Effects of hydraulic architecture and spatial variation in light on mean stomatal conductance of tree branches and crowns


B. E. Ewers. Fax 307 766 2851; e-mail: beewers@uwyo.edu


In a Pinus taeda L. (loblolly pine) plantation, we investigated whether the response to vapour pressure deficit (D) of canopy average stomatal conductance (GS) calculated from sap flux measured in upper and lower branches and main stems follows a hydraulically modelled response based on homeostasis of minimum leaf water potential (ΨL). We tested our approach over a twofold range of leaf area index (L; 2–4 m2 m−2) created by irrigation, fertilization, and a combination of irrigation and fertilization relative to untreated control. We found that GS scaled well from leaf-level porometery [porometry-based stomatal conductance (gs)] to branch-estimated and main stem-estimated GS. The scaling from branch- to main stem-estimated GS required using a 45 min moving average window to extract the diurnal signal from the large high-frequency variation, and utilized a light attenuation model to weigh the contribution of upper and lower branch-estimated GS. Our analysis further indicated that, regardless of L, lower branch-estimated GS represented most of the main stem-estimated GS in this stand. We quantified the variability in both upper and lower branch-estimated GS by calculating the SD of the residuals from a moving average smoothed diurnal. A light model, which incorporated penumbral effects on vertical distribution of direct light, was employed to estimate the variability in light intensity at each canopy level in order to explain the increasing SD of both upper and lower branch-estimated GS with light. The results from the light model showed that the upper limit of the variability in individual branch-estimated GS could be attributed to incoming light, but not the variation below that upper limit. A porous medium model of water flow in trees produced a pattern of variation below the upper limit that was consistent with the observed variability in branch-estimated GS. Our results indicated that stems acted to buffer leaf- and branch-level variation and might transmit a less-variable water potential signal to the roots.


sapwood-to-leaf area ratio


cumulative leaf area density


empirical coefficient


vapour pressure deficit


transpiration per unit leaf area


actual transpiration of a branch


maximum EV


porometry-based stomatal conductance


canopy average stomatal conductance


maximum GS, GSref, GS at 1 kPa vapour pressure deficit


sap flux per unit xylem area


hydraulic restriction coefficient


extinction coefficient for an ellipsoidal leaf distribution


leaf area index


leaf area density


light restriction coefficient


above-canopy photosynthetic photon flux density


photosynthetic photon flux density in the ith canopy layer


maximum Qi


root area index


projected leaf area per unit dry mass


leaf absorbtivity of Q0


sensitivity of GS to D


volumetric soil moisture


volumetric soil moisture normalized by root area


fraction of incident beam radiation


water potential


leaf water potential


zenith angle


clumping factor of leaf distribution


Stomatal conductance (gs) is needed in models of carbon and water flux at scales ranging from leaf to globe (e.g. Running & Coughlin 1988; McMurtrie, Rook & Kelliher 1990; Bonan 1991; Foley et al. 1996; Landsberg & Waring 1997; Sellers et al. 1997; Lai et al. 2000, 2002; Baldocchi, Wilson & Gu 2002; Schäfer et al. 2003); however, uncertainties remain regarding the parameterization of gs (Mackay et al. 2003; Ewers et al., in press). Measurement in gas exchange chambers provide the standard for estimating gs at the leaf level (Jarvis 1995), and sap flux and eddy covariance measurements provide estimates of gs at scales ranging from branches to ecosystems (Köstner et al. 1992). Stomatal conductance is incorporated into models of biosphere-atmosphere water, energy and carbon exchange using two approaches to canopy specification: (1) multilevel canopies that explicitly account for vertical gradients in gs conductance and other gas-exchange parameters, and therefore require detailed information for use; and (2) bigleaf models that use canopy average stomatal conductance (GS) and are simple to apply but suffer from errors in averaging over processes that affect photosynthesis and transpiration non-linearly (Raupach & Finnigan 1988). A hybrid approach distributes GS vertically based on modelled light profile and uses both profiles for calculating an effective canopy mean leaf internal-to-external CO2 concentration (DePury & Farquhar 1997; Schäfer et al. 2003).

Estimates of GS based on sap flux are at a spatial scale intermediate to leaf chamber estimates of gs and eddy covariance-based estimates of GS (Köstner et al. 1992; Granier et al. 1996; Martin et al. 1997; Saugier et al. 1997; Pataki et al. 1998a; Ewers & Oren 2000). Sap flux, typically measured at the base of a tree, provides high-frequency measurements of GS and captures the response to temporal variations in environmental conditions. Moreover, sap flux of individual branches allow estimates of branch-estimated GS, representing an additional refinement of the scale between leaf-level estimates of gs and main stem-estimated GS, thus preserving the spatial variance of gs both horizontally and vertically (Oren et al. 1998; Ewers & Oren 2000; Martin et al. 2001). Furthermore, calculations of GS from the basally measured fluxes in branches are less influenced than measurements in the main stem by the effect of tissue water storage.

The external drivers that generate vertical variation in gs and GS are D and Q0. Gradients in D can be quite large in broadleaved and in coniferous forests that are poorly coupled to the atmosphere and have small boundary-layer conductances compared to gs (Jarvis & McNaughton 1986; Magnani et al. 1998). In coniferous forests supporting relatively low-canopy L and coupled strongly to the atmosphere, such as Pinus taeda L. plantations, gradients in D are undetectable (Ewers & Oren 2000). The gradient in average Q0 through the canopy leads to complimentary gradients in leaf structural characteristics and nitrogen, and thus in the photosynthetic capacity of leaves (Oren et al. 1986a; Ellsworth & Reich 1993; Palmroth & Hari 2001; Gunderson et al. 2002; Lai et al. 2002; Meir et al. 2002). However, high-frequency temporal and spatial variation in photosynthesis and gs can be expected at one canopy level because of frequent changes in direct Q0 originating from self-shading among and within shoots (Oechel & Lawrence 1985; Stenberg 1998). As intra-crown shading reduces gs and transpiration of some needles or branches, more water may immediately be available to support higher gs of better-illuminated branches (Shinozaki et al. 1964; Brooks et al. 2003). This effect has been shown for leaf blades of wheat (Buckley & Mott 2002a). Although results of investigations on whole trees are mixed (Brooks et al. 2003), hydrodynamic models can be employed to assess the effects of within-crown variation in light on potential variation in gs.

Determining gs response to environmental variables is further complicated by the potential autocorrelation between the variables (e.g. Q0 and D) and interactive effects on physiology. One approach to unraveling the interactive effects is to reduce data to parameters that describe a known relationship between stomatal conductance and D within sequential and narrow intervals of light and soil moisture (Schäfer, Oren & Tenhunen 2000). Plants that maintain transpiration at a rate that does not permit minimum leaf water potential to drop below a threshold well above the value that would lead to excessive cavitation (isohydric; Tardieu & Davies 1993) show sensitivity of GS to DGS/ΔlnD) that is proportional to GSmax (or its proxy GSref, defined as GS at a reference value of D = 1 kPa; Oren et al. (1999). This proportionality can be captured by


where the slope −δ (dGS/dlnD) is the sensitivity of GS to D. The slope of the correlation between −δ and GSref has been shown empirically and theoretically to be near 0.6 for a wide range of plants that homeostatically regulate the minimum water potential to prevent excessive cavitation (Oren et al. 1999; Ewers, Oren & Sperry 2000; Schäfer et al. 2000; Ewers et al. 2001b; Oren & Pataki 2001; Lai et al. 2002; Wullschleger et al. 2002; Addington et al. 2004; Ewers et al. 2005; Ewers, Mackay & Samanta 2007). Such a semi-empirical model (Eqn 1) can be utilized to reduce large data sets to a few representative coefficients, two for each combination of, for example, light and water availability, and to quantify differences in stomatal conductance response to light and soil moisture (Ewers et al. 2001a).

Pinus taeda (L.) (loblolly pine) grows across a wide range of soil and environmental conditions, and shows considerable hydraulic plasticity reflecting differences in climatic and edaphic conditions (Teskey et al. 1987; Ewers et al. 2000; Hacke et al. 2000; Maier et al. 2002). We assessed the coherence in scaling from gs to branch- and main stem-estimated GS and the consistency of stomatal conductance response to D at these scales using Eqn 1 and a data set of water fluxes in a P. taeda stand subjected to irrigation, fertilization and a combination of irrigation and fertilization for comparison with untreated control.

Our present study builds on previous work by the investigators (Ewers et al. 2000; Ewers et al. 2001a). We present this work briefly further to set the stage for our hypotheses. Previous studies show that fertilization doubles leaf area index (L) relative to control trees without an increase in root surface area. In response to the drastic decrease in root-to-leaf area ratio, fertilized trees operate under a greater water stress and develop fine roots with low xylem vulnerability to cavitation, permitting the extraction of soil moisture to lower concentrations. The tradeoff for this acclimation is a lower maximum root hydraulic conductance in fertilized trees. Thus, when moisture of the coarse, sandy soil of the site was increased by continuous irrigation in all treatments, GS among treatments ranked as fertilized < control = irrigated < combination (irrigated and fertilized), but minimum leaf water potential was similar in all treatments, indicating a homeostatic, isohydric behaviour. Indeed, the response of GS to D followed the 0.6 ratio between −δ and GSref (Eqn 1). In this study, we hypothesized that the above ranking would be maintained at all scales, and that the response to D would be similar regardless of scale. Furthermore, given the spatial variability in light within the canopy, we expected that the residuals of individual branch-level GS from the average branch-level GS at each point in time would increase with light intensity, reflecting the spatial variability in light, coupled with redistribution of water flow away from poorly illuminated branches. We then tested the relative contribution of light response and redistribution of water to branch-level GS variability through the use of a hydrodynamic crown model (Bohrer et al. 2005).


The Southeast Tree Research and Education Site (SETRES; Albaugh et al. 1998) was established in 1992 in the Sandhills of North Carolina (34°48′N 79°12′W) on an infertile, well drained, sandy, siliceous, thermic Psammentic Hapludult soil (Wakulla series). Annual precipitation averages 1210 mm with occasional growing season water deficits.

Sixteen 50 × 50 m treatment plots (including a 10 m wide buffer) separated by 10 m untreated buffers were established in a mixed families stand of North Carolina Piedmont P. taeda planted in 1984 in a 2 × 3 m spacing. Treatments were a 2 × 2 factorial combination of nutrition and water additions. Nutrient treatments have been maintained since March 1992 and water addition since April 1993. Nutrient treatment consists of optimal nutrition (fertilized) defined as maintaining nitrogen concentration of 1.3% in upper canopy foliage with phosphorus, potassium, calcium and magnesium balanced with nitrogen levels. Boron is also added to maintain foliar concentrations above 12 ppm. Foliar nutrient status was monitored monthly and fertilizer applied annually to meet target values. Water addition was made to keep available soil moisture between field capacity and 40% of available soil water in the upper 0.50 m of the soil profile corresponding to 30 mm soil water in that layer as measured with time domain reflectometry (TDR). For more details on nutrition and water treatments, see Murthy et al. (1996) and Albaugh et al. (1998).

From 25 July to 8 August 1998, we continuously irrigated a portion of a measurement plot in all four treatments to compare GS at similar maximum water content. This study was presented in detail in Ewers et al. (2000, 2001a), but a brief summary is included here. The continuous irrigation raised soil moisture to levels above the standard irrigation treatment, and above the highest that rains can naturally maintain, even temporarily, in this fast draining soil. A sprinkler was placed such that it could reach three trees within each treatment on which sap flux and leaf water potential (ΨL) were measured (see further). In all treatments, the maximum flow of water permitted a minimum coverage of 2.5 m radius that included the immediately neighbouring tree of each of the three individuals plus halfway to the next tree. We analyzed the irrigated trees as replicates, and thus used = three within each treatment.

Plant biomass measurements

We calculated projected L using allometric relationships, adjusting for seasonal trends based on canopy surface analyzer measurements (LAI-2000; Li-Cor Co., Lincoln, NE, USA; see Albaugh et al. 1998; Ewers et al. 1999). Whole tree sapwood to leaf area ratio (AS : AL) was calculated from L and sapwood area determined from breast-height cores (Ewers et al. 1999). All sided root area index (R) was calculated from root biomass and specific root area (Ewers et al. 2000). Roots were excavated to 1.9 m in a 1 × 1 m pit incrementally to 0.15, 0.30, 0.50, 0.90, 1.1, 1.3, 1.5, 1.7 and 1.9 m. Roots from each depth interval were sorted into < 1, 1–2 and 2–5 mm size classes and oven dried at 65 °C. Total R was determined from a surface area-to-mass relationship constructed from subsamples of each size class.

Each branch measured for sap flow (see further) was harvested to determine leaf and sapwood areas. Nine fascicles throughout the branch were subsampled to determine specific leaf area (SLA). The length and width of each needle in the fascicle were measured to determine the projected area. Needles were then oven dried to 65 °C to determine mass. Sapwood areas of each branch were measured at the midpoint of the sap flux measurement gauge, approximately 15 cm away from the first fascicles. Because there was no observable heartwood in branches, sapwood area was calculated by subtracting the bark and pith cross-sectional area from the total branch cross-sectional area.

Sap flux and environmental measurements

The analyses in this study were made on three irrigated trees, a subset of the eight trees measured in previous studies located in the buffer zone of each of the four treatment plots in one block, and measured for sap flux using constant heat sensors (Granier 1987; Ewers et al. 1999; Ewers et al. 2000). Previous work detailed the circumferential and radial trends in each treatment (Ewers et al. 1999; Ewers & Oren 2000) used to scale between individual sap flux sensors and stem-level sap flux. Sap flow was also measured with compensating heat Kucera-type sensors (Cienciala et al. 1994; Ewers et al. 2000; Ewers et al. 2002) on an upper (∼ 7 m) unshaded branch and a lower (∼ 2 m) shaded branch of the three trees in each treatment.

A relative humidity and temperature probe (Vaisala HMP 35C; Campbell Scientific, Logan, UT, USA) was positioned at a height of 7 m in the centre of each treatment and provided data for calculating D (Goff & Gratch 1946). Q0 was measured above the canopy (Li-190s, Li-Cor). Soil volumetric water content (θ) was measured using automated TDR probes with 6 cm long steel rods (Theta Probe; Delta-T Devices, Cambridge, UK) at 0.05, 0.1, 0.25, 0.5, 1.0 and 2.0 m depth. A weighted θ by root area (θR) was calculated using the root area profile to 1.9 m (Ewers et al. 2000).

Xylem flux and all environmental sensors were sampled every 30 s, and 15 min average values were logged (CR21X, Campbell Scientific).

Light transmittance modelling

We determined a weighting scheme for branch-estimated GS that separated the canopy into two layers based on leaf area distribution using leaf biomass and leaf area data from Albaugh et al. (1998, 2004) and Q0 attenuation. Light in our stands decreased rapidly and nearly linearly with height from the canopy top, before changing to a more curvilinear and slow decrease at greater depths. We partitioned the canopy to upper and lower levels using the midpoint of the zone in which the linear decrease occurred, ∼ 3 m from the top. Upper branch-estimated GS measurements were weighted by the tree leaf area in the upper 3 m of the canopy and lower branch estimates of GS by the rest of the crown leaf area (Table 1), producing a weighted mean canopy GS. Direct and diffuse beam Q0 transmission through the canopy was computed based on the adaptation of the Campbell & Norman (1998) model by Schäfer et al. (2003). Briefly, this model treats sunlit and shaded portions separately to estimate Q0 absorbed at each canopy level and incorporates self-shading and penumbral effects that are particularly common in conifers (Stenberg 1998; Stenberg et al. 2001). The fraction of incident beam radiation from a zenith angle penetrating through the canopy is given by

Table 1.  Canopy leaf area index (L), weight of sap flux calculated stomatal conductance (GS) in the upper and lower branches as determined from light transmission modelling; specific leaf area (SLA) in upper and lower branches and the sapwood-to-leaf area ratios (AS : AL) in upper and lower branches in the four treatments
TreatmentL (m2 m−2)GSUpper weightsGSblLower weightsSLAUpper (cm2 g−1)SLALower (cm2 g−1)AS : ALUpper (cm2 m−2)AS : ALLower (cm2 m−2)
  1. Letters indicate significant differences within a column and values in parentheses are one SE of the mean (= 3).

Control2.00.050.9531.8a (1.4)39.3a (1.3)2.3a (0.6)16.5a (1.1)
Irrigated2.10.050.9532.0a (1.5)38.7a (1.0)1.8ab (0.4)16.7a (1.5)
Fertilized3.80.030.9732.3a (1.5)39.1a (1.1)1.5b (0.2)18.7b (1.0)
Combination4.00.100.9031.5a (1.9)39.5a (1.0)2.6a (0.4)21.0b (0.9)

where α is the leaf absorptivity for Q0; Kbe(Π) is the extinction coefficient for an ellipsoidal leaf distribution; a1 is the cumulative LAD integrated from the canopy top, and Ω is the clumping factor of leaf distribution (Gower, Kucharik & Norman 1999). We used an α of 0.83 and an Ω of 0.5 based on a measurement on loblolly pine (Thérézien et al. in press). The model was used to determine the mean and SD of irradiance on branches at each level in the canopy. The SD was determined by quantifying the probability distribution of Q0 on needles and then by dividing the distribution into the same number of probability classes as the number of branches monitored for sap flux. This approach assumes that clumping and the penumbral effect generates variation in light among branches rather than generates large variation among needles on branches without affecting the mean light on the branch.

Calculations of EL and GS

To calculate transpiration per unit leaf area (EL), sap flux per xylem area (JS) was combined with sapwood-to-leaf area ratio (AS : AL; Oren et al. 1998; Pataki et al. 1998a; Ewers & Oren 2000). We calculated GS from EL and D, at both the branch and main stem level, using the simplification of the inversion of Penman–Monteith model as suggested by Monteith & Unsworth (1990) and evaluated for P. taeda by Phillips & Oren (1998). The simplified calculation was permitted because in all treatments, (1) D is close to the leaf-to-air vapour pressure deficit, that is, boundary-layer conductance is high; (2) D has no vertical gradient through the canopy; and (3) the amount of water stored in the trees above the probes is negligible (Ewers & Oren 2000). Relative errors in GS, caused by instrument limitations, were kept to < 10%, by limiting GS calculations to D > 0.6 kPa (Ewers & Oren 2000). Previous work at the site had shown that fertilized trees had significantly lower EL and GS than the other treatments (Ewers et al. 1999; Ewers et al. 2000; Ewers et al. 2001b), so data from control, irrigated and combination were analyzed together to increase the tree and branch sample size from 3 to 9.

Leaf gas exchange measurements

In 1998, on 24 July (before the continuous irrigation experiment) and 29 July (during the experiment), porometric measurements of gs (Li-Cor 6400, Li-Cor) were made on the branches fitted with the Kucera-type sensor. One fascicle from each branch was excised and the cut end placed in water for measurements at 8.5, 10.0, 13.0 and 15.5 h on each day. Preliminary measurements at the site had shown that there were no differences between excised and attached needle gas exchange, under constant chamber conditions, when measurements were restricted to less than 15 min after excision. Environmental conditions inside the chamber were matched with the previous 15 min average ambient CO2, D and Q0. For the lower branches, Q0 was set at the estimated level of attenuated Q0 based on the Beer–Lamber law, extinction coefficient of 0.5 and treatment L (Table 1).

Hydrodynamic model

To test the relative contribution of light and water redistribution to branches on variability of branch-estimated GS, we simulated sap flux with the finite elements tree crown hydrodynamics (FETCH) model (Bohrer et al. 2005). FETCH uses a modified set of Richards' equations for porous media flow, similar to that in Chuang et al. (2006), to resolve water movement in the crown branch system. The model assumes, as boundary conditions, a specified water pressure at the top of the root-system and canopy-top environmental conditions (wind speed, air temperature and humidity, Q0 and D). Water potential (Ψ) of roots was assumed to be near saturation (−8 cm) because the trees were continuously watered. Observations for mean wind conditions during July and actual Q0 and air temperatures and humidity (and thus D) at 15 min intervals during the days of the experiment were used as environmental conditions for simulations. To calculate the maximal potential transpiration (EVmax) at each vertical level of the tree crown, the FETCH model uses a first-order turbulent closure routine, assuming no hydraulic limitations on water flux from stomata (i.e. stomata are fully open) and that the canopy is composed of tree crowns with the same structure as the modelled one. This maximal potential transpiration is used to drive water out of the branch system, along leaf supporting ‘end branches’. FETCH also estimates the actual amount of water leaving an end branch (i.e. actual transpiration –EV) by accounting for Ψ in the branch, and the light conditions at the branch, modelled as described earlier.

As Ψ in the branch decreases, the conductivity of the xylem is decreasing along a Weibull shaped curve often described as the xylem vulnerability curve (Sperry et al. 2002). As discussed earlier, to limit the risk of cavitation, the stomata control the rate of water loss to the atmosphere in order to regulate water pressure. The stomata of many tree species were observed to reduce conductance by 90% when xylem conductivity was reduced by 10% (Cruziat, Cochard & Ameglio 2002). The stomata response to Ψ, therefore, leads to a ‘hydraulic restriction’ coefficient (HC). This coefficient is a fraction between 0 and 1 following a Weibull-shaped function driven by the branch water tension, with an inflection point at −2 MPa derived from minimum leaf Ψ and vulnerability curve measurements on these trees (Ewers et al. 2000).

The model also accounts for stomatal response to light. When Qi (i.e. the light modelled at the surface of a ‘statistical’ branch) is low, demand for CO2 is low and stomata would partially close in order to minimize water loss. A ‘light restriction’ coefficient (LC; a fraction between 0 and 1) is calculated for EVmax assuming that stomata are fully open when Qi levels are maximal [Qimax (mmol m−2 s−1)], and water pressure in the branch is not limiting. Thus,


where b is an empirical coefficient; = 0.0011 at the canopy top and 0.0023 below 6 m in the canopy based on the light model.

The shape and structure of the model tree is based on an L-system fractal tree (Bohrer et al. 2005), with four lateral branches at each split. The tree was 12 m high (highest tree height; Ewers et al. 2000) with first-order branches splits at 2.4, 4.8, 8.4 and 9.6 m above the ground, each axis symmetric around the stem. Second-order branches split at 1.2 and 2.4 m from the branch base of the bottom two first-order branches. The LAD of end branches is assumed proportional to their length. The simulations here assumed that all branch elements were of the same length.

A light level per branch was updated every 15 simulated minutes. The distribution of light levels at each update time was obtained from light estimates in the canopy at the same day and time as the simulation time using the light model described earlier. Light levels at each end branch and time step were randomly drawn from a binned distribution (six bins for the canopy branches below 6 m and 12 bins for the canopy above 6 m).

FETCH simulated the transpiration and sap flux from 9 trees per day, from 0800 to 1400 h, during 10 d of the experiment. The model time step was 1 s, and Ψ and HC were updated every time step. Environmental variables and LC were updated every 15 min. Sap flux at the base of the first-order branches was calculated by the model and converted to branch-level GS using the same approach as with measured flux. The SD of GS estimated from branches at the bottom (2.4 m) and top (8.4 m) was calculated as the SD of residuals resulting from the difference between the actual branch GS and the smoothed branch GS from a moving average of three time points (45 min).

Statistical analyses

Statistical analyses were performed using SAS procedures general linear model (GLM) and MIXED (version 8.0; SAS Institute, Cary, NC, USA). When data were collected repeatedly on the same tree, repeated measures analysis using MIXED was used. Non-linear curve fits were performed in SigmaPlot (version 8.0; SPSS, San Rafael, CA, USA). We constructed boundary lines of GS response to D by partitioning D into 0.2 kPa ranges, determining the mean of GS and SEs within each 0.2 kPa range, and estimated the boundary line from a regression through the combined points that fell above the mean GS plus one SE within each D range (Ewers et al. 2000; Schäfer et al. 2000). This boundary line represents the best possible conditions for stomatal conductance at any given value of the independent environmental variable, controlling for the effects of other variables (Martin et al. 1997). Although we show means of treatments for clarity, all analyses were performed on the raw data obtained from measurements of fascicles, branches or trees. Separation of mean values were performed using the LSMEANS, Tukeys statement at an α value of 0.05 in the MIXED procedure of SAS.


Modelling attenuation of Q0

GS estimated from upper and lower branches was weighted by L in the upper and lower crown for scaling to GS estimated from branches for comparison with GS estimated from main stem measurements (Table 1). Based on the light model, approximately half of the light attenuated throughout the canopy was absorbed within the top 3 m of the canopy in both high and low Q0 conditions regardless of whether L was low (control and irrigated) or high (fertilized and combination; Fig. 1, Table 1). Because L was virtually identical in stands receiving different water supplies but similar nutrient supply (Table 1), Fig. 1 shows L and Q0 attenuation profiles for only one representative treatment from each fertilization level (control and fertilized). In each canopy level, SLA was similar in all treatments (> 0.3 for each comparisons), and as expected, SLA in upper branches was lower than in lower branches (< 0.001, Table 1). AS : AL in upper branches was lower than in lower branches (P < 0.01). Fertilized trees had lower AS : AL of upper branches, reflecting more leaf than wood production, and both fertilized and combination trees had higher AS : AL of lower branches than control and irrigated trees, reflecting greater leaf retention along the branch (P < 0.01 for each). The results of the light distribution, together with the similarity in needle and branch characteristics within each canopy layer, despite the large difference in L, produce a branch-estimated GS with a weight ratio of ∼ 9:1 for upper and lower branch GS, respectively.

Figure 1.

The distribution of leaf area index (L) with height in control (a) and fertilized trees (b) and the attention of photosynthetic photon flux density at each layer in the tree canopies (Qi) with different above-canopy photosynthetic photon flux density (Q0) in control (c) and fertilized (d) trees.

Smoothing of time series and comparison among scales of GS

Based on the raw 15 min branch sap flux data, there was no relationship between upper or lower branch GS and environmental conditions (Q0, D and θ) in any treatment (P > 0.6 for all). In addition, there was no relationship between the branch- and main stem-estimated GS (P > 0.5 for all) or between the branch-estimated GS and gs (P > 0.4 for all). The apparent lack of response to environmental conditions was due to high-frequency noise in branch-estimated GS (Fig. 2). In order to extract the diurnal signal from the high-frequency variability in branch-estimated GS, we performed a moving average smoothing of the data. An illustration of a representative day is shown in Fig. 2, where panels a–c depict raw data, and the corresponding 45 min moving average windows are shown in panels d–f. The differences between the smoothed and unsmoothed data for the branches and stems are shown in Fig. 2 (g–i). We used both the smoothed average and the residuals from the average in subsequent analyses.

Figure 2.

Canopy average stomatal conductance (GS) on a 15 min time step from nine upper branches (a), lower branches (b) and main stems (c). GS from nine upper branches (d), lower branches (e) and main stems (f) using a moving average over three time steps from (a–c), respectively. The differences between unsmoothed and smoothed data from (a & d), (g); (b & e), (h); and (c & f), (i) for each branch.

A strong, positive relationship was found between the smoothed branch-estimated GS data and main stem-estimated GS in all treatments (Fig. 3, P < 0.001, slope between 1.0 and 1.05, intercept P > 0.50 for each treatment). The comparison between branch- and main stem-estimated GS includes both high and low Q0 (sunny or cloudy conditions, respectively) and high and low θR (constantly irrigated or not, respectively). There was a general increase in variability between branch- and main stem-estimated GS as the magnitude of both increased (Fig. 3).

Figure 3.

Relationships between canopy average stomatal conductance (GS) calculated from main stems and scaled from light attenuation (Fig. 1)-based weights (Table 1) and upper and lower branch-estimated GS in control, irrigated, fertilized and combination trees. Each point represents the mean of three sap flux measurements.

Smoothed upper and lower branch-estimated GS in all treatments was highly correlated to gs measured on the same branch during the same time period (Fig. 4, P < 0.001). Comparisons were made using the mean values for the three trees measured four times in each of the four treatments on 2 d, 1 d before the irrigation experiment (24 July, θR ≈ 0.06 m3 m−3) and 1 d during the irrigation experiment (29 July, θR ≈ 0.10 m3 m−3). In all treatments, the ratio of gs to branch- or main stem-estimated GS or GS showed no time dependence (P > 0.60 for all). Main stem-estimated GS was not significantly different (P > 0.3) from branch-estimated GS in all treatments, but upper branch-estimated GS was greater than both (P < 0.01).

Figure 4.

Comparison of porometry-based stomatal conductance (gs) and canopy average stomatal conductance (GS) in control, irrigated, fertilized and combination trees from three tree measurement positions: upper branches (open symbols), lower branches (gray symbols) and main stems (closed symbols). The dashed line represents the 1:1 line.

Response of GS to D

We tested whether the coefficients from Eqn 1, used on smoothed GS in all measurement locations (main stem, lower and upper branches), conformed to the theoretical expectations of the response of GS to D. In all treatments and GS measurement locations, −δ related to GSref with a slope not different from the theoretical slope of 0.6 (P > 0.50, intercept P > 0.2 in all treatments, all R2 > 0.95; Fig. 5). θR did not affect the ratio between −δ and GS in any treatment (P > 0.2); lower θR caused the GSref values to decrease in all treatments with a proportional drop in −δ (Fig. 5).

Figure 5.

The relationship between canopy average stomatal conductance (GS) calculated at vapour pressure deficit equals 1 kPa (GSref) and the sensitivity of GS to vapour pressure deficit (δ) in control (a), fertilized (b), irrigated (c) and combination (d) trees in different measurement positions and soil volumetric moisture levels (θ). High θ was greater than 0.07 m3 m−3 and low θ was less than 0.06 m3 m−3. The dashed line represents the theoretical relationship between GSref and δ based on Eqn 1.

Variation in Q0 and GS

Although the smoothed data facilitated testing of the hypothetical relationships between the sensitivity of stomatal conductance to D under varying soil moisture and light, the residuals of branch-estimated GS from the smoothed diurnal pattern (Fig. 2g–i) suggest additional influences. Branch-estimated GS, even within one canopy level, changes asynchronously from one measurement to the next, with some branches increasing and others decreasing from one measurement period to the next, generating a large degree of variability in the unsmoothed data (80 and 60% coefficient of variation for lower and upper branch-estimated GS, respectively) such that unsmoothed branch-estimated GS was unrelated to D, Q0 or θR.

We normalized the SD of both branch-estimated GS and Q0 by their highest value in each canopy level to facilitate a joint qualitative comparison – the absolute values of the SDs of branch-estimated GS in the upper canopy were appreciably higher. Both direct and diffused light increased with increasing Q0, but the maximum amount of variation in both occurred during cloudy conditions as seen by the group of points below the line of clear sky direct radiation and above the line of diffuse radiation (Fig. 6a). The combination of the direct and diffused light variability lead to the increased SD of absorbed Q0 during cloudy conditions (Fig. 6b). The SD of branch-estimated GS, calculated as difference from the individual branch moving average smoothing depicted in Fig. 2, increased with increasing Q0 in both upper and lower branches (Fig. 6c). At each canopy level, the highest SD of branch-estimated GS occurred on two cloudy days when maximum Q0 was only 1300 µmol m−2 s−1. The FETCH model captured the increased variability in SD observed at each light level, and qualitatively captured the increased variability in branch-estimated GS on cloudy days with larger deviation of residuals from the smoothed diurnal occurring on the two cloudy days (Fig. 6d).

Figure 6.

Relationship between photosynthetically active radiation by crown layer (Qi) to the total incoming photosynthetically active radiation (Q0) (a), modelled SD of absorbed light in upper or lower branches (b), measured SD of GS residuals from a moving average smoothed diurnal in upper and lower branches (c) and finite elements tree crown hydrodynamics (FETCH) modelled SD of GS residuals from a moving average smoothed diurnal in upper and lower branches (d).


Branch-estimated GS reflected the influences of signals at two different frequencies – a low-frequency signal originating from the diurnal pattern in light and vapour pressure deficit, observable at a 45 min moving window of average flux, and a higher frequency fluctuation consistent with expected variation in light because of mutual shading among crown elements within and among canopy layers, observable at the 15 min average data (Figs 2 & 6). As shown elsewhere (Oren et al. 1998), the low-frequency branch-estimated GS behaved similarly to the smoother pattern observed at the high-frequency stem sap flux (Fig. 3), indicating that the main stems of P. taeda are not compartmentalized and integrate over the high-frequency fluctuations observed in individual branches. It is beyond our capability to obtain continuous measurements of leaf-level gs at representative positions on a single branch. It is likely, however, that while some needles on a branch are exposed to high light for a few moments, others are deeply shaded, with the conditions quickly changing. This would have resulted in additional high-frequency fluctuations beyond those observed in individual branches.

The observation of greater fluctuations in branches than main stems (Oren et al. 1998; Ewers & Oren 2000; Martin et al. 2001) means that non-compartmentalized transport organs smooth over the variability in flux. Consequently, the GS calculated from flux measured in stems and even branches masks larger variation in gs occurring among the many, smaller dimension organs attached downstream. Photosynthetic parameters tend to saturate with increasing stomatal conductance (Katul, Leuning & Oren 2003). Because the fluctuation in GS is greatly reduced relative to the fluctuations actually occurring at the leaf surfaces, crown or canopy photosynthesis calculated based on mean stomatal conductance (e.g. Schäfer et al. 2003) may be overestimated relative to calculations of photosynthesis that preserve the spatial distribution of stomatal conductance at each timestep.

Spatial variation in stomatal conductance

For simplicity of presentation, spatial variation in gs can be discussed in relation to the vertical trend in average light conditions, reflected in gross characteristics of leaf and shoot structure and function, and variation in gs at each layer in the canopy, reflecting the interaction between the variation in light and hydraulic architecture.

Low-frequency variation

Light exerts a strong influence on leaf structure (Oren et al. 1986a; Larcher 2003). Although fertilized treatments doubled L and altered the LAD profile, the effect on the vertical distribution of light was small (Fig. 1). Thus, SLA of needles from the four treatments was higher at the bottom of the canopy, but, reflecting similar light profiles, there were no differences in SLA among treatments at either canopy levels (Table 1).

Lower AS : AL in upper than lower canopy branches in all treatments (Table 1) seems contrary to findings in other studies showing that the cross-sectional transport area of better-illuminated branches is higher per unit of transpiring leaf area (Oren, Werk & Schulze 1986b). The reason for the apparently different finding here is that the ratio is with AS near the stem and not at the base of the needle-carrying portion of branches. In upper branches, needles are distributed along the entire branch, while in lower branches, needles are supported only by secondary and tertiary twigs, quite a distance from the stem. Thus, the more relevant information provided by the ratio in this study is that it was similar among treatments at a given level in the canopy. The slightly higher values at the bottom of the canopy in fertilized treatments reflect reduced needle retention in lower canopy branches of fertilized P. taeda stands (Amponsah et al. 2004).

The vertical variation in light, needle and shoot characteristics were complimented with variation in stomatal conductance at the leaf and branch levels. Regardless of treatment, gs in upper and lower needles was similar to corresponding branch-estimated GS (Fig. 4). When gs measured higher and lower in the canopy were weighed by the corresponding leaf area, partitioned based on the light distribution (Fig. 1), the agreement with main stem-estimated GS was also noteworthy (Fig. 4). This correspondence also reflects the fact that gs measurements were made on excised needles exposed to constant light conditions in the chamber, effectively smoothing gs. Thus, instantaneous gs measurements on intact needles made at ambient light conditions (which would be changing more rapidly than could be measured by a porometer because of crown light conditions) would not have corresponded to GS as well. More data were available for comparison of branch- to main stem-estimated GS, and again after weighing branch estimates by leaf area, the agreement of stomatal conductance between these two very different scales was remarkable in all treatments (Fig. 3). These results agree with leaf-level photosynthesis measurements taken the year after this study in the same stands showing almost twice the photosynthesis in upper-grown positions as lower regardless of canopy nutrition (Gough et al. 2004).

The scaling of gs and individual branch-estimated GS to main stem-estimated GS (Figs 3 & 4, respectively) necessitated weighing lower canopy measurements approximately nine times more heavily than upper canopy measurements (Table 1). This weighting indicates that in P. taeda, over a twofold range in L, measurements lower in the canopy are more representative of main stem-estimated GS than measurements taken in the upper canopy. The dominance of either gs or GS from lower branches in the scaling to main stem-estimated GS may reflect the shade-intolerant nature of P. taeda – despite a sharp decrease with depth from the top in light and photosynthetic rate (and presumably gs; Oren et al. 1986a), shade-intolerant species can retain only foliage with relatively high gas exchange rates. Indeed, the gas exchange of leaves at the canopy bottom of five forests, including broadleaf and needleleaf stands, had a large effect on the gas exchange rate of the entire canopy (Meir et al. 2002).

GS estimated from main stem measurements and those from smoothed branch data showed an expected sensitivity to D that was proportional to the GS at low D (GSref at D = 1 kPa, Fig. 5). Such behaviour is proposed or observed in other studies (Kaufmann 1982; McNaughton & Jarvis 1991; Yong, Wong & Farquhar 1997; Oren et al. 1999), and is consistent with a homeostatic, isohydric regulation of water potential. The response of main stem-estimated GS to D, the difference among treatments and the effect of soil moisture on the response has been previously shown at this site (Ewers et al. 2000, 2001a). The combination treatment showed highest Gs and sensitivity to D, and the fertilized treatment the lowest, and increasing soil moisture with continuous irrigation increased gs and proportionally the sensitivity to D. Here we demonstrate that these patterns are conserved at the finer scale of branches once the low-frequency diurnal signal is extracted from the high temporal frequency signal (Fig. 5), and that the ratio between −δ and GSref at the branch and crown level (∼ 0.6) is similar to the average of many species (Oren et al. 1999).

High-frequency variation

Vapour pressure deficit was uniform through the canopy profile in all treatments because of a high degree of canopy coupling to the atmosphere reflected in negligible leaf-to-air temperature differences (Ewers & Oren 2000). Variations of light on leaf surfaces increase with solar elevation angle. When the sun is high, direct radiation penetrates deep in the canopy, but shading by shoots increase the variability in the intensity of light on leaf surfaces. However, cloud cover decreases the variation in light at a given solar elevation angle because the amount of direct radiation decreases relative to diffuse radiation, and the latter does not produce shading (see e.g. mid-range Q0 in Fig. 6a). Such interplay between direct and diffused light depending on solar angle and cloud conditions generates high SD of absorbed light in both upper and lower branches (Fig. 6b), and is consistent with the observed increase in the maximum variation branch-estimated GS with increasing Q0, as well as the high variation in branch-estimated GS on cloudy days (Fig. 6c). Further, more of the needles would be exposed to Q0 below 700 µmol m−2 s−1 (threshold for Pinus teada stomatal response to Q0; Schafer et al. 2003) on cloudy days, further contributing to increased variability.

The high-frequency variation of branch-estimated GS (Fig. 2) in response to Q0 is possible if (1) gs response to Q0 is sufficiently fast relative to the frequency of data collection, and (2) water storage capacity of branches is low. A fast time constant for stomatal response to light was measured in P. taeda (shading, 15 min; shade removal, 25 min; Whitehead & Teskey 1995), as well as small total capacitance (Phillips et al. 1997), meaning that little water can be stored in branches. Thus, the high-frequency fluctuations in branch-estimated GS (Fig. 2) probably reflect the spatial variation in light intensity (Vesala et al. 2000; Stenberg et al. 2001; Zweifel, Böhm & Hösler 2002). Indeed, the variation in Qi at both levels in the canopy was large and qualitatively similar in relation to Q0 to the variation in upper and lower branch-estimated GS (Fig. 6). The similar variation of upper and lower branch-estimated GS (Fig. 3) agrees with the findings of Zweifel et al. (2002) that similar variability in gs can be found in upper and lower branches at a given light level even though gs is much higher in upper branches.

Although Q0 and cloud conditions can set an upper limit (Fig. 6a) to the high-frequency variability in gs at each canopy level (Fig. 6b), these external controls cannot explain the variation in the SD of gs below that upper limit. However, a hydraulic feedback to quick changes in light conditions on individual branches at one level in the canopy can increase the range in stomatal conductance among leaves and branches. The feedback can occur as stomatal conductance decreases in some needles or branches as a result of shading within or between branches, allowing a greater proportion of the upstream transport capacity to become available for better-illuminated needles and branches (Buckley & Mott 2002b; Brooks et al. 2003).

Leaf specific hydraulic conductivity exerts an influence on stomatal conductance (Saliendra, Sperry & Comstock 1995; Ewers et al. 2000; Salleo et al. 2000; Hubbard et al. 2001; Mencuccini 2003). Some defoliation studies show that the resulting increase in average leaf specific hydraulic conductivity is associated with an increase in gs up to the physical maximum (Meinzer & Grantz 1990; Pataki, Oren & Phillips 1998b), yet others have found no such response (Troeng & Langstrom 1991; Syvertsen 1994). Similarly, in some studies of partial shading, the unshaded portion of the leaf or crown showed increased gs relative to expected value (Meinzer & Grantz 1990; Buckley et al. 2002b; Pepin, Livingston & Whitehead 2002), but other studies did not observe such a response (Syvertsen 1994; Phillips, Bond & Ryan 2001; Brooks et al. 2003). These inconsistencies may reflect the gs of both the shaded and illuminated portions relative to their respective maximum at the time of the experiment, the total transpiration in each portion and water flow pattern within the xylem. For example, shading the lower three-fourths of a Pinus radiata crown increased gs in the downstream crown fourth, while shading the upper crown fourth had no impact on the upstream lower crown because of the small contribution of upper foliage to total transpiration (Whitehead et al. 1996). Similarly, shading most of the foliage on branches of Pseudotsuga menziesii trees did not increase the gs of the remaining sunlit foliage (Brooks et al. 2003), presumably because of a high degree of hydraulic autonomy even among shoots on a given branch (Shinozaki et al. 1964; Sprugel, Hinckley & Schaap 1991). Thus, at one canopy level, the extra variation in GS contributed by hydraulic feedback to the spatial variation in light is likely to differ among species, change with light, vapour pressure deficit, soil moisture and vertical distribution of leaf area, and would be difficult to separate from the direct effects of light on GS.

Our experimental design and, thus, data density in space and time are insufficient to directly assess the effect of hydraulic feedback to variability in light on the variability of branch-estimated GS at a given time and canopy level. However, the hydraulic model translated the increased upper limit of variation in Qi with Q0 to a variability in branch-estimated GS (Fig. 6d) consistent with that observed in the data (Fig. 6c).

In summary, this study demonstrated that the main stem of P. taeda operates as a low pass filter, smoothing over the high-frequency variability generated among branches within and among layers in the crown by the variation in light and possibly hydraulic feedback. The consequence is that the stem homogenizes the water potential signal that reaches the roots. Another potential consequence is that using sap flux-based stomatal conductance measured in non-compartmentalized stems would result in overestimation of photosynthesis relative to a method that preserves the spatial variability (e.g. frequency distribution) of stomatal conductance in each canopy layer.


This research was funded by the US Department of Agriculture (USDA) Forest Service, Westvaco Co., the National Science Foundation though Grant BIR-9512333 and the US Department of Energy through the Southeast Regional Center at the University of Alabama (Cooperative Agreement No. DE-FC03-90ER61010). The authors are grateful to P. Anderson, G. Burkland, J. Butnor and C. Gough for technical assistance. This work contributes to the Global Change and Terrestrial Ecosystem (GCTE) core project of the International Geosphere-Biosphere Program (IGBP).