MPa. Trees with VS maintained higher stem water potential than trees without VS and, with or without VS, stem water potential increased with fR leaf (e). Similarly, trees with VS maintained higher leaf water potential than trees without VS, whereas effect of varying fR leaf on leaf water potential small (f). With VS, K total (g) and transpiration (h) declined with higher fR leaf .

Given that fR leaf likely affects tree performance and that its variation among trees is not well understood, we assessed fR leaf in trees from various biomes and habitats. We estimated fR leaf from previously reported and new measurements of pressure differences in situ within tropical and temperate trees. We also followed Sack et al. (2003) in exploring the K total -K leaf allometry with an expanded data set of trees. We compared methods for estimating fR leaf in trees and tested whether fR leaf varies among biomes, clades (conifers vs.

angiosperms), tree sizes, and with soil moisture. We also addressed the hydraulic segmentation hypothesis by testing whether trees with higher fR leaf maintained higher Ψ stem .

According to Darcy's Law, the water flux (F) through a system is equal to the pressure difference between endpoints divided by its hydraulic resistance. In transpiring trees, around midday, the hydraulic resistance between soil and leaf (R total ) is:

where Ψ soil and Ψ leaf_md are soil and midday leaf water potential, respectively, and hρ g w is the hydrostatic gravitational pressure, where h is tree height, ρ w is the density of water and g is gravitational acceleration (Domec et al., 2009;Whitehead, 1998).

The F term in Equation 1 can be measured with several techniques. Transpiration from individual leaves can be measured with a porometer or portable leaf gas-exchange system to estimate leaf-area-specific R total (MPa mmol -1 m 2 s). However, estimating F with leaf-level measurements is generally problematic because of the difficulty in matching boundary layer conditions between measurement chambers and ambient conditions, where they are highly variable temporally within and among tree canopies (Pearcy et al., 1989). F can also be measured within roots, trunks, and branches with sap flow probes (e.g., Granier, 1985). These measurements are generally scaled by cross-sectional area of the measured organ or by the whole tree. They can be converted to give leaf-areaspecific R total through allometric relationships or by directly measuring whole-tree leaf area.

One assumption in Equation 1 is that midday F is in a steady state (i.e., F is at equilibrium within the SPAC). This assumption does not hold when water stored within the plant is in flux. However, the contribution of stored water to F is generally negligible around midday as water stored within plant tissues enters the transpiration stream in the morning and is recharged with soil water in the afternoon and night (Goldstein et al., 1998;Loustau et al., 1998;Maherali & DeLucia, 2001).

Ψ soil in the rooting zone is often estimated with predawn leaf water potential (Ψ leaf_pd ) as:

Equation 2 relies on the assumption that, except for the hydrostatic gravitational pressure difference, Ψ leaf_pd is in equilibrium with Ψ soil . This assumption has theoretical support if low nighttime vapour pressure deficit (VPD) and closed stomata combine to drive F to zero. However, when R total is very high, nighttime VPD is high, or nighttime stomatal conductance is high, then Ψ leaf_pd and Ψ soil do not equilibrate on diurnal cycles (Bucci et al., 2005;Donovan et al., 2003;Kavanagh et al., 2007).

Darcy's Law can also be used to assess the combined hydraulic resistance of the SPAC components located proximally to the leaf, that is, between the soil and terminal branches (R soil-branch ). R soil-branch can be measured similarly to R total by replacing Ψ leaf_md with midday branch water potential (Ψ branch_md ) Then,

Ψ branch_md is commonly measured with the pressure chamber technique on leaves that have been put into Ψ equilibrium with their proximal branches by stopping transpiration. This is achieved by sealing leaves in plastic bags and protecting the bags from solar radiation with reflective foil, usually for at least 1 h before the leaves are detached from the branch for measurement. This stops transpiration by placing the leaves in vapour-saturated air and closing stomata (Begg & Turner, 1970).

Since resistances in series are additive, that is, R total = R leaf + R soil-branch , R leaf can be calculated by combining Equations 1, 2, and 3 to give

Then, dividing Equation 4 by Equation 1 gives fR leaf . Since F is assumed to be in steady state within the SPAC, the equation simplifies (Equation 5), producing an estimate of fR leaf without the need for F measurements and laboratory based K leaf measurements.

(5)

We measured fR leaf in trees at two sites in central Panama where the Smithsonian Tropical Research Institute operates cranes that each access the canopy of ca.8000 m 2 of the forest. One site is in a seasonally dry tropical forest on the Pacific side of the isthmus in the Parque Natural Metropolitano that averages 1850 mm of rainfall annually with a dry season from December through April (Pivovaroff et al., 2021). The other site is in a wet tropical forest in the Bosque Protector San Lorenzo on the Caribbean side of the isthmus that averages 3300 mm of rainfall annually with a dry season from January through March (Pivovaroff et al., 2021).

We measured nine trees at each site, each of a different species (Supporting Information: Table S1). We selected canopy trees that were exposed to full sunlight and supported few or no lianas. Diameter at breast height (dbh) and height of the focal trees ranged 19-132 cm and 17-39 m, respectively (Supporting Information: Table S1). datapoints from these trees were outliers (i.e., >3 SD from the mean; Supporting Information: Figure S1) and were excluded from the tree-level mean fR leaf .

On four of the measurement campaigns at each crane site, we also measured in situ transpiration with a portable photosynthesis machine (LI-6400XT, LI-COR Inc.) on 6-8 leaves that were also measured for Ψ leaf_md between 10:00 and 15:00 h. We set cuvette conditions to closely match ambient conditions. Further details of the transpiration and Ψ leaf_md measurements are described in Wu et al. We combined these with the Ψ leaf_pd measurements to calculate leafarea-specific K total (i.e., inverse of Equation 1).

We measured fR leaf on three 5-year-old Pinus taeda (loblolly pine) trees planted in full sun at the Louisiana State University AgCenter Burden Experimental Station, located in Baton Rouge, Louisiana, USA, on 4 February 2021. The trees were 167-191 cm height and 2.0-2.1 cm dbh. On each tree, we measured Ψ leaf_pd , Ψ leaf_md , Ψ , and Ψ branch_md as described above. Each Ψ was the average of five leaves. We used the measurements to compute fR leaf for each tree with Equation 5. Each leaf that was measured for Ψ leaf_md was measured for transpiration with a portable photosynthesis machine (LI-6400XT, Li-Cor Inc.) with cuvette conditions set to closely match ambient conditions. P. taeda leaves grow in fascicles of three needles (generally, rarely two, or four needles). For each transpiration measurement one fascicle was placed in the cuvette (6400-02B, Li-Cor Inc.). Afterwards, the fascicles were imaged with a flatbed scanner and their diameter was measured with ImageJ. Leaf area was calculated from fascicle diameter following Blazier et al. (2018) and transpiration was scaled to leaf area. Then the inverse of Equation 1 was used to calculate K total for each tree.

We measured fR leaf on three Persea borbonia (red bay) trees growing in the forest understory at the margins of bayheads (i.e., stream bottoms) and pine forests at the H. G. Lee Memorial Forest in Washington Parish, Louisiana, USA, on 30 March 2021. The trees were 1.5-1.7 m tall and 1.4-1.5 cm dbh. On each tree, we measured Ψ leaf_pd and Ψ leaf_md , Ψ branch_pd , and Ψ branch_md as described above.

Each Ψ was the average of five leaves. We used the measurements to compute fR leaf for each tree with Equation 5 and took the specieslevel mean.

We used Web of Science, Google Scholar, and citations within publications to search for studies that reported Ψ leaf_pd , Ψ leaf_md , and Ψ branch_md to compute fR leaf . We excluded shrubs, tree seedlings (defined here as <1 m height), and potted plants. Only publications that concurrently measured all three components of Equation 5(i.e., Ψ leaf_pd , Ψ leaf_md , and Ψ branch_md ) were included. For studies that reported values in figures but not in tables or text, we extracted the values using Web Plot Digitizer (Rohatgi, 2020). We took mean values of fR leaf for species within each study unless the trees were measured under various conditions within a study (e.g., different habitats or experimental treatments), then we took the mean for each condition.

Combined with data that we collected (described above), we compiled 101 values of fR leaf for 99 tree species (Supporting Information: Table S1). Among the samples, 36 included a measurement of Ψ branch_pd . We used these to test the assumption of Ψ branch_pd-Ψ leaf_pd equilibrium by fitting an SMA through the points (Supporting Information: Figure S2). Disequilibrium between these values would indicate a pressure difference that violates the assumption of Ψ soil -Ψ leaf_pd equilibrium in Equation 2. Samples from temperate forests, tropical seasonal forests, and tropical rainforests did not vary from the 1:1 line, indicating equilibrium. However, samples from tropical savannas were above the 1:1 line, indicating a predawn pressure difference. Indeed, Bucci et al. (2005), reported Ψ soil -Ψ leaf_pd disequilibrium in this system due to nocturnal transpiration. Therefore, we excluded the 11 tropical savanna fR leaf values (Supporting Information: Table S1) from our analyses.

To assess the allometric relationship between K total and K leaf with a data set that was expanded from Sack et al. (2003), we searched the literature as described above for studies that reported K total and K leaf .

We found 19 paired values of K total and K leaf from 12 tree species across 5 studies (Supporting Information: Table S2). Note that K leaf was measured in the laboratory on detached leaves, independently of K total . Since K leaf is generally considered a species-level trait (Scoffoni & Sack, 2017), we also paired K leaf and K total from disparate studies that measured only one of them on the same species. This produced a much larger sample size, 50 additional K leaf -K total pairs (Supporting Information: Table S2). However, compared to the K leaf -K total pairs from single studies, the pairs from disparate studies had higher variance (i.e., lower R 2 ) and a higher intercept (Supporting Information: Figure S3). K leaf -K total pairs from single studies are likely more reliable since K leaf can vary within species among sites (Taneda et al., 2016), so we present only the analyses of K leaf -K total pairs from single studies.

Before testing for differences in fR leaf among biomes and between tree clades, we noted that fR leaf was negatively related to tree height.

Therefore, to test for differences in fR leaf among biomes and between clades, we used ANCOVA with tree height as a covariate. Seven samples did not have values of tree height and were excluded from these analyses (Supporting Information: Table S1). To test whether fR leaf was associated with Ψ branch_md , Ψ leaf_pd , and Ψ leaf_md , we used Spearman's rank correlation analysis. Spearman's was used instead of Pearson's correlation analysis because variables were nonlinearly related. These correlations were confounded by the component variables appearing in both the x and y-axes, a statistical nuisance called the shared variables problem. To account for this, we compared the observed Spearman's correlation coefficients (r s ) to null correlations created with randomisation tests (Jackson & Somers, 1991). We randomly sampled without replacement the component variable from all observations and, along with the other two components, calculated fR leaf with Equation 5. We then calculated r s between the component variable and fR leaf with the randomized data. This procedure was repeated 1000 times to create the null correlation. We then used z-tests compute the p value (⍺ = 0.05) for the observed r s in relation to the r s obtained in the randomisation test.

To assess fR leaf among the 19 paired values of K total and K leaf (Supporting Information: Table S2), we followed Sack et al. (2003).

We fit an SMA through log 10 -transformed values and interpreted the mean fR leaf as the inverse of 10 to the power of the intercept.

Additionally, we used the smatr package in R (Warton et al., 2012) to test for differences in the intercept, slope, and position along a common axis among biomes and tree clades.

Among the samples measured with the in situ method using Equation 5, the mean fR leaf was 0.51 (standard deviation [SD] = 0.24; 95% CI = 0.46-0.57) and fR leaf declined with tree height (R 2 = 0.30, P = 8e-8; Figure 2). When four samples of conifer trees that were outliers in terms of height in the temperate forest (Figure 2) were excluded from the regression analysis, the result of declining fR leaf with tree height remained (R 2 = 0.23, P = 4e-6). Although tree height differed among biomes (Figure 3a), fR leaf did not differ among biomes 3b). fR leaf was strongly correlated with Ψ branch_md (r s = 0.61) and, to a lesser extent, Ψ leaf_pd (r s = 0.41), while Ψ leaf_md was not correlated with fR leaf (r s = -0.04) (Figure 4a,c,e). The correlation between fR leaf and Ψ branch_md was not any stronger than expected by random chance (Figure 4f) while the correlation between fR leaf and Ψ leaf_pd was significantly greater than expected by random chance (Figure 4b).

The SMA that was fit on paired K leaf -K total values showed an isometric relationship with a mean fR leaf of 0.40 (95% CI = 0.29-0.56)

F I G U R E 2 leaf as a function of tree height. Points represent species or species by treatment combinations (Supporting Information: Table S1). The line represents a least squares linear regression fit through all the points, fR leaf = 0.70 (SE = 0.041) -0.011 (SE = 0.002) × tree height. R 2 = 0.30. Shading represents the 95% confidence interval.

F I G U R E 3 Distribution measured tree heights (a) and fR leaf values (b) among forest biomes. Boxes extend to the 25th and 75th quartiles and are bisected by the median. Bars extend to the most extreme data point that is no more than 1.5 times the length of the box away from the box. Points represent samples of species or species by treatment combinations (Supporting Information: Table S1). Symbols are drawn as in Figure 2.

(Figure 5). Among biomes, the SMA slope did not differ (likelihood ratio = 3.5, p = 0.06), but the intercept did (Wald = 10.1, p = 0.002).

The intercept was larger for the tropical seasonal forest biome (0.57, 95% CI = 0.35-0.79; fR leaf = 0.27, 95% CI = 0.16-0.45) than for the temperate forest biome (0.30, 95% CI = 0.13-0.48; fR leaf = 0.50, 95% CI = 0.34-0.74). Comparisons between conifers and angiosperms found no differences in slope (likelihood ratio = 0.06, p = 0.81) or intercept (Wald = 0.96, p = 0.33), but conifers had lower K leaf and K total values than angiosperms along the common SMA line (Wald = 42.6, P = 6e-11).

Our multibiome assessment of in situ measurements found a mean fR leaf of 0.51 among trees (Figures 3,4). This value for trees is double the commonly cited value of 0.25 that Sack et al. (2003) presented for K total -K leaf allometry across multiple plant forms. However, it is relatively similar to the fR leaf of 0.45 obtained from the 11 trees and shrubs within the Sack et al. (2003) data set (see Introduction) and the fR leaf of 0.40 obtained from the 19 trees in our expanded K total -K leaf allometry data set (Figure 5). Overall, these results suggest that fR leaf is considerably higher among trees than seedlings and herbs. fR leaf may be higher in trees because their roots and stems contain proportionally more secondary xylem than those of seedlings and herbs. Anatomical features in secondary xylem enable higher hydraulic conductance than in primary xylem (Evert, 2006) and proportionally higher hydraulic conductance in roots and stems would increase fR leaf .

Among trees, the in situ method produced a higher mean fR leaf than the K total -K leaf allometric method. This is likely because the insitu method inherently accounts for dynamic K leaf by measuring leaf and branch pressure differences simultaneously. In contrast, the K total -K leaf allometric method based on measurements of excised organs assumes that K leaf measured in the laboratory is equal to K leaf when K total is measured in situ. However, K leaf is dynamic on diurnal et al., 2009, 2018;Sack & Holbrook, 2006;Simonin et al., 2015;Zhang et al., 2016). Our K total -K leaf allometric analysis and that of Sack et al. (2003) used maximum K leaf obtained under laboratory conditions. This approach lowers the apparent fR leaf if in situ K leaf is lower than laboratory maximum K leaf , which is common. For this reason, the in situ method likely produces a more accurate descriptor of fR leaf .

fR leaf obtained with the in-situ method was highly variable among samples, with SD = 0.24. This result contrasts with the expectation that fR leaf is consistent among trees (e.g., De Cáceres et al., 2021;Wolfe et al., 2016). The wide range of fR leaf values also calls into question the rule of thumb that hydraulic resistance is consistently partitioned between roots and shoots in a 50-50 split (e.g., Sperry et al., 1998). Our modelling exercise demonstrates how fR leaf can vary within trees in response to environmental conditions and how this response depends on the leaf hydraulic vulnerability relative to other SPAC components (Figure 1d). Measured values of leaf and stem Ψ 50 show a wide range of relative vulnerability; for example, among 63 angiosperm species the R 2 between leaf Ψ 50 and stem Ψ 50 was only 0.16 (Scoffoni & Sack, 2017). Hydraulic conductance of all SPAC components (i.e., soil, roots, stems, and leaves) is highly variable in response to environmental conditions (Domec et al., 2006(Domec et al., , 2010(Domec et al., , 2021;;Johnson et al., 2009). Therefore, it is likely that fR leaf is not a reliable species-level trait, but rather is dependent on current and past environmental conditions experienced by individual trees.

We found that taller trees tend to have lower fR leaf (Figure 2). This result is consistent with the results of von Allmen et al. (2015), who modelled fR leaf in oak and maple trees and found that fR leaf decreased as stem diameter increased. For maple and oak, respectively, they found fR leaf decreased from 0.42 to 0.19 and from 0.36 to 0.17 among trees with trunk diameters 5 to 30 cm. This pattern could result through several processes. If K leaf remains constant with tree height and K total declines with tree height due to the increasing pathlength, this would result in a decrease of fR leaf with tree height.

Hydraulic conduit tapering can partially compensate for the effect of increasing pathlength but is unlikely, in itself, to prevent a decline in K total (Savage et al., 2010;Zaehle, 2005). Indeed, among the trees in our analysis for which we had height and K total data (Supporting Information: Table S2), we found no relationship between the two traits for angiosperms but a declining K total with tree height for conifers (Supporting Information: Figure S4).

All else being equal, trees with higher leaf area to sapwood area ratios would have lower fR leaf since leaves are arranged on branches analogously to resistors in parallel. In other words, as the leaf area to sapwood area ratio increases, hydraulic conductance in roots and stems is partitioned into smaller portions for each leaf, which corresponds to a lower fR leaf . Yet, taller trees tend to have lower leaf area to sapwood area ratios (McDowell et al., 2002), which would have an effect in the opposite direction of our result of decreasing fR leaf with tree height. However, as von Allmen et al. ( 2015) noted, as long as K total declines with height faster than the leaf area to sapwood area ratio, then fR leaf will decline with height. It is also possible that K leaf declines with tree height (Zhang et al., 2009), which would counteract the effect of K total decline with tree height. The relationship between height, fR leaf , and these competing influences is likely highly variable among trees.

fR leaf was positively correlated with Ψ branch_md (Figure 4e). This result is predicted by the hydraulic segmentation hypothesis (Zimmermann, 1983) and our modelling exercise (Figure 1e). It suggests that fR leaf can influence drought performance by preventing K stem loss. However, because Ψ branch_md and fR leaf are not independent in our analyses (Equation 5), the correlation is susceptible to spuriousness (Jackson & Somers, 1991), which was confirmed with a randomisation test (Figure 4f). Therefore, our results do not directly support the hydraulic segmentation hypothesis. To do so would require other experimental approaches in which fR leaf is assessed independently of Ψ branch buffering. A study of excised shoots found F I G U R E 5 Allometric relationship between total hydraulic conductance (soil to leaf; K total ) and leaf hydraulic conductance (K leaf ). Each circle represents a tree species or species by treatment combination (Supporting Information: Table S2). Circles and triangles represent samples from temperate and tropical seasonal forests, respectively. Filled symbols represent conifers and open symbols represent angiosperms. The dashed 1:1 is shown for reference. The solid line represents a standardized major axis fit through all points.

Log 10 (K leaf ) = 1.03 (95% CI = 0.75-1.40) × K total + Log 10 (2.47; 95% CI = 1.77-3.45). R 2 = 0.62. Note that both axes are log scaled. CI, confidence interval.

that drought-tolerant trees have proportionally higher R leaf than drought-sensitive trees (Drake et al., 2015). Further, among aridenvironment shrubs, K leaf is positively correlated with the branch hydraulic safety margin (i.e., stem Ψ 50 minus Ψ branch_md ) (Pivovaroff et al., 2014). However, direct evidence that high fR leaf protects K stem is lacking.

The finding that fR leaf declined with Ψ leaf_pd (Figure 4a) constrasts with our model prediction that fR leaf remains relatively constant or increases as soil dries (Figure 1d). However, fR leaf would be expected to decline with Ψ leaf_pd if a nonleaf component contributed significantly to R total and was more vulnerable than leaves. Roots and the rhizosphere likely follow this pattern (Bourbia et al., 2021;Rodriguez-Dominguez et al., 2018). The root-soil interface, root cortex and rhizosphere are generally very vulnerable to drying and can become hydraulic bottlenecks unless plants compensate by investing in sufficient absorptive root area (Cuneo et al., 2016;Lo Gullo et al., 1998;North et al., 2008;Rodriguez-Dominguez et al., 2018;Sperry et al., 2016). We explored how variation in rhizosphere hydraulic resistance interacts with fR leaf to influence Ψ stem buffering by comparing our simulations in Figure 1 with simulated soil dry downs where the average rhizosphere hydraulic resistance was set to 50% of average R total (the default is 5%; Sperry et al., 2016) (Supporting Information: Figure S5). These simulations predicted reduced fR leaf during initial Ψ soil decline (Supporting Information: Figure S5d), consistent with observations (Figure 4a).

Together, these results suggest that the influence of the root-soil interface, root cortex and rhizosphere on R total increases as soil dries.

In theory, upstream hydraulic vulnerability moderates fR leaf buffering of Ψ stem , but in any case, higher fR leaf has higher Ψ stem buffering capacity (compare Supporting Information: Figure 1e andS5e).

If hydraulic segmentation acts to protect stems, an important implication of declining fR leaf with tree height (Figure 2) is that larger trees have less protection. Loss of K stem is a key predictor of tree mortality during droughts (Adams et al., 2017). Large trees tend to suffer higher mortality rates than small trees during droughts (Bennett et al., 2015). In addition to other factors, including the higher VPD that larger trees experience (McDowell & Allen, 2015), lower fR leaf may contribute to the trend for higher drought mortality in larger trees.

Our literature review and field measurements combined data sets of Ψ leaf , Ψ branch , K total and K leaf that were taken with varying techniques and sample sizes within and among trees and species (Supporting Information: Tables S1,S2). Ψ leaf and Ψ branch were often averaged over several days before they were input into Equation 5 to calculate fR leaf . These discrepancies may have contributed to the high variation that we found in fR leaf (Figures 2,3). More standardized measurements with high replication are needed to better quantify fR leaf and its variation within and among trees in association with plant traits and in response to environmental conditions. Attention is also warranted to verify the assumptions in K total and fR leaf measurements: that Ψ leaf_pd can quantify Ψ soil in the rooting zone and that F is at steady state at midday. Our result of high mean fR leaf suggests that leaf hydraulics play an even more outsized role in tree water relations than typically considered. Therefore, research focused on leaf hydraulics is likely to improve understanding of whole-plant water relations.