The study was conducted in the church forest Gelawdiwos (also Gelawdios)—located in the district Dera Woreda, Amhara National Regional State, north-central Ethiopia (11°38′25″N, 37°48′55″E). The forest covers c. 68 ha at c. 2500 meters above sea level. The forest is considered an old-growth, seasonally dry-evergreen Afromontane forest growing on slopes of 10-40° (Assefa et al., 2018; Belay et al., 2018), and is embedded in a mosaic of (degraded) grass- and croplands (Assefa et al., 2022). The climate of the mountainous study area is seasonally dry subtropical, with warm summers and dry winters (Köppen–Geiger classification ‘Cwb’, Peel et al., 2007). Annual precipitations ranges between 1200 and 1600 mm a^-1, heavy rainfall occurring mainly from June to September (Summer) and occasional showers from March to May (Spring); the annual mean air temperature is c. 19°C (Abebe, 2017; Belay et al., 2018). Prevailing soil types are Cambisols; soil texture is 8-9% sand, 39-40% silt and 52% clay, pH(H₂O) of 6.3, and c. 120 mg C g^-1 and c. 11 mg N g^-1 in the top soil (Eshete, 2007; Assefa et al., 2017).

In the Gelawdiwos forest, at least 41 woody species of 29 families have been recorded, dominantly woody angiosperms (Wassie et al., 2010; Assefa et al., 2022). The only woody gymnosperm is Afrocarpus falcatus (syn. Podocarpus falcatus); at present occurring at relatively low density, it was likely more dominant in the past (Assefa et al., 2022). In 2014, the mean timber volume was estimated as 92.4 m³ ha^-1 with an annual increment of 3.5 m³ ha^-1 (Sisay et al., 2017). The density of trees with a diameter at breast height >10 cm was 315 ± 24 ha^-1 and average tree height was 10.2 ± 0.7 m.

Fourteen frequent, native woody species ( Table 1 ) of the Gelawdiwos forest were selected according to (assumed) abundance/relevance, following advice by local expert (Dessie Assefa, personal comment) and literature information (Eshete, 2007). Five species were deciduous and eight evergreen angiosperm trees, and one, in light of its assumed high natural abundance, the gymnosperm A. falcatus. Leaf habit classification of angiosperms followed literature information ( Table 1 ); some species classified here as evergreen have, however, occasionally been described as semi-deciduous (e.g. E. capensis (Tilney et al., 2018) and Prunus africana (Seyoum et al., 2012)), and C. macrostachyus has been described as facultative deciduous (Seyoum et al., 2012). Within the study area, six plots similar in slope, canopy closure and tree height were established in August 2015. At each plot, one individual per species was selected based on following criteria: i) trees were in upper canopy layer, ii) lacked visible injuries, and iii) had no signs of recent disturbance in the surrounding. The minimum distance between plots was 20 m, in most cases c. 50-100 m; the individuals were thus treated as true replicates. Height of the selected trees was measured with a clinometer (Silva Sweden AB, Stockholm, Sweden); a tape was used for assessing diameter at breast height ( Table 1 ).

From each of the 84 tree individuals (n = 6 per species), two coarse root samples and three branches were collected. Root systems where carefully exposed, starting at the tree bole for correct identifications (Rewald et al., 2012), until a coarse root section with a diameter of c. 3-5 mm was discovered—usually at a soil depth of 10-30 cm. The positions of coarse root segments along the flow path (i.e., distance to tips, root order etc.; Freschet et al., 2021) remained unknown as large excavations were not tenable. Sun exposed branches (from the upper canopy) were collected by local tree climbers. For standardization, two-year old branch sections were identified and selected for all species; here, current-year sections were defined as zero-years-old. Subsequently, all distal leaves were collected and stored in plastic bags to prevent dehydration. Excising larger samples first, branch and root samples were recut to a length of 0.35 m, immediately rinsed in tap water and immersed in a silver chloride solution (Micropur, Katadyn, Germany) to reduce bacterial growth (Lübbe et al., 2022). The long branch was kept moist while being transported to the laboratory. Stem cores were collected from 13 tree species (n=2-6 per species) with a hand increment corer (5 mm diameter; Suunto, Vantaa, Finland) to determine wood density. Cores were extracted from the northern sector of the stem at breast height until reaching the center. Because of its very hard wood, no samples could be taken from Teclea nobilis. See Supplementary Table S1 for a list of xylem traits, with acronyms and units, covered by this study.

At the Amhara Regional Agricultural Research Institute (ARARI; Bahir Dar, Ethiopia) root and branch samples were submerged in tap water within 24 h after collection. Central and regularly shaped branch segments were shortened to a length of 50 mm and re-cut with a razor blade, while roots were cut to a length of 100 mm and likewise re-cut with a razor blade. The mean sample diameters were 2.8-4.9 mm in roots and 3.9-5.9 mm in branches (without bark; Supplementary Table S2 ).

Eighteen branch and 12 root samples were measured per species using the XYL’EM device (Bronkhorst, Montigny-Les-Cormeilles, France). Hydraulic conductance was measured following Sperry et al. (1988). Axial hydraulic conductivity (K_h, kg m MPa^-1 s^-1) is given as mass flow rate (m s^-1) per hydrostatic pressure gradient, created through a defined static head in the tubing system, and sample length (Δp l^-1). The basal end of small branches (after debarking) and the distal end of coarse roots were attached to the tubing system. The mass flow rate was determined with a high precision flowmeter (Liquiflow, Instrutec, France; 0.2-10 g h^-1) using degassed and filtered (0.2 µm) KCl solution (10 mM). For measuring the initial K_h (i.e., before flushing, to detect potential plugging of conduits) a hydrostatic pressure of 2 kPa was used for branches, and 1 kPa for roots. Xylem-specific conductivity (   KS hydr , kg m^-1 MPa^-1 s^-1) was calculated by dividing K_h by the xylem area (A_xylem, m²). A_xylem was calculated as the area of an ellipse from two crosswise diameter measurements with a caliper ( ± 0.003 mm) at the basal (branch) or distal (root) end—minus the pith area in branches ( Supplementary Table S2 ). The maximum xylem-specific conductivity (   KS hydr ) was determined after flushing; branch samples were repeatedly flushed at 150 kPa for 30 s, root segments with a pressure of 200 kPa for 60 s until a stable value (k_max) was reached (Martin-StPaul et al., 2014). Length, flushing pressures and time intervals were determined empirically during a pilot phase (data not shown). For A. schimperiana roots, no trustworthy   KS hydr values (i.e. ~3x greater values than in any other species, and a very high variability) could be obtained, probably due to leakages - these measurements were discarded.

Leaf area-specific conductivity (K_L, kg m^-1 MPa^-1 s^-1) of branches was calculated as maximum K_h (after flushing) per distal leaf area (m², one-sided) per branch. See Supplementary Methods for details on leaf area, and K_L.

The coarse root and branch samples used above were labeled and conserved in 60% EtOH. For anatomical analyses, three samples per organ and species were selected, representing mean values of   KS hydr best. Subsequently, semi-thin cross sections (20 µm; at basal sides of branches and distal sides of roots) were cut using a sliding microtome (Reichert-Jung, Austria) and were embedded in Euparal (Carl Roth, Germany). Cross-sections were digitalized using a light microscope (Leica DM 5500B; Leica, Switzerland) equipped with a digital camera (Leica DMC 2900), and analyzed with the software ImageJ v. 1.37 (NIH, USA). Conduit lumen was estimated in a randomly selected radial sector (‘wedge’, opening angle of 20°-70°, bordered by ray parenchyma; (Hacke and Sperry, 2001)), with 57 ± 27 conduits per sector. The conduit density (CD, n mm^-2) was calculated by relating n to the total area of the sector. The lumen fraction (F), i.e. the lumen-to-wood area ratio, was determined by relating conduit lumen area to total woody area of the respective radial sector. Mean conduit diameter (data not shown), hydraulically weighted conduit diameter (D_h, µm), and potential conductivity per xylem area (   KS pot , kg m^-1 MPa^-1 s^-1) were calculated according to Sterck et al. (2008).   KS pot calculations used the Hagen–Poiseuille equation as K S p o t = ((π × ρ × ∑D⁴)/(128 η × A_xylem)), where η is the viscosity (1.002 10^-9 MPa s) and ρ the density of water (998.2 kg m^-3) at 20°C. The Huber value (HV), i.e. the wood-to-leaf-area ratio, was calculated for branches by dividing the downstream leaf area (one-sided) by the area of sapwood (Tyree and Ewers, 1991).

Wood density (WD) was determined as the dry mass per fresh volume (g cm^-3) of branch and coarse root segments used for measuring K_h (n = 12-18), as well as of the stem cores (n = 3-6). The bark was removed prior to determining wood density. Crosswise diameters at both sample ends and sample length were measured to ± 0.003 mm. Xylem volume (cm³) was calculated as the average of the elliptic woody area of both ends multiplied with the sample length, and corrected for the volume of the pith (in branches; Supplementary Table S2 ). The volume of stem cores was estimated using the corer inner dimension (5 mm) multiplied with the length of the core, measured after remoistening the cores for 12 h with tap water. WD_stem of Teclea nobilis could not be determined. The samples were dried at 100°C to a constant weight, and dry mass was determined to 0.1 mg.

Calculations were conducted with the free programming language R, version 4.2.0 (R Core Team, 2019), interfaced with Rstudio (version 2022.2.2.485, RStudio Team, 2016). Normal distribution of each trait was assessed via histograms and qqnorm plots;   KS hydr ,   KS pot , and K_L were log-transformed for parametric tests. Data from different samples of the same tree were averaged per organ to avoid pseudo-replication. Linear models, followed by Post hoc Tukey tests were performed using the R package ‘emmeans’ (Lenth, 2022) to test for differences between species, organs and leaf habits. Letters indicating significant differences were created using the R package ‘multcomp’ (Hothorn et al., 2008). Pearson correlations of each trait between organs, as well as correlations of different traits within organs were calculated using species-wise means, and illustrated using the R package ‘corrmorant’ (Link, 2020). Root-to-branch ratios (R:B) of specific traits were calculated using either tree-wise means (   KS hydr and WD) or species-wise means (for traits with fewer replicates). To investigate the impact of tree height on hydraulic traits, as studied by Olson et al. (2021) linear mixed-effects models (Bates et al., 2015) were fitted. These models were used to examine the correlation between D_h and   KS hydr with tree height, with species considered as a random effect. To calculate P-values for the linear mixed-effects models, we followed the method described in Kuznetsova et al. (2017). The significance level used for all tests was p<0.05 *, partially tendencies (p<0.1; (*)) are reported. Values are given as mean ± standard error (SE) or ± standard deviation (SD) as indicated.

Hydraulically-weighted conduit diameters (D_h) varied >4-fold across coarse roots (p<0.001, Figure 1A ). D_h:root did not differ significantly between deciduous and evergreen angiosperms—holding mean D_h:root of 68 µm and 58 µm, respectively ( Table 2 ). Yet, conduit density (CD_root) of coarse roots tended to be lower in deciduous trees (p=0.099), where 80 conduits mm^-2 were observed, compared to 139 n mm^-2 in evergreen angiosperms ( Figure 1B ); the diameter of the measured roots was similar in both groups (3.82 ± 0.62 mm vs. 4.10 ± 0.44 mm in deciduous and evergreen species, respectively (mean ± SD); Supplementary Table S2 ). Angiosperms’ CD_root varied between 31 mm^-2 in A. schimperiana and 236 mm^-2 in E. capensis; CD_root of A. falcatus was 947 mm^-2. The conduit lumen fraction (F_root) ranged from 6.2% in C. mildbraedii to 36.8% in C. macrostachyus ( Supplementary Table S3 ); mean F_root did not differ between leaf habits (21-22%, Table 2 ). In contrast to roots, the variation in D_h in branches was lower, with species-means between 25.3 µm (T. nobilis) and 60.0 µm (C. macrostachyus) ( Figure 1A ). D_h:branch of deciduous (45.5 µm) were significantly greater than in evergreen angiosperms (34.5 µm; Table 2 ; Figure 1 ); D_h:branch of A. falcatus was 11.1 µm. CD_branch varied widely among species, e.g. 83 mm^-2 in C. macrostachyus to 208 mm^-2 in P. africana ( Figure 1B ). F_branch of angiosperm species varied between 7.4% in C. mildbraedii and 18.2% in E. capensis ( Supplementary Table S3 )—evergreen and deciduous trees held similar F values ( Table 2 ). F_branch in A. falcatus was 26.2%. See Figure 2 ; Supplementary Figures S1 , S2 for example cross sections.

Potential xylem-specific conductivities of coarse roots ( K S : r o o t p o t ) ranged from 1.5 kg m⁻¹ MPa⁻¹ s^-1 (C. mildbraedii), over 9.6 kg m⁻¹ Mpa⁻¹ s^-1 (A. falcatus), to 115 kg m⁻¹ MPa⁻¹ s^-1 (A. dimidiata; Figure 3A ). A. dimidiata branches had a K S : b r a n c h p o t of 6.5 kg m⁻¹ MPa⁻¹ s^-1, the average value among angiosperms was 9.1 kg m⁻¹ MPa⁻¹ s^-1. Lowest K S : b r a n c h p o t were estimated for A. falcatus (1.5) and the evergreen T. nobilis (2.2 kg m^-1 MPa^-1 s^-1). K S : b r a n c h p o t was significantly greater in deciduous as compared evergreen angiosperms (p=0.045, Table 2 ); however, some species showed a high deviation from leaf habit averages ( Figure 3A ). Empirically determined conductivity K S : r o o t h y d r of angiosperms ranged between 1.4 kg m^-1 MPa^-1 s^-1 in C. mildbraedii, to 17.4 kg m^-1 MPa^-1 s^-1 in C. macrostachyus ( Figure 3B ). In general, K S h y d r was significantly lower in branches (2.3 kg m⁻¹ MPa⁻¹ s^-1) compared to coarse roots (9.5 kg m⁻¹ MPa⁻¹ s^-1; p<0.001, Table 3 ). Noteworthy, coarse roots and branches of e.g. C. mildbraedii had similar K S h y d r . K S h y d r was strongly positively correlated with K S p o t (p<0.001, R²=0.67, Supplementary Figure S3 ); K S h y d r was on average c. 25% of K S p o t .

Leaf area-specific conductivities (K_L) of angiosperms ranged from 313 kg m⁻¹ MPa⁻¹ s⁻¹ in C. aurea to 16,368 kg m⁻¹ MPa⁻¹ s⁻¹ in A. schimperiana ( Supplementary Table S3 ). K_L tended to be greater in deciduous than in evergreen angiosperms (p=0.085; Table 2 ). Lowest Huber values (HV) were observed in A. falcatus (0.696 10^-4 m² m^-2), followed by much greater values in evergreen (13.9 10^-4 m² m^-2) and deciduous angiosperms (26.4 10^-4 m² m^-2, Supplementary Table S3 ).

In angiosperms, average D_h, F, K S p o t , and K S h y d r in coarse roots were significantly greater than in branches; CD tended (p=0.096) to be lower in roots ( Table 3 ). For example, D_h of angiosperm branches (39 µm) were on average 37% lower than those in roots (61 µm; p<0.001). These differences were similar between deciduous (D_h:root = 68 µm, D_h:branch = 46 µm) and evergreen (D_h:root = 57 µm, D_h:branch = 35 µm) angiosperms. In accordance, F_root was on average 9.3% greater than F_branch (p=0.002). Average root-to-branch ratios (R:B) of K S h y d r were comparable, with 5.1 for deciduous and 5.8 for evergreen angiosperms ( Figure 4 ; Supplementary Table S4 ). Including A. falcatus, linear models applied on species-means revealed significant positive correlations between root and branch traits: D_h (p=0.026), CD (p<0.001), F (p=0.013), K S p o t (p=0.04), and a trend for K S h y d r (p=0.07; Supplementary Table S5 ). When calculated for angiosperms only, however, only marginally positive correlations were found ( Figure 4 ; Supplementary Table S5 ). Pearson coefficients indicated clear correlations between some anatomical and hydraulic traits of angiosperms ( Figure 5 ). In particular, K S : r o o t h y d r showed a significant positive correlation to F_root. K S : b r a n c h h y d r was significantly correlated to D_h:branch. Correlations between the hydraulic traits of different organs were scarce and only trends were detected ( Figure 5 ). Significant correlations were not observed between the hydraulic properties K [ S ] [ h y d r ] and D[_h] of branches and roots and tree height ( Supplementary Table S6 ).

Differences in wood densities (WD) were found ( Table 4 ). For example, C. mildbraedii held the densest coarse roots (0.77 g cm^-3) while S. abyssinica had a WD_root of 0.38 g cm^-3. WD_branch ranged between 0.50 and 0.78 g cm^-3 in angiosperms ( Table 4 ), significantly denser than WD_root ( Table 3 ). In contrast to WD_branch, species-means of root and stem WD were highly positively correlated (p=0.002, Figure 5 ). WD_stem was significantly greater in evergreen compared to deciduous trees (p=0.031, Table 2 ). WDs of roots, stem and branches were significantly negatively correlated to K S : r o o t p o t ( Figure 5 ). Negative correlations were found between WD_branch and K S : b r a n c h h y d r , as well as between WD_stem and K S : b r a n c h p o t , D_h:branch, and F_branch.

Our results revealed a strong decline in hydraulic efficiency in most Ethiopian highland tree species, with measured conductivity in coarse roots 5.1- or 5.8-times greater than in similar-sized branches in deciduous and evergreen angiosperms, respectively. Although hydraulic efficiency is largely determined by vessel diameter, the vessel lumen fraction (F) was closely associated with hydraulic conductivity K_S in roots and branches. Narrower conduits were (partially) compensated for by higher conduit densities (CD) in branches of a majority of species (Ewers et al., 2007). The results on hydraulic tapering follow patterns already observed by e.g. Maherali et al. (2006), whereas in their study K S h y d r of roots was on average approximately 6-times greater than in branches. Similar, Plavcová et al. (2019) recently revealed significantly higher mean xylem cell cross-sectional areas in roots than in branches of temperate trees. Reducing hydraulic efficiency along the hydraulic flow paths has been observed in many temperate (Lübbe et al., 2022), and tropical tree species (Schuldt et al., 2013; Kotowska et al., 2015), but also some woody species in Mediterranean or semi-arid systems (Martínez-Vilalta et al., 2002; Pratt et al., 2007; Longui et al., 2012; Santini et al., 2018). The phenomenon has been related to interrelated gradients of e.g. turgor pressure (Woodruff et al., 2004; Woodruff and Meinzer, 2011) and cambial age (Li et al., 2019; Rodriguez-Zaccaro et al., 2019) along the flow path. The substantial difference of hydraulic efficiency between roots and branches may be explained by a trade-off between hydraulic efficiency and safety against cavitation, whereas roots were thought to be more vulnerable to cavitation and closer to their hydraulic limit (Martínez-Vilalta et al., 2002). However, this view is increasingly challenged (Gleason et al., 2016; Santini et al., 2018). Recently, Lübbe et al. (2022) reported that xylem embolism resistance did not differ significantly between roots and branches in three out of four temperate tree species. However, in contrast to continuous vessel tapering along the hydraulic pathway from roots in branches observed in temperate and semi-arid systems, a hump-shaped vessel diameter variation has been found in mature trees from tropical moist forests. Largest vessel diameters were present in the coarse root and stem xylem, while smaller vessels were found in small roots and branches-the reason for this observation remains speculative (Schuldt et al., 2013; Kotowska et al., 2015). In our data set, species-specific root-to-branch ratios of c. 1 were found between similar-sized roots and branches. For example, hydraulically weighted conduit diameter (D_h) in roots of C. mildbraedii were only marginally greater than in branches ( Figure 2 ) and remained largely similar in E. capensis, pointing towards species-specific tradeoffs between hydraulic and other (e.g. mechanical) traits. Plavcová et al. (2019) indicated that the mechanical properties of roots and stems within species are rather independent of each other. While Schuldt et al. (2013) suggested that vessel diameter in above- and belowground organs may be related to segment diameter in trees growing in a perhumid tropical environment, a ΔD_h of 30-40% between similar-sized root and branches of this study clearly points towards additional effects in the studied highland forest. However, different tissue differentiation (‘secondary growth’) due to a potentially greater age of sampled coarse root segments compared to 2^nd-year branches cannot be excluded—roots are notoriously hard to classify (Freschet et al., 2021). In sum, our data provides convincing evidence that the coarse root xylem holds a greater hydraulic efficiency than the branch xylem in subtropical tree species of the seasonally dry Ethiopian highlands.

We had hypothesized that drought-deciduous and evergreen angiosperms differ in their hydraulic traits and have different relationships between root and branch hydraulic traits. Deciduous species invest less mass per unit leaf area while evergreen species have higher leaf mass per area, constituting different costs and benefits of leaf construction (Eamus, 1999; Powers and Tiffin, 2010; Souza et al., 2020). Both leaf habits often differ in physiological leaf traits (Mediavilla and Escudero, 2003; Ishida et al., 2006) and overall growth strategies (Tomlinson et al., 2014). Recently, however, Ribeiro et al. (2022) found no difference in leaf traits between evergreens and deciduous species in neotropical dry forests. As leaf shedding in seasonally dry environments may present an relative effective mechanism (‘hydraulic fuse’) to prevent or limit damage to a hydraulically vulnerable xylem and overall plant desiccation (Wolfe et al., 2016), we expected that different leaf habits may go along with different hydraulic traits along the flow path. In accordance with (Choat et al., 2005, for branches), our anatomical comparison showed significant greater D_h in branches of drought-deciduous trees and tendencies towards greater conduit density (i.e. ‘hydraulic redundancy’) in evergreens’ roots and branches. While Hoeber et al. (2014) did not find significant differences between root and branch traits of leaf habit groups in the seasonally dry tropics of Costa Rica, their trees were only c. 6 years old and large changes in hydraulic strategies have been reported with ongoing tree maturation (Osazuwa-Peters et al., 2017). However, while we hypothesized differences in root-to-branch ratios between leaf habits, i.a. greater root-to-branch ratios in evergreen species, no such differences were observed. This indicates that drought-deciduousness does not translate into a less marked difference between root and branch hydraulic efficiencies compared to evergreen angiosperms. While one could then speculate on a tight coordination between coarse roots and branch traits, this is not supported by our data—which indicated rather weak correlations of hydraulic traits between lateral woody organs. Instead, the higher values of potential K_S and leaf area-specific conductivity in deciduous species may allow for an increased carbon gain, and the lower wood density of stems of deciduous trees would enable a high volumetric growth during a shortened growing season (Eamus and Prior, 2001). While Markesteijn et al. (2011) pointed out that drought sensitivity in terms of critical xylem tension is not necessarily lower in deciduous species, Fu et al. (2012) reported greater embolism resistance in evergreen species of Asian tropical dry forests.

While the absence of vulnerability measurements does not allow clarifying this for the studied species, our findings demonstrate that deciduous and evergreen species show marked differences in hydraulic strategies but also highlight a great variability within leaf habit groups, similar to earlier findings in moist tropical forests (Kraft et al., 2008; Hoeber et al., 2014). Here, potential reasons may lay in the unstable precipitation patters in the region, where global climatic events such as El Niño can cause severe rainfall deficits in the Ethiopian summer (Gleixner et al., 2017), potentially resulting in year-to-year differences in most successful hydraulic strategies. Additional traits such as rooting depth (Hasselquist et al., 2010), canopy size and leaf-to-sapwood area (Mencuccini et al., 2019; Rosas et al., 2019), and particular additional leaf traits (e.g., degree of isohydry; (Hartmann et al., 2021); length of leave-less periods (Borchert et al., 2002)), are needed to further unravel species-specific patterns.

Wood density (WD) is a key plant property, related to mechanical and (eco-)physiological performance (Martínez-Cabrera et al., 2009; Ziemińska et al., 2013). The relationship between wood density and the underlying anatomical traits is shaped by multiple, interrelated functions of xylem—water transport, storage, and mechanical support (Chave et al., 2009; Messier et al., 2017). A strong negative relationship of WD with lumen fraction of stems and partially across organs has e.g. been demonstrated for different temperate and Mediterranean taxa (Preston et al., 2006; Zanne et al., 2010; Lübbe et al., 2022). This trade-off is a corner-stone of the wood economic spectrum, where denser wood is expected to provide better mechanical support while typically involving restrained sap transport capacities (Messier et al., 2017). However, studies on tropical trees did not find a close linkage between WD_stem and vessel characteristics (Fan et al., 2012; Schuldt et al., 2013; Hietz et al., 2017), because WD_stem is more controlled by fiber than by conduit structure (Martínez-Cabrera et al., 2009; Ziemińska et al., 2013). Similar, Fortunel et al. (2014) found no relationship between WD and vascular traits at the branch level, suggesting no tradeoff between WD_branch and hydraulic efficiency in trees of the moist tropics. In accordance, our pairwise correlation analyses revealed that WD_branch had a strong negative relationship with empirically determined K S h y d r , but no significant correlations with traits related to either conduit size or frequency. In contrast, angiosperm roots held strong negative relationship between WD_root and K_S, mediated through the negative effect of WD_root on D_h and lumen fraction F. The higher variation in F_root as compared to F_branch might explain the good correlation with WD_root (R² = 0.61). The correlation between F_branch and WD_branch revealed a lower goodness-of-fit, with R² being 0.11. WD_root of trees in the studied ecosystem may thus to be both determined and restricted by the cumulative conduit volume in the stele, although the scatterplots illustrate the large dependency of WD_root on D_h and not conduit density in angiosperms. In contrast to our hypothesis, the hydraulic efficiency of coarse roots can thus be predicted reasonably well from WD due its large dependency on conduit properties. WD_root of the studied species may thus not depend strongly on fiber properties, as reported for Mediterranean, temperate and tropical tree species (Martínez-Cabrera et al., 2009; Schuldt et al., 2013; Fortunel et al., 2014; Plavcová et al., 2019).

Earlier, Nakagawa et al. (2016) reported that root wood traits were highly correlated with corresponding stem wood traits in South-East Asian tropical trees. Based on these findings, we had hypothesized that WD of coarse roots, stems and branches are generally related. Indeed, our analyses showed that WD_stem was highly negatively correlated to root hydraulic conductivity and significantly positively to WD_root. However, while Fortunel et al. (2014) reported that WD was strongly correlated between branches and roots in 113 Amazonian rainforest tree species across a large environmental gradient, we found no correlations between root and branch WD or between WD_branch and WD_stem. In contrast to our hypothesis, our result thus indicate fundamental differences between above- and belowground xylem: fibers may play a greater role in shaping WD in branches, likely because of their role for mechanical support, while vessels may play a greater role in shaping WD_root where the soil matrix provides mechanical support (Pratt et al., 2007). However, as the arrangement of xylem traits and the total fraction of tissues (i.e. fiber wall fraction/lumen fraction, fraction of parenchyma) can also influence wood density (e.g. Ziemińska et al., 2013; Lachenbruch and McCulloh, 2014), this has to be considered in future studies.