Original article

Scaling xylem sap flux and soil waterbalance and calculating variance:

a method for partitioning water flux in forests

Ram Oren Nathan Phillips Gabriel Katul Brent E. Ewers

Diane E. Pataki

Nicholas School of the Environment, Duke University, Durham, NC 27708-0328, USA

(Received 15 January 1997; accepted 16 July 1997)

Abstract - To partition evapotranspiration between canopy and subcanopy components in a12-m-tall Pinus taeda forest and to assess certain aspects of environmental regulation of canopytranspiration, we quantified water flux in a forest using three approaches: 1) measuring water fluxin xylem of trees, and scaling to stand transpiration of canopy trees (EC); 2) measuring soil watercontent and saturated conductivity, and modeling drainage to estimate total evapotranspiration (ET)during rainless days based on a local water balance (LWB); and 3) using an eddy correlationapproach to estimate total ET. We calculated variances for each estimate, and proposed anapproach to test for differences between estimates of EC and ET. Diurnal ’patterns’ in wateruptake were similar using direct measurements in stem xylem and LWB. However, LWB wasfound to be inappropriate for estimating ’absolute’ ET diurnally when changes in soil moisturebetween consecutive measurements were small. Eddy correlation estimates of ET are of a highertemporal resolution than xylem flux measurements made in branches. Diurnal flux patterns inbranches are more similar to the pattern generated by eddy correlation than those in stems. How-ever, differences between the patterns indicate that patchiness in branch transpiration may pre-clude using branch xylem flux measurements to estimate canopy conductance. In one stand,daily EC accounted for ca 70 % of total ET estimated by either LWB (in a separate study) or theeddy correlation approach; the difference between ET and EC was significant based on vari-ances calculated to account for spatial variation in each. Regardless of the vapor pressure deficit,EC decreased linearly with soil moisture from 2.5 to 1.5 mm d-1 over a 9-d drying cycle, as soilmoisture in the rooting zone (ca 0.35 m depth) declined by 23 mm. (© Inra/Elsevier, Paris.)

canopy transpiration / subcanopy evapotranspiration / sap flux / soil water balance /variability

* Correspondence and reprintsTel: (1) 919 613 8032; fax: (1) 919 684 8741; e-mail: ramoren@duke.edu

Résumé - Une méthode pour séparer les flux hydriques en forêt basée sur l’extrapolationdes mesures de flux de sève, le bilan hydrique et le calcul des variances. Dans le but de sépa-rer, dans l’évapotranspiration d’une forêt de Pinus taecla de 12 m de hauteur, la participationdes arbres de celle du sous-étage, et d’évaluer les caractéristiques de la régulation de la transpi-ration des arbres, les flux hydriques ont été quantifiés à partir de trois approches complémentaires :a) la mesure du flux de sève brute dans les arbres, permettant de calculer la transpiration du peu-plement (EC) ; b) le bilan hydrique local (LWB), à partir de la mesure de la teneur en eau du sol,de la conductivité hydraulique du sol à saturation, et de la modélisation du drainage pour estimerl’évapotranspiration totale (ET) pendant des périodes sans pluie ; c) la mesure, d’ET au moyen dela méthode des corrélations turbulentes. Les variances de chaque estimation ont été calculées, etune approche pour tester les différences entre EC et ET a été proposée. Les variations journalièresde consommation en eau à partir du flux de sève étaient similaires à celles obtenues à partir du bilanhydrique. Néanmoins, la méthode LWB s’est montrée inadéquate pour estimer les variationsabsolues d’ET lorsque les variations de teneur en eau du sol étaient faibles. L’estimation de ETau moyen des corrélations turbulentes a montré une plus forte résolution temporelle que celle baséesur la mesure de flux de sève dans les branches. Une plus grande similitude des variations jour-nalières d’ET a été montrée entre les méthodes des corrélations turbulentes et du flux de sève dansles branches qu’avec le flux de sève mesuré dans les troncs. Toutefois, des différences dans lesallures de courbes semblent indiquer qu’il existe une hétérogénéité de la transpiration desbranches, ce qui exclut l’utilisation de cette méthode pour évaluer la conductance du couvert. Dansune des parcelles, EC journalier a atteint environ 70 % d’ET estimé par la méthode LWB (cf. uneautre étude), ou par les corrélations turbulentes. À partir du calcul des variances, pour prendre encompte la variabilité spatiale des différentes estimations, on a pu montrer que la différence entreET et EC était significative. En dehors de l’effet du déficit de saturation de l’air, EC a montré unediminution linéaire avec l’humidité du sol, pour passer de 2,5 à 1,5 mm/j, sur une période de 9 jde dessèchement, tandis que l’humidité du sol dans la sphère racinaire (qui s’étend sur environ0,35 m de profondeur) diminuait de 23 mm. (© Inra/Elsevier, Paris.)

transpiration du couvert / transpiration du sous-étage / flux de sève / bilan hydrique /variabilité

Nomenclature

1. INTRODUCTION

In a series of publications beginningwith an introduction to water balance

dynamics, Eagleson [18] stated that mod-eled land-atmosphere interactions mustretain both the underlying physical deter-minism and the uncertainty that plays alarge role in nature. Thus, any approachto modeling or estimating water balancemust rely on probability distributions,which apply to the values of model param-eters and independent variables, and tothe information, or observations, used toevaluate the system’s behavior.

In recent years, a rapidly increasingnumber of papers on water flux in tree

xylem reflects the introduction of afford-able, easy-to-use instruments. Most usersconsider some of the variability in themeasured water flux when values are

extrapolated to the stand. However, gen-erally less attention is paid to calculatingthe uncertainties about the estimated stand

transpiration, which is based on sensorsoften representing less than 1 % of the leafarea of monitored trees. Several authorshave suggested methods to improve scal-ing measurements to stand-level [ 13, 17,26, 28, 48]. Thus far, stand-level transpi-ration estimates based on xylem water flux(JS) measurements have been assumed tovary mostly with time, implicitly assumingthat the variation in space is captured inthe variability in JS among measured trees.As a result, differences between canopytranspiration (EC) values obtained by thismethod and, e.g. eddy correlation esti-mates of total evapotranspiration (ET) havebeen attributed, under certain conditions,to the subcanopy contribution to standtranspiration and evaporation (ESC; [2, 17].

In order to account for some variabilityin JS, sample trees have been chosen torepresent different size classes [2, 17].Based on this approach, EC is calculated asthe product of JS in each tree diameter

class, or of water flow per unit of circum-

ference, and the area of xylem in eachclass, or the total circumference of all treesin the class, respectively. EC is calculated

by summing over all classes. The motiva-tion behind this scaling approach is thattrees of different size classes have differ-ent flow rates per unit of xylem area ortree circumference. However, while insome situations JS may be affected by theintensity of competition around measuredtrees [2], or by spatial variation in root-ing volume and soil moisture availability[27, 28], this is not always the case [54].

Here we suggest a method for incor-

porating the variability in both JS and

hydroactive xylem area (AS) per unit ofground area (AG; sapwood area index) toestimate the variance about the calculated

EC. The result is an interval of valueswithin which EC is likely to occur at aselected probability level. This approach isuseful for two lines of investigations: 1)comparison of estimated EC based on

xylem water flux measurements with otherEC estimates based on modeling [64] ormass balance (e.g. eddy correlation, waterbudget; [2, 10]; and 2) partitioning of tran-spiration between canopy and subcanopyin forests [17, 40]. When subtracting ECobtained with scaled xylem water fluxmeasurements from ET obtained, for

example, by eddy correlation in order toestimate ESC, it is necessary to accountfor the sources of, and sinks for, waterunique to each method. The sources andsinks are likely to differ, not only in thevertical axis, but also horizontally. Thus,for eddy correlation measurements theyare determined by the footprint size anddirection, and for JS measurements theyare affected by the sampling design. Inthis study, we address only the variabil-ity associated with estimated EC.

In most situations, JS measurements

represent water uptake rate rather thantranspiration. When comparing ET with

EC estimated diurnally, the resulting time-lag may introduce large errors into the par-

titioning of fluxes between canopy andsubcanopy [27, 43]. Using Granier-Typexylem flux sensors inserted in the mainstem of trees and in branches, we com-pared diurnal patterns of water uptake andtranspiration, respectively, to the sourceof water in the canopy, estimated with

eddy correlation flux measurements abovethe canopy combined with measurementsof water vapor concentration within the

canopy volume. A companion paper [50]describes the lag effect in terms of poten-tial errors in estimating canopy conduc-tance.

2. MATERIALS AND METHODS

The study was undertaken in a 12-year-old,uniform plantation of loblolly pine (Pinus taedaL.; 1000 m x 300 m) located near Durham,NC, USA, in the Blackwood Division of DukeForest (32°52’N, 79°59’W), a transitional zonebetween the coastal plain and the Piedmontplateau. One-year-old seedlings were planted in1983 at 2.4 m spacing, and at the time of thestudy, formed the main canopy, reaching aheight of 12 m. Below, occasionally reachinginto the main canopy, were naturally seededloblolly pines, a small number of individuals ofother species, including Liquidambar styracu-flua L., Juniperus virginiana L. and Quercus

phellos L., as well as herbaceous, grass andvine species. The natural regeneration increasedthe density of individuals in the stand from theplanted 1 736 ha-1 to over 5 200 ha-1, withmost of the natural regeneration in the sub-canopy, small diameter classes. The stand islocated on soils of the Enon Series, moderate-fertility, acidic Hapludalfs of low potential forerosion due to less than 2 % slope. Soil mois-ture is extracted mostly from the top 35 cm (ca90 %; Oren et al. [49]) and maximumextractable water to this depth is ca 120 mm.

In the summer of 1994, 66 circular plots of44.5 m2 were positioned in the stand, and thediameter at 1.3 m above ground for all indi-viduals with a diameter greater than 20 mmwas measured. In addition, three plots wereestablished in 1993 and 1994 as part of a long-term study of water flux in forests, one plot ona portion (111 m2) of the stand encircled bythe free air CO2 enrichment (FACE) proto-type, and the others as reference plots (Ref. 1and Ref. 2) nearby (53 and 117 m2, respec-tively), and diameters were measured as above.Individual tree and plot characteristics areshown in table I.

2.1. Overstory transpiration- scaled xylem flux measurements

Overstory transpiration was estimated usingmeasurements of xylem water flux in the outer

20 mm of the xylem in trees of two plots,FACE and Ref. 1, taken with Granier-Typesensors [22]; and at two depths (0-20 mm fromthe cambium and 20-40 mm) in trees of Ref. 2,using modified sensors, as described in detailin Phillips et al. [51]. Ten individuals weremeasured in each plot; here we report on datacollected between June 1993 and September1996. Measurements of JS provide values inunits of gH2O m-2xy lems-1, which must be con-verted to units of flux for the stand using thehydroactive xylem area (sapwood) per unit ofground area (AS:AG, unitless) as a scaling vari-able. Scaling was based on measured JS inouter xylem only (Jout), while JS in the inner

xylem (Jin) was estimated based on Jout and acorrection factor obtained by comparing Jinwith Jout in Ref. 2. The flux in each xylemregion was scaled using its respective AS:AG to

obtain plot-level EC for comparison with 1)total evapotranspiration (ET) obtained with thelocal water balance (LWB) calculation in thatplot. The flux in each xylem region was alsoscaled using average inner and outer sapwoodin the 66 plots to a stand-level EC for compar-ison with 2) ET obtained with the eddy corre-lation calculation in the whole stand. Due tothe young age of the stand, sapwood comprisedall the cross-sectional area inside bark [51];bark thickness was estimated from a relation-

ship developed on site with outside bark diam-eter.

Vapor pressure deficit (D) was calculatedfrom relative humidity and temperature mon-itored with sensors (RHA 1, Delta-T DevicesLtd., Cambridge, UK) positioned in both FACEand Ref plots 7 m above ground, a point cor-responding to the peak in the stand leaf areaindex (L) profile as estimated with a canopyarea analyzer (LAI 2000, Li-Cor., Lincoln, NE,USA). Temperature, relative humidity and theGranier-Type sensors were interrogated every30 s, and half-hour means were stored in a

data-logger (DL2, Delta-T Devices). On 19September 1994, Granier-Type sensors weremonitored in six branches, originating 5, 7 and9 m above ground in each of two trees, anddata were stored every 20 min for further com-

parison with eddy correlation estimates. Duringthe study, photosynthetically active quantumflux (LI-193SA spherical quantum sensor, Li-Cor) and shortwave radiation (LI-200SA pyra-nometer) were also measured above the canopyusing the same logging procedure.

2.2. Evapotranspiration - local waterbalance (LWB) approach

The LWB method, shown to produce rea-sonable estimates of ET over a large portionof one growing season [35, 59], was evaluatedfor use in conjunction with xylem flux mea-surements to partition evapotranspirationbetween canopy and subcanopy componentsover short periods of time, such as portions ofa single day. The method is based on mea-surements of throughfall precipitation (PT,mm), and volumetric soil moisture content (&thetas;)in the rooting volume. Volumetric soil moisturemeasurements are used to calculate changesin soil moisture, ΔSW, between measurementtimes, and estimate drainage from the soilbelow the rooting zone. Evapotranspiration iscalculated as the balance between PT, ΔSW and

drainage. The method assumes negligible evap-oration from wet surfaces and soil. The exper-iment was conducted over a single rainless day(see below for details on instruments and cal-culations). We chose a day without precipita-tion, several days after the last rain event, toavoid all errors associated with estimates of

PT, one of the three components in the LWBapproach, as well as to strengthen the assump-tion that little evaporation occurs.

Estimates of ΔSW with respect to time,dSw/dt, and drainage (see below), wereobtained from measurements of &thetas; [12, 45, 60].At five locations along a radial transect in the30 m diameter FACE plot, metallic timedomain reflectometer (TDR; Tektronix 1502-B, Redmond, Oregon, USA) rod pairs, 0.10,0.20, 0.35 and 0.70 m in length, were installedvertically from the surface to four depths. Timedomain reflectometer measurements were takenat 1-h intervals throughout 29 September 1994.

An empirical relationship is typicallyderived to convert κ measurements to &thetas;. Toppet al. [61] derived a general relationship fordifferent mineral soils which was found to besuitable for soils with low clay content. Thesoil in this study was a clay loam in the upperhorizons, and Topp’s equation was tested andfound to quantify &thetas; accurately [59]. Soil mois-ture content was calculated for each interval

by subtracting total moisture estimated fromtwo rod pairs of consecutive length and divid-ing by the difference in length.

The drainage can be estimated as a Darcianflux q = K(&thetas;) ▿H, where K(&thetas;) is hydraulic con-ductivity at &thetas;, H is the matric potential and V

is the gradient operator. For deeper layers, themoisture content depth variation is not largeand in a first order analysis the drainage isstrictly gravitational so that ▿H ≅ 1 (see, e.g.[33, 66]). Hence q = K(&thetas;), and an equationdeveloped by Clapp and Hornberger [16] maybe used to calculate drainage. Thus, thedrainage flux is

where KS is saturated hydraulic conductivity,and &thetas;S is the saturated moisture content. The

exponent b is empirical, and must be estimatedbased on the soil type. We estimated b using aroot exclusion monolith over a large part ofone growing season (nearly 5 months), mea-suring PT and ΔSW, and calculating drainage bydifference (assuming evapotranspiration andlateral flow to be negligible; see Todd [59] fora full description of the process). Calculateddrainage from the monolith was made to equalK(&thetas;) in the monolith by changing b iteratively.

Saturated conductivity, KS, necessary forestimating drainage [equation (3)], was mea-sured with a compact constant head perme-ameter (Amoozemeter, Ksat, Inc., Raleigh,North Carolina) which quantifies the steadystate flow of water through soil [1]. Measure-ments of KS were made at 0.20-0.35 m and0.55-0.70 m soil depths at four of the five posi-tions where TDR values were collected and

averaged.Saturated soil moisture, &thetas;S, also necessary

to estimate drainage [equation (1)], was esti-mated in different soil levels based on the fol-

lowing procedure: a trench was excavated in astepwise fashion, representing 0.1-m incre-mental intervals from the ground surface to0.6 m in depth. At each step, 0.1-m TDR rodswere installed and &thetas; was measured after thesoil was brought to saturation. The value of &thetas;Swas similar, 54 % (SE = 2 %) in all layers.

2.3. Evapotranspiration- micrometeorological approach

In a uniform and extensive canopy, under

steady-state conditions, the mean source ofwater vapor from the stand, ET, can be esti-mated from the flux above the canopy, account-

ing for changes in absolute humidity in thecanopy volume. Total evapotranspiration canthen be compared to EC estimated using the

Granier-Type sensors. During the summer of1994, water vapor flux was measured on 9 daysin the FACE plot using an eddy correlationsystem positioned above the canopy [34]. Themeasurements at z/h = 1 (where z and h arethe instrument and canopy height, respectively)were made with a Gill triaxial sonic anemome-ter (Gill Instruments, Hampshire, UK) mounted0.25 m from a Campbell Scientific kryptonhygrometer (Campbell Scientific, Logan, Utah,USA). The velocity, temperature and watervapor concentration measurements from theseinstruments were sampled at 10 Hz (21XCampbell Scientifc micrologger). The rawmeasurements were then fragmented into 27.5-min runs, with each run comprising 16 384(= 214) measurements per variable per run. Thefriction velocity (u*), latent heat flux (LE) wasthen computed as described in Katul et al. [34].In brief, LE = LV<w’q’>, where LV is the latentheat of vaporization, w’ is the vertical velocityturbulent fluctuation, and q’ is the water vaporconcentration fluctuation, and <> denotes time

averaging. The FACE prototype was notenriched with CO2 in four of the days, andatmospheric CO2 in the canopy was maintainedat an average of 55 Pa for five of the days.

3. RESULTS

3.1. Canopy transpiration

To estimate EC, information on JS and

AS:AG is combined [26, 65]. In measuringJS, we quantified the effects of two sourcesof variability: I) depth in the xylem, and 2)competition.

Within-tree variability in JS is appar-ent in figure 1, where diurnal JS in the outer

xylem was more than twice that in the innerxylem (compare Jin and Jout in Ref 2) dur-ing most of the day. Inner xylem alsorequired a longer night-time period torecharge its storage. The standard errorbars displayed in figure I demonstrate thelarge between-tree variability in Jout. Thevariability among individuals is shown indetail in the inset in figure I, where diurnalaverage Jout of the 30 trees in the stand

displays a non-normal distribution; the dis-

tribution of the natural log transform ofthe data is not different from normal

(P > 0.7). Between-tree variability in JSis, in part, explained by competition, indi-cated as AS:AG in a 3-m-radius plot sur-rounding and including a monitored tree(figure 2), but was uncorrelated with treeheight or diameter (P > 0.1). The nega-tive correlation between JS and AS:AG isused in the calculations of the variance of

EC, according to equations (4) and (6)below. Differences among plots potentiallyincorporated a small effect of elevated CO2at FACE. However, direct responses ofstomates to elevated CO2 in this site werefound to be weak (&sim; 5 % reduction in stom-atal aperture with double ambient CO2;[19]).

When measurements of JS are made atdifferent xylem depths, canopy transpira-tion is estimated based on J S in eachdepth, JSi (gH20m-2sapwoods-1) and the sap-wood area index for a given xylem depthin a plot, including the measured individ-uals, ASi:AG (m2sapwoodm-2ground):

Average canopy transpiration in thestand, EC, is the sum of ECi for all cate-

gories (e.g. depth, azimuth, size class,species).

The variance around the estimate foreach category, &sigma;2EC, is given by

where &sigma;2Ji is the variance of JSi (i.e. vari-ance of flux in a given depth) and ASi:AGhas zero variance for a single plot case.The combined variance of two categories(or more), such as the variance of the fluxin inner (&sigma;2ECin) and outer (&sigma;2ECout) xylem,about EC is

where rECinECout is the correlation coeffi-cient between the two categories. Thisestimate of EC and its variance on a plotlevel is suitable for comparison with esti-mates of ET based on the LWB approachin the same plot. Using equations (2)-(4),EC and &sigma;2EC were calculated. EC and &sigma;2EC,the latter converted to standard error, areshown in figure 3, both without account-ing for the radial pattern with depth in JS,and with correction (figure 3). It is clear

that not accounting for categorical differ-ences in JS causes a large over-estimationof EC, a discrepancy that becomes moresignificant as EC decreases (uncorrectedEC = -0.10 + 0.83 EC corrected; r2 = 0.99).

3.2. Evapotranspiration- LWB method

In order to compare estimates of EC to

those of ET based on the LWB approach ata high temporal resolution, measurementswere conducted in the FACE plot from0830 to 2030 EST on 29 September 1994,a cloudless day with a maximum air tem-perature of 25.5 °C and a minimum of11.1 °C. Throughout the day, &thetas; in the topsoil layer (0-0.1 m) changed from 11.65 to10.52 % (one-way ANOVA, P < 0.0001;SAS, Cary, NC, USA), while in the other

three layers to a depth of 0.7 m, &thetas;remained constant (P > 0.05; figure 4a).Converting &thetas; at the top layer to &Delta;SW, andtesting for the difference between themeans (LSD) identified four distinctly dif-ferent means (P < 0.05; figure 4b). Thesewere used to construct the diurnal patternin soil moisture and ET (based on Vogt etal. [63]). The diurnal pattern in &Delta;SWreflected a reduction of 1.13 % in &thetas;.Cumulative drainage, calculated withequation (1), was very low (0.058 mm),reflecting the low drainage rate in thesesoils under unsaturated conditions (fig-ure 4c). A diurnal pattern in EC was com-

pared to ET estimates based on the LWB

approach. After calculating drainage[equation (1)], diurnal ET was estimatedfrom &Delta;SW and drainage (figure 4b, c), andis depicted in figure 4d, along with diurnalEC, calculated with its variance as

described before. The variance estimatefor ET was restricted to the variance inmeasurements among TDR rods, carriedthrough the calculations. (Variance esti-mates for a longer measurement periodmust also include the spatial variance inPT.) The estimates of diurnal ET were

indistinguishable from those of EC, indi-cating that ESC was negligible, or that thesensitivity of TDR (ca 1 % measurementerror) is insufficient to detect small

changes in soil moisture between frequentmeasurements in zones from which 1)drainage is low, and 2) soil moisture isnot taken up by roots at high rates.

Daily EC during the study day, basedon the scaled measurements of JS, was1.28 mm, while the estimate of ET, basedon the LWB, was 1.25 mm. This reflectsthe insensitivity of the TDR to smallchanges in &thetas;, which can translate to a large

amount of soil moisture depletion whenintegrated over a thick rooting zone. If therequirement that only significantly differ-ent measurements of &thetas; are used for esti-

mating ET is relaxed so that changes in &thetas;are considered regardless of their statisticalsignificance, daily ET becomes 1.74 mm,of which EC accounts for ca 76 %.

3.3. Evapotranspiration- micrometeorological approach

Comparisons of EC to ET, the latterestimate based on eddy correlation mea-surements, must scale JS measurements

to a larger footprint than a single plot of30 m radius or less. Optimally, JS mea-

surements should be made in trees selected

randomly in an area large enough to rep-resent the footprint of the eddy correla-tion measurement point, and AS:AG should

be estimated in plots positioned aroundthe selected trees. In practice, factors suchas the number of data-loggers availableand the maximum sensor-to-logger dis-tance, which maintains signal integrity,confounds the number of JS measurements

that can be made and their spatial distri-bution. Then, a few clusters for JS mea-

surements are established, AS:AG is mea-sured around each of the randomlyselected trees to assess the correlationbetween the two variables, and additional

plots selected randomly in the stand areused to capture the variability in the scal-ing variable, AS:AG. (In species and situ-ations where there is evidence for a rela-

tionship between JS and tree size,stratification by size classes may also benecessary; Phillips et al. [51] demonstratedthat there was no relationship between JSand size class in our stand.) Because ofthe experimental nature of our stand, theclusters and plots were not selected ran-domly, however, between cluster vari-ability in JS is illustrated in figure 1.

Because scaling measurements basedon Granier-Type sensors to the standinvolve two variables, equation (2)changes so that the means of JS and AS:AGin each category (e.g. xylem depth) areused

and EC is the sum of ECi. The varianceabout ECi is calculated as

where rJS i(AS i:AG) is the correlation coeffi-

cient between JSi and ASi:AG (e.g. fig-ure 2). The variance of canopy transpira-tion in each category, &sigma;2ECi, can becombined to variance about EC, &sigma;2EC, usingequation (4). In general, we use the covari-ance term when two variables (A and B)satisfy the condition (rAB·&sigma;2A·&sigma;2B)/(A· B)>0.1.

The contribution of the variability in

ASi.:AG to &sigma;2ECi can be seen in figure 5,which shows wide ranges in both innerand outer xylem area index. Although thestudy was performed in a stand planteduniformly to ca 1 800 trees ha-1, estab-lishment of natural regeneration increasedthe density to over 5 200 trees ha-1 in

some plots, and natural variability inresources within the apparently uniformsite (e.g. depth to a clay pan), resulted invariation in growth rate. Combined, these

factors result in a total xylem area (roughly90 % of basal area) that is relatively evenlydistributed from 2 to 19 m2 ha-1 (figure 5c).

Using the information in figures 2 and5, and JSi measured over 9 d in which

eddy correlation measurements were madeabove the canopy, EC and its variancewere calculated. EC and its standard error

are shown in figure 6 in comparison to ETestimated based on eddy correlation mea-surements. Over 9 d, eddy correlationbased estimates of ET ranged from 0.6 to3.8 mm d-1. The relationship between theestimates based on the two methods waslinear (R2= 0.89; P < 0.0001), with a zerointercept (P > 0.1), and was not affectedby atmospheric CO2 concentration. Theestimated EC accounted for 69 % of ET. Itis possible that not all the differencebetween ET and EC can be attributed to

evapotranspiration from the subcanopy;we did not attempt to match the footprintof the eddy correlation system with thecorresponding AS:AG plots at each point intime during the 9-d comparison, and, thus,the sources of water vapor may not havebeen exactly the same for the two meth-ods. Such a detailed analysis would refinethe comparison between ET and EC.A major assumption in using JS to infer

transpiration is that water flux at the sen-sor reflects water transpired by leaves, anassumption which may be correct only forbranch level measurements. By comparingdiurnal EC patterns, obtained usingGranier-Type sensors to measure JS in

stems and branches, with diurnal ET pat-terns obtained using eddy correlation, itis possible to evaluate the utility of xylemflux measurements for estimating tran-

spiration. The comparison is based on theassumption that flux of water vapor abovethe canopy can be converted to the sourceof water vapor in the canopy by account-ing for changes in absolute humidity inthe canopy volume (using data on tem-perature and relative humidity collectedat 7 m, and assuming that there is no ver-tical profile; [64]). Moreover, it is assumedthat the behavior of canopy trees is similarto the behavior of all the vegetation in thestand; thus, while EC may account for onlyca 70 % of LE at any time, the pattern inboth is assumed to be similar. This

assumption is supported by the high cor-relation between flux measurements made

using eddy correlation systems at twocanopy levels (above the canopy and at0.7 canopy height; R2 = 0.72; P < 0.001; N= 133 during 8 d, each datum representinga 27.3-min sum; P intercept > 0.1), butmay not be valid both early and late in theday.

On a clear day ( 19 September 1994),shortwave radiation showed a nearly sym-metric pattern, while the pattern in D waslagged relative to radiation (figure 7a).

Latent heat flux was converted to JS usingthe average AS:AG in the stand (see fig-ure 5c), and summed over 20-min inter-vals. (A conservative thermal equilibriumresponse time of the Granier-Type sensoris ca 30 s.) Eddy correlation measurementswere made at the canopy-atmosphereinterface, and, based on the previous find-ings (figure 6), were not expected to pro-duce similar values. Thus, comparison ofvalues from eddy correlation measure-ments, stems and branches were made

only to assess whether the high frequencyvariation in these measurements are sim-ilar. For this purpose, only JS of two trees

adjacent to the tower, and their sixbranches, were compared to LE. In fig-ure 7b, it is apparent that the three diur-nal patterns are similar, but that the highfrequency fluctuations are smoothed inthe mean JS of the two stems; the patternis even smoother when more trees are

averaged. Even the average diurnal pat-tern in branch JS is smooth, althoughbranches are certainly less affected by thestorage component in the xylem relative tostems. Thus, although certain branchesmay display large variation on time scalesof 5-20 min, more similar to that mea-sured with eddy correlation (figure 7c), itis difficult to identify the branches whosebehavior might reflect that of the wholecanopy.

4. DISCUSSION

Difficulties in scaling fluxes arise whenvariables that exhibit both high spatial andtemporal variability are sampled [31, 42].For this reason, there is an increased ten-

dency to use methods that integrate over’either’ spatial or temporal variability.Micro-meteorological methods provide afoot-print level of spatial integration [3,23], but their utility for assessing the effectof forest structure on flux is limited, andthey are unsuitable to study different phys-iological responses of co-occurring species

[40]. Therefore, when the purpose of aninvestigation is to assess the effect of vari-ation in environmental conditions on dif-ferent sources and sinks within the canopy,or to calculate canopy stomatal conduc-tance of different components in the forest[3], other approaches must be employed.Sapflow measurements allow evaluation ofspatial variability and, in combination withlocal water balance or micro-meteorolog-ical flux estimates, can partition fluxesamongst distinct sources. These sourcesmay represent strata within the canopy, orpatches of different species within a strata.Partitioning water flux into strata may alsobe accomplished by positioning eddy cor-relation instruments at a desired number oflevels within the canopy [17, 24], how-ever, the compatibility of foot-prints (i.e.similarity in the spatial variation withinsource areas) must be assured before mea-surements from a pair of instrument at dif-ferent heights are compared [39].

4.1. Scaling sap flux and localwater balance

Scaling of sapflow to the stand levelhas often been carried out by first scalingflux measurements to the tree using treecircumference (Cermák-Type sensors; [2,38]), tree sapwood area (heat pulse veloc-ity and Granier-Type sensors; [5, 28, 30,41, 43, 62]), or projected crown area [48].In each of these studies, a relationship wasshown between tree transpiration and oneor more variables describing a measure oftree size, e.g. stem circumference, diame-ter, basal area, sapwood area, projectedcrown area or tree leaf area. This, ofcourse, is to be expected, because all ofthese measures are strongly auto-corre-lated. Because one of these measures isused to scale heat flux or pulse measure-ments to tree sapflow, they are, in turn,auto-correlated with tree transpiration. Theapparently good relationship between ameasure of tree size and tree transpiration,

especially convincing when the range intree size is large, has been used to justifycalculating stand flux by combining thenumber of individuals per hectare in eachsize category with the flux density in arepresentative individual(s) [10, 17, 26].However, this may mask potentiallyknown and accountable sources of vari-

ability in the original sap flux data.

Many investigations have found that,after standardizing tree sapflow by a con-ducting or transpiring surface area (i.e.sapwood or leaf area), or an index of theseareas (e.g. projected crown area), differ-ences in flow among individual trees

decrease, and become generally unrelatedto a measure of tree size [2, 17, 43, 54,58]. In closed stands, some of the resid-ual variability originated from emergent[40, 62] or very suppressed [ 10, 17] posi-tions in the canopy, position in relation toneighboring crowns [2] and competition(figure 2). Here we demonstrate that, evenin a so-called homogeneous plantation ofpine, a patchy distribution in sapwood areaindex (figure 5) causes high variation incompetition among patches, and, in turn,a large variability in sap flux densityamong individuals (figure 1). Large vari-ability in sapwood area index has beenreported in other, apparently homogeneousstands [27]. The standardized flow, or flux,may not be distributed normally, reflecting forest structure ([2]; inset figure 1).

To account for a proportion of the vari-ability associated with canopy positionsof individual crowns, Granier et al. [26]recommended using a proportional sam-pling procedure, whereby trees to be sam-pled are allocated into categories (e.g.diameter classes) of a scaling variable (e.g.tree sapwood area) in proportion to theratio of the sum of the values in the cate-

gory (e.g. total sapwood area in a class)to the sum of values in the stand. This, ora similar procedure has been used by manyinvestigators [10, 13, 15, 40, 54, 62]. Itsmain weakness is that it implies that vari-

ation in JS is related to tree size, althoughit has been shown that JS (unlike JSAS ofindividual trees) is rarely related to size,but is related to exposure and competi-tion. Thus, Cermák [13] showed that ECestimates had appreciably lower standarderrors, and were less prone to systematicerrors, when scaling was based on a solarequivalent leaf area rather than on basalarea, stem volume, projected crown area,projected leaf area, or leaf dry mass.Weighing leaf area of individual trees bythe time integrated relative irradianceaccounted for the variability in JS caused

by exposure, which was only partiallyaccounted for by the other, size-relatedvariables.

A simpler approach is to assume thatEC is the product of two variables, JS and

AS:AG. The variance of EC, &sigma;E, can becalculated after the degree of correlationbetween the two variables is quantified[equation (6)]. The approach does notattempt to reduce the variability causedby a known factor - the degree of expo-sure of individual trees - as recommended

by Cermák [13]. Thus, in stands in whichdistinct strata are formed in the canopy,it may be necessary to perform this scalingprocedure in combination with the

approach outlined by Granier et al. [26],thereby proportionally partitioning sam-ple trees according to relative strata impor-tance, and then accounting for the effect ofcompetition within each stratum. Althoughthis approach is not as elegant as that pro-posed by Cermák [13], it is statisticallysound because it allows error to be prop-agated throughout all steps, and includesthe correlation between variables for vari-ance calculation. In this study, an inversecorrelation between JS and AS:AG (fig-ure 2) resulted in a downward adjustmentin &sigma;E [equation (6)]. Furthermore, theresults shown in figure 2 demonstrate thatthe reason that JS was not distributed nor-

mally (inset in figure 1) is that it wasaffected by another variable, namely, com-

petition. Removing the effect of competi-tion resulted in a normally distributed JSvalue.

There are several other sources of vari-

ability in JS that may require an explicittreatment when &sigma;E is calculated. For

example, it has long been recognized thatin many species there is a radial patternin JS ([25]; see summary in Phillips et al.[51]). In the case of species displaying aradial pattern in JS and a hydroactivexylem, which is greater in depth than thelength of the sensor used, scaling to thestand should account for the variability inJS in each depth interval and variabilityin AS:AG representing each depth inter-val. Eventually, a variance combiningthose of all intervals can be calculated for

canopy transpiration [equation (4)].Recently, investigators using Granier-Typesensors have begun to account for radialpattern in estimating long-term EC by cor-recting the flux in the xylem beyond thereach of sensors by a factor based on ashort-term, unreplicated study [7, 26, 41].The correction was made to permit com-parisons with two Cermák-Type sensors,to estimate canopy transpiration based oneach sensor, and to calculate the number ofsensors necessary to achieve a mean thatfalls within 10 % of the mean of all 24sensors [7, 41]. Such a correction must beperformed with the understanding that itdoes not take into account the dynamicnature of the conducting tissue. Under con-ditions of decreasing water availability,reduction in JS in the inner xylem wasgreater than that in the outer xylem, andoccasionally ceased entirely [5, 51]. Over9 rainless days, the inner xylem JS ofPinus taeda decreased, first gradually andthen rapidly, from 44 to 36 % of the outerxylem [51 ]. Thus, to estimate EC usingsensors that do not transverse the hydroac-tive xylem, it is best to quantify the radialpattern by positioning sensors across thehydroactive xylem [26], and to assess thesensitivity of JS at all hydroactive depths

to soil water availability. Doing so, wedemonstrated in figure 3 that the com-monly assumed uniformity in sap fluxthroughout the xylem can cause largeerrors in estimates of EC, errors thatincrease with decreasing soil moisture.

A key issue in experimental design,with important implications for both dataanalyses and project costs, concerns thenumber of replicated experimental units.Granier et al. [26] summarized the resultsof several studies and showed that thecoefficient of variation (CV) in JS is ca10-15 % in temperate forests, but is muchlarger in tropical forests (35-50 %).Diawara et al. [17] reported a 30-40 %CV (N = 10) in P. pinaster Ait. Further-more, variability in JS increases with

drought, because large trees probably usewater faster, and progress into droughtstress (i.e. low JS) earlier than smallertrees [26, 28, 44]. Variability in JS alsoincreases with manipulation of stand struc-ture and resource availability [44, 58].Using a total of 24 sensors, two variants ofCermák-Type and a Granier-Type, Köst-ner et al. [41 ] concluded that the mean ofrandomly selected subsets of sensors didnot converge towards the overall meanwhen more than eight sensors wereemployed. Granier et al. [26] concludedthat ten sensors should suffice for esti-

mating canopy transpiration in temperateforests. We approached the question dif-ferently, asking how many sensors arenecessary to obtain a reasonable CV?

Choosing a CV of 15 %, we demonstratein table II that the number of sensors nec-

essary to quantify flux at the selected vari-ability is highly variable among species, ishigher for broadleaf than conifer species,and may be doubled by cultural treatments.For P. taeda, the number of sensors nec-essary to obtain the same CV in the inner

xylem was similar to that in the outerxylem, showing a similar variability in JS,as is demonstrated by the high correlationbetween Jin and Jout. However, variabil-

ity in Jin and Jout was not similar inbroadleaf species, where the variabilitymay increase or decrease with depth inthe xylem. Additional sources of varia-tion on sloping sites include a differentdiurnal pattern of JS, depending on theaspect and the azimuth sampled [26]. It is clear that comparison among species, andwithin a species between stands on sitesdiffering in quality, may require a differ-ent number of monitored individuals. Opti-mizing the use of resources, therefore, maynot call for a balanced design with equalnumber of sensors in each forest type orstand.

Capitalizing on the simplifications aris-ing from integration of flux measurementsat the whole-individual level requires adetailed evaluation of potential sources of&sigma;2Js [21]. Not accounting for radial andazimuthal variability may create errors inscaling flux density to the whole tree [23 ];errors which will be carried to canopy-level estimates (figure 3). Additionalerrors, and larger &sigma;E, will result if com-petition and exposure are not considered in

the sample allocation design. It is likelythat in open stands, competition for watermay dominate tree-to-tree variability [17],but that as stands close, and perhapsbecome less coupled with the atmosphere,variation in irradiance, and thus exposure,becomes more important (Jarvis, per.comm.). The use of such information,should provide guidance for appropriatesampling design depending on the objec-tives of each study. If the objective of astudy is to estimate a component flux bydifference, then scaling to the same sourcearea, and minimizing variance, are bothrequisite steps in the process.

4.2. Comparing EC with LWBestimate of ET

In this study, we used two methods toestimate ET (LWB and micro-meteorol-ogy), and one to estimate EC. The LWB issimilar in principle to that used by Rambal[53] and Kelliher et al. [36]. After scalingeach of the estimates, we tested whether

ESC can be estimated from the differencebetween ET and EC. Our diurnal soil mois-ture extraction pattern, obtained with theLWB approach, was very similar to thediurnal of water uptake obtained fromscaled JS (figure 4). Scaling JS was basedon Jin and Jout, and the inner and outersapwood area index of the entire FACEplot [equation (4)]. Thus, only one sourceof variation was included in this scaling,which compared water uptake in the plotwith soil moisture extraction in the same

plot. The sensitivity of the TDR was suf-ficient to detect significant changes in &thetas;

only in the upper 0.1 m of the profile.Higher variability in measurement in lowersoil layers, coupled with smaller changesin &thetas; over the day (figure 4a), preventedthe obvious trend from being significant.Had the changes been significant, ECwould have accounted for 76 % of the

daily soil moisture extraction, implyingthat ca 24 % of ET may be ESC, a similarestimate to that from the difference in eddycorrelation measurements above andbelow the canopy (> 20 %). Using theLWB method, long-term data from boththe FACE and Ref. 2 plots showed a sig-nificantly lower EC relative to ET [59].The difference, reflecting evapotranspira-tion below the canopy, accounted for 36 %in the FACE and 25 % in Ref. 2, at a timewhen L in FACE was ca 15 % lower thanin Ref. 2.

We propose that 1) two estimates offluxes, from the same or different com-ponent, may be compared only if errorsabout the estimates account for the vari-

ability from the same source area; 2)before a difference between two compo-nents is attributed to a third, it is neces-

sary to show first that it is significantlygreater than zero, or, in other words, beforeaccepting a budget as closed, the devia-tion from closure must be shown not todiffer from zero. Employing these crite-ria in the comparison between EC andLWB-estimated ET (figure 4), we con-

eluded that the LWB approach is not suit-able to estimate ET for the total stand or,

by difference with EC, for the subcanopy,on time scales less than 1 day. The con-clusion is based on lack of significant diur-nal difference between the ET and EC esti-

mates, although such differences wereapparent in both long-term LWB basedestimates of ET, and in diurnal and dailyeddy correlation estimates of ET.

Bréda et al. [10] had performed a sim-ilar comparison between water source(rooting volume) and sink (uptake bytrees). Over three study years, wateruptake accounted for ca 0.95 of the soilwater balance. The authors attributed thesmall difference to soil evaporation, per-haps because understory was absent underthe high canopy L (= 6). However, pre-cipitation input and drainage were notexplicitly considered in that study. In amixed stand of mature oaks, Bréda et al.

[11 ], using a similar approach, identifiedthat, in spring, the soil water balanceshowed greater water loss than wasaccounted for by scaled water uptakeobtained with Granier-Type sensors. Inspring, water uptake accounted for 0.73of soil water loss, but the ratio increased to0.85 by early summer, and 1.43 by latesummer. The difference at the beginningof the season was attributed to evapora-tion from the soil. However, as in the pre-vious study, there was no treatment of pre-cipitation input or drainage. It is possiblethat drainage could account partially forthe early season discrepancy, and uptakefrom layers below the measured depth forthe late season discrepancy.

Rambal [53] studied the dynamics ofwater extraction by roots using, in princi-ple, the same approach used in this study.His study demonstrated that water uptakefrom deep horizons increased with

decreasing growing season precipitation;up to 23 % of soil moisture in his ever-

green oak scrub was absorbed below 2 mof soil depth during the growing season.

While theoretically, LWB can be appliedto great depths in the soil [32], the nor-mal functioning of the forest will be dis-rupted if trenches must be dug to installsoil moisture sensors in great depths and innumbers sufficient to quantify the changein &thetas;.

We therefore recommend that the LWBmethod be used for time scales greaterthan a day, unless water uptake is highand concentrated in a thin soil layer, anddrainage is relatively low in comparison tomoisture extraction. In addition, the LWBrequires that the variability in rainfall inputto the soil is considered. In general, themethod is difficult to use in rocky soils,or in vegetation supporting deep roots(> 2 m). The LWB is useful for estimatingET in small plots, and over sloping ground,both situations unsuitable for the applica-tion of micro-meteorological techniquesfor estimating ET.

4.3. Comparing EC with micro-

meteorological estimates of ET

Briefly, micro-meteorological approachesto estimate component flux in forestsinclude: 1) estimating ET based on eddycorrelation measurements above the

canopy; and 2) estimating evapotranspi-ration from the soil, litter, and understorybased on i) lysimeters, ii) eddy correla-tion measurements above the understory,or iii) process or empirical models forestimating the contribution to ET fromeach subcanopy component. In manyinvestigations, ET is estimated above the

canopy and ESC above the understory.Although each estimate typically repre-sents a different source area, and some-times the areas differ in size by severalorders of magnitude, it is common to sub-tract 2) from 1) and attribute the differ-ence to EC. Often, in addition to one orboth of the other components, EC is esti-mated directly, using scaled JS, or indi-

rectly, using models and input of envi-ronmental variables and parameterizationwith porometry data.

When either ESC or ET is measured orestimated with EC, the unmeasured com-ponent is often calculated without con-

sidering source areas. If all three compo-nents are obtained, the closure in thebudget is evaluated [2, 7, 8, 17, 24, 26,36, 37, 39, 41, 64]. The closure is thenjudged based on assumed or estimatederrors [6, 36, 37, 39] or on subjective cri-teria (e.g. difference from closure relativeto ET; [7]). If the closure is consideredunacceptable, the budget is adjusted basedon some post facto criteria [2]. When errorestimates are used, they often 1) includecomponent errors that are based onassumptions; 2) do not attempt to includethe spatial variability from the same sourcearea for all budget components; 3) useensemble variance for one component and

spatial variance for another; and 4) are notthe product of a study design with an apriori objective to obtain reasonably smallstandard errors of estimates. The effect ofnot designing a study so as to minimizethe standard error can be seen in figure 4.Replicates were sufficient to produce avery small standard error in soil moistureof the top layer, but were too few for pro-ducing small errors in deeper layers. Gen-erally, without estimating standard errors,one relies on subjective criteria to judgethe closure in the budget; with inappro-priately estimated standard errors, onemisjudges the closure; and with artificiallylarge standard errors, one may always findclosed budgets.

Comparing EC obtained from JS, withthat using a combination of energy bal-ance and understory lysimeters, Arneth etal. [2] found a difference of ca 30 %between the two estimates. Although stan-dardizing whole-tree water flow by pro-jected canopy area removed systematicvariation in flux among crown classes, theauthors excluded subcanopy individuals

< 110 mm in diameter from EC calcula-tions, which represented 60 % of trees and15 % of the plot basal area. This reducedthe difference between the two estimates

of EC to only 5-10 % which, in turn, wasused to propose that trees of the small sizeclass transpire less per unit of stem area.Furthermore, the authors suggested that,given the 10-20 % coefficient of varia-tion for the two estimates, the differencesbetween them are probably not statisti-cally significant. However, an analysis oftheir results (table III in Arneth et al. [2])showed that, in each of the six study days,estimates of EC based on the two meth-ods differed at P < 0.03 to 0.0001. Fivemeasured variables were involved in theestimation of EC based on the two meth-ods: mass flow rate of xylem sap for a por-tion of tree circumference was combinedwith tree circumference into one variable,mass flow per tree (assuming no circum-ferential variance), and divided by crownarea, to produce mass flow per unit ofcrown area. This facilitated a comparisonwith the following three variables: avail-able energy, sensible heat flux density andsubcanopy evaporation rate. The authorsstated that some uncertainty may be asso-ciated with estimates based on scalingmeasurements and combining estimates.The implicit assumption in comparing thetwo EC estimates is that the four estimatedfluxes represent the same source area, orthat the spatial average is represented bythe source areas of each of the flux vari-ables.

However, the failure to close the budgetwithout excluding the smaller treesdemonstrates that this fundamental

assumption has not been met, as may oftenbe the case [39]. Thus, an alternativechoice may be to ignore the lysimeter-based estimate of subcanopy evaporationbecause, due to its very limited spatialsampling, it may not be representative ofthe mean, and retain the transpiration bysubcanopy trees in the calculations. In

doing so, the estimate of ET is greater thanthe estimate of EC by 0.3 mm on average[2]. The difference is significantly greaterthan zero (P < 0.01 or less for each day),and is thus attributable to the unmeasured

budget component, ESC. If this value bet-ter represents ESC, then the flux estimatedwith lysemeter measurements may haveover-estimated ESC three-fold on average.

We have estimated daily EC based on

Granier-Type sensor measurements,expanded to the stand [equation (5)]. Thevalues were compared to ET obtained with

eddy correlation from a similar source area(figure 6). Except for the day of the lowestmeasured flux, when apparent ECexceeded ET probably due to recharge oflong-term depletion of stem water reserve[50], EC was significantly lower than ET.The slope of the relationship between ECand ET indicates that on average ca 31 %

of ET originates below the canopy, notvery different from the long-term estimateusing LWB (36 %; [59]). However, thedaily ratio indicates that EC / ET can varyfrom 1.13 to 0.57, with an average of 0.73(1 SE = 0.05). Subcanopy contributionvaries appreciably among forests, anddaily within forests, depending on standstructure, species composition, canopy L,soil moisture availability and atmosphericconditions [9, 47]. In certain forests, ESCbecomes a larger proportion of ET as soildries [37].

On a clear day, ET, and JS of stems andbranches showed a similar pattern ofincreased flux in the morning, but the pat-terns diverged thereafter (figure 7b),reflecting the dynamics of hydraulic resis-tance in the stem and of water storage [26].The time constant between transpirationand water uptake has been evaluatedexplicitly to permit estimation of EC frommeasurements of water uptake [43, 51,52]. The effect of the resulting time-lagwill be addressed here only in the contextof its effect on partitioning flux betweencontributing components within forests.

Comparing sapflow-based estimates of ECwith ET diurnally, even when scaled tothe same source area and including appro-priate error calculation, may result in dif-ferences between the two components dur-

ing certain hours, presumably attributableto ESC, but potentially resulting from thetime-lag between the two estimates. Atime-lag not chosen carefully, may artifi-cially reduce the difference at certain timesand increase it at others. The time-lag notonly shifts the course of uptake relativeto EC, but also changes its shape andbuffers the high-frequency fluctuation inEC.We attempted to evaluate the use of

branches, presumably containing little stor-age, for assessing the lag between ET, EC(i.e. JS in branches) and uptake (JS in

stems). Unrelated to their position in thecanopy, four of the six branches began totranspire without a time-lag and twobranches lagged behind the stem (fig-ure 7c). All branches continued to

recharge into the night after transpirationstopped, although they completed therecharge before the stem. Most impor-tantly, however, was the clear asyn-chronous behavior of branch JS duringmost of the day (figure 7c), resulting ina smooth course of mean JS (figure 7b),and indicating that micro-climate vari-ability is large in the canopy. Selectingbranches to monitor in order to follow thediurnal course of EC may not be possible.

Thus far, time-lag in stem JS has beenselected mostly by subjectively lagginguptake relative to ET, radiation, or D, until there appears to be an acceptable match[26, 41]. Other approaches include: across-correlation analyses between uptakeand these variables; resistance-capaci-tance formulations; and, recently, estima-tion of the stem storage dynamics and uti-lizing the information to translate uptaketo EC [43, 51, 52]. Regardless of themethod used, errors in matching uptaketo EC affect not only calculations of

canopy conductance [43, 50], but also con-found the estimation of diurnal ESC. Thus,due to the time-lag between transpirationand uptake, the shortest time interval forsafely estimating EC from JS may be 1 d

[15], and choices made in selecting thetime over which measurements are inte-

grated may also affect the results of com-parison on a daily scale [17].

4.4. EC in relation to estimated soilmoisture depletion

We used our estimate of averageESC = 0.31ET to make a similar analysis ofEC response to soil moisture depletionover a period for which &thetas; was not mea-sured. Over a 9-d drying cycle, weincreased EC to account for the ESC of ca31 % of ET, and expressed EC as a func-tion of cumulative ET (figure 8). We con-sider cumulative ET to only approximatesoil moisture depletion, because, whiledrainage is quite negligible over such ashort period during the growing season[36] - < 1 mm in total (figure 4c; [59]),daily transpiration of understory may beaffected less by soil drought than over-story [9]. The variable contribution of sub-canopy to ET requires a different correc-tion for EC every day. We did not havesuch information, and therefore appliedthe average correction (figure 6) to alldays.

The effect of soil moisture on stomatalconductance and transpiration is well doc-umented [29, 55]. Reduction in soil mois-ture also affects the dynamics of waterflux in stems, and increases the contribu-tion of water stored in stems to EC [43,51]. Reduction in the total conductivityof the soil-tree system with soil moisturewas explained by increased soil-root resis-tance at higher (&thetas; = 0.33; [4]) and lower(&thetas; = 0.17; [57]) levels, perhaps resultingfrom reduced root extension growth [56]and root-soil contact [46].

Daily transpiration is correlated to soilmoisture during short drying periods andseasonally [11, 14, 20, 43, 46, 57, 58].Granier and Loustau [23] showed thatincreasing soil moisture deficit during 8 dresulted in a decrease in EC of maritime

pine to a sixth of the initial value. In standsgrowing over deeper soils and transpiring at lower rates, the rate of decline in ECwith drought may be considerably lower[43].

In this study, approximately 90 % ofthe water used for transpiration is absorbedin the upper 0.35 m of the soil (figure 4);when &thetas; decreases below 0.18, canopy con-ductance decreases rapidly with &thetas; [59].Using the calculated cumulative ET, &thetas; inthe main rooting layer decreased by ca0.07, as EC decreased to 60 % of its orig-

inal value, a much lower sensitivity of ECto change in &thetas; than reported by Granierand Loustau [23]. Because daily EC duringthis period was not related to D (P > 0.3),it is likely that &thetas; was within the range inwhich it strongly influences canopy con-ductance, as demonstrated in Oren et al.[49].

ACKNOWLEDGMENTS

This research was funded by the US Depart-ment of Energy (DOE) through the NationalInstitute for Global Environmental Change(NIGEC) Southeast Regional Center at theUniversity of Alabama, Tuscaloosa (DOECooperative Agreement DE-FC03-

90ER61010). We thank Philip Todd for hisassistance in collecting and processing the timedomain reflectometer data.

REFERENCES

[1] Amoozegar A., A compact constant-head per-meameter for measuring saturated conduc-tivity of the vadoze zone: comparison of theGlover solution with simultaneous-equationsapproach for measuring hydraulic conduc-tivity, Soil Sci. Soc. Am. J. 53 (1989)1356-1361.

[2] Ameth A., Kelliher F.M., Bauer G., HollingerD.Y., Byers J.N., Hunt J.E., McSeveny T.M.,Ziegler W., Vygodskaya N.N., Milukova I.,Sogachov A., Varlagin A., Schulze E.-D.,Environmental regulation of xylem sap flowand total conductance of Larix gmelinii treesin eastern Siberia, Tree Physiol. 16 (1996)247-255.

[3] Baldocchi D.D., Luxmoore R.J., Harfield J.L.,Discerning the forest from the trees: an essayon scaling canopy stomatal conductance,Agric. For. Meteorol. 54 (1991) 197-226.

[4] Barataud F., Moyne C., Bréda N., GranierA., Soil water dynamics in an oak stand. II. Amodel of the soil-root network comparedwith experimental data, Plant and Soil 172(1995) 29-43.

[5] Becker P., Sap flow in Bornean heath anddipterocarp forest trees during wet and dryperiods, Tree Physiol. 16 (1996) 295-299.

[6] Berbigier P., Bonnefond J.M., Loustau D.,Ferreira M.I., David J.S., Pereira J.S., Tran-spiration of a 64-year-old maritime pine standin Portugal. 2. Evapotranspiration and canopystomatal conductance measured by an eddycovariance technique, Oecologia 107 (1996)43-52.

[7] Bernhofer Ch., Gay L.W., Granier A., JossU., Kessler A., Köstner B., Siegwolf R., Ten-hunen J.D., Vogt R., The HartX-Synthesis:An experimental approach to water and car-bon exchange of a Scots pine plantation,Theor. Appl. Climatol. 53 (1996) 173-183.

[8] Bernhofer Ch., Blanford J.H., Siegwolf R.,Welder M., Applying single and two layermodels to derive conductances of Scots pineplantation from micrometeorological mea-surements, Theor. Appl. Climatol. 53 (1996)95-104.

[9] Black T.A., Kelliher F.M., Processes con-trolling understory evapotranspiration, Phil.Trans. Soc. Lond. B 324 (1989) 207-231.

[10] Bréda N., Cochard H., Dreyer E., Granier A.,Water transfer in a mature oak stand (Quercuspetraea):seasonal evolution and effects ofsevere drought, Can J. For. Res. 23 (1993)1136-1143.

[11] Bréda N., Granier A., Barataud F., MoyneC., Soil water dynamics in an oak stand. I.Soil moisture, water potentials and water

uptake by roots, Plant and Soil 172 (1995)17-27.

[12] Cassel D.K., Kachanoski R.G., Topp G.C.,Practical consideration for using a TDR cabletester, Soil Technol. 7 (1994) 113-126.

[13] Cermák J., Solar equivalent leaf area: an effi-cient biometrical parameter of individualleaves, trees and stands, Tree Physiol. 5(1989) 269-289.

[14] Cermák J., Huzulák J., Penka M., Waterpotential and sap flow rate in adult trees withmoist and dry soils as used for the assess-ment of root system depth, Biol. Plantarum(Praha) 22 (1980) 34-41.

[15] Cienciala E., Lindroth A., Cermák J., Häll-gren J.-E., Kucera J., The effects of wateravailability on transpiration, water potentialand growth of Picea abies during a growing season, J. Hydrol. 155 (1994) 57-71.

[16] Clapp R.B., Hornberger G.M., Empiricalequations for some soil hydraulic properties,Water Resour. Res. 14 (1978) 601-604.

[17] Diawara A., Loustau D., Berbigier P., Com-parison of two methods for estimating evap-oration of a Pinus pinaster (Ait.) stand: Sapflow and energy balance with sensible heatmeasurements by an eddy correlation method,Agric. For. Meteorol. 54 (1991) 49-66.

[18] Eagleson P.S., Climate soil, and vegetation: 1.Introduction to water balance dynamics,Water Resour. Res. 14 (1978) 705-712.

[19] Ellsworth D.S., Oren R., Huang C., PhillipsN., Hendrey G.R., Leaf and canopy responsesto elevated CO2 in a pine forest under free-airCO2 enrichment, Oecologia 104 (1995)139-146.

[20] Goulden M.L., Carbon assimilation andwater-use efficiency by neighboring Mediter-ranean-climate oaks that differ in wateraccess, Tree Physiol. 16 (1996) 417-424.

[21] Goulden M.L., Field C.B., Three methods formonitoring the gas exchange of individualtree canopies: ventilated-chamber, sap-flowand Penman-Monteith measurements on ever-

green oaks, Funct. Ecol. 8 (1994) 125-135.

[22] Granier A., Une nouvelle méthode pour lamesure de flux de sève brute dans le troncdes arbes, Ann. Sci. For. 42 (1985) 193-200.

[23] Granier A., Loustau D., Measuring and mod-eling the transpiration of a maritime pinecanopy from sap flow-data, Agric. For. Mete-orol. 71 (1994) 61-81.

[24] Granier A., Bobay V., Gash J.H.C., Gelpe J.,Saugier B., Shuttleworth W.J., Vapor fluxdensity and transpiration rate comparisons ina stand of maritime pine (Pinus pinaster Ait.)in Les Andes Forest, Agric. For. Meteorol.51 (1990) 309-319.

[25] Granier A., Anfodillo T., Sabatti M., CochardH., Tomasi M., Valentini R., Bréda N., Axial

and radial water flow in the trunk of oak trees;

a quantitative and qualitative analysis, TreePhysiol. 14 (1994) 1383-1396.

[26] Granier A., Biron P., Bréda N., Pontailler J.-Y., Saugier B., Transpiration of trees and for-est stands: short and long-term monitoringusing sapflow methods, Global Change Biol.2 (1996) 265-274.

[27] Granier A., Biron P., Köstner B., Gay L.W.,Najjar G., Comparisons of xylem sap flowand water vapour flux at the stand level andderivation of canopy conductance for Scots

pine, Theor Appl. Climatol. 53 (1996)115-122.

[28] Hatton T.J., Wu H.-I., Scaling theory toextrapolate individual tree water use to standwater use, Hydrol. Processes 9 (1995)527-540.

[29] Hinckley T.M., Running S.W., Lassoie J.P.,Temporal and spatial variations in the waterstatus of forest trees, For. Sci. Monogr. 20(1978) 1-72.

[30] Honeysett J.L., Beadle C.L., Turnbull C.R.A.,Evapotranspiration and growth of two con-trasting species of eucalyptus under non-lim-iting and limiting water availability, For. Ecol.Manag. 50 (1992) 203-216.

[31] Jarvis P.G., Scaling processes and problems,Plant Cell Environ. 18 (1995) 1079-1089.

[32] Jipp P.H., Nepstad D.C., Cassel D.K., Reis-de-Carvalho C., Deep soil moisture and tran-spiration in forests and pastures of season-ally-dry Amazonia, Climate Change (1998).

[33] Katul G.G., Wendroth O., Parlange M.B.,Puente C.E., Nielsen D.R., Estimation of insitu hydraulic conductivity function from non-linear filtering theory, Water Resour. Res. 29(1993) 1066-1070.

[34] Katul G., Hsieh C.-I., Oren R., Ellsworth D.,Phillips N., Latent and sensible heat flux pre-dictions from a uniform pine forest using sur-face renewal and flux variance methods,Boundary-Layer Meteorol. 80 (1996)249-282.

[35] Katul G., Todd Phillip, Pataki D., Kabala Z.J.,Oren R., Soil water depletion by oak treesand the influence of root water uptake on themoisture content spatial statistics, WaterResour. Res. 33 (1997) 611-623.

[36] Kelliher F.M., Black T.A., Price D.T., Esti-

mating the effects of understory removal fromDouglas-fir forest using two-layer canopyevapotranspiration model, Water Resour. Res.22 (1986) 1891-1899.

[37] Kelliher F.M., Whitehead D., McAneney K.J.,Judd M.J., Partitioning evapotranspirationinto tree and understory components in twoPinus radiata D. Don stands, Agric. For.Meteor. 50 (1990) 211-227.

[38] Kelliher F.M., Köstner B., Hollinger D.Y.,Byers J.N., Hunt J.E., McSeveny T.M.,Meserth R., Weir P.L., Schulze E.-D., Evap-oration, xylem sap flow, and tree transpirationin a New Zealand broad-leaved forest, Agric.For. Meteor. 62 (1992) 53-73.

[39] Kelliher F.M., Hollinger D.Y., Schulze E.-D., Vygodskaya N.N., Byers J.N., Hunt J.E.,McSeveny T.M., Evaporation from easternSiberian larch forest, Proc. Int. Symp. For-est Hydrol., Tokyo, Japan, 1994.

[40] Köstner B., Schulze E.-D., Kelliher F.M.,Hollinger D.Y., Byers J.N., Hunt J.E., McSev-eny T.M., Meserth R., Weir P.L., Transpira-tion and canopy conductance in a pristinebroad-leaved forest of Nothofagus: an anal-ysis of sap flow and eddy correlation mea-surements, Oecologia 91 (1992) 350-359.

[41] Köstner B., Biron P., Siegwolf R., GranierA., Estimating water vapor flux and canopyconductance of Scots pine at the tree levelutilizing different xylem sap flow methods,Theor. Appl. Climatol. 53 (1996) 105-113.

[42] Leverenz J., Deans J.D., Ford E.D., JarvisP.G., Milne R., Whitehead D., Systematicspatial variation of stomatal conductance inSitka spruce plantation, J. Appl. Ecol. 19(1982) 835-851.

[43] Loustau D., Berbigier P., Roumagnac P.,Arruda-Pacheco C., David J.S., Ferreira M.I.,Pereira J.S., Tavares R., Transpiration of a64-year-old maritime pine stand in Portugal.1. Seasonal course of water flux through mar-itime pine, Oecologia 107 (1996) 33-42.

[44] Lu P., Biron P., Bréda N., Granier A., Waterrelations of adult Norway spruce (Picea abies(L) karst) under soil drought in the Vosgesmountains: water potential, stomatal con-ductance and transpiration, Ann. Sci. For. 52( 1995) 117-129.

[45] Nikodem H.J., Effects of soil layering on theuse of VHF radio waves for remote terrain

analysis, Proc. 4th Symp. Remote SensingEnv., Univ. Mich. Ann. Arbor., 1966, pp.691-703.

[46J Nnyamah J.U., Black T.A., Tan C.S., Resis-tance to water uptake in a Douglas-fir forest,Soil Sci. 126 (1978) 63-76.

[47] Oren R. Waring R.H., Stafford S.G., BarrettJ.W., Twenty-four years of Ponderosa pineGrowth in relation to canopy leaf area and

understory competition, For. Sci. 33 (1987)538-547.

[48] Oren R., Zimmermann R., Terborgh J., Tran-spiration in upper Amazonia floodplain andupland forests in response to drought break-ing rains, Ecology 77 (1996) 968-973.

[49] Oren R., Ewers B.E., Phillips N., Todd P.,Katul G., Soil moisture affects canopy con-

ductance at depths delineated with local waterbalance. Ecol. Appl. (1998).

[50] Phillips N., Oren R., A comparison of dailyrepresentations of canopy conductance basedon two conditional time-averaging methods,Ann. Sci. For. (1998) 217-235.

[51] Phillips N., Oren R., Zimmermann R., RadialPatterns of xylem sap flow in non-, diffuse-and ring porous tree species, Plant Cell Env-iron. 19 (1996) 983-990.

[52] Phillips N., Nagchaudhuri A., Oren R., KatulG.G., Time constant for water uptake inloblolly pine trees estimated from time seriesof stem sapflow and evaporative demand,Trees 11 (1998) 412-419.

[53] Rambal S., Water balance and pattern of rootwater uptake by a Quercus coccifera L. ever-green scrub, Oecologia 62 (1984) 18-25.

[54] Sala A., Smith S.D., Devitt D.A., Water useby Tamarix ramossisima and associatedphreatophytes in a Mojave Desert floodplain,Ecol. Appl. 6 (1996) 888-898.

[55] Schulze E.-D., Hall A.E., Stomatal responses,water loss and CO2 assimilation rates of plantsin contrasting environments, EncyclopediaPlant Physiology, New Series 12B, Springer-Verlag, Berlin, 1982, pp. 181-229.

[56] Squire R.O., Attiwill P.M., Neales T.F.,Effects of changes of available water andnutrients on growth, root development andwater use in Pinus radiata seedlings, Aust.For. Res. 17 (1987) 99-111.

[57] Tan C.S., Black T.A., Factors affecting thecanopy resistance of Douglas-fir forest,Boundary-Layer Meteorol. 10 (1976)475-488.

[58] Teskey R.O., Sheriff D.W., Water use byPinus radiata trees in a plantation, Tree Phys-iol. 16 (1996) 273-279.

[59] Todd P.H., Use of a local mass balancemethod to determine the effects of soil waterconditions on canopy conductance and to esti-mate forest stand evapotranspiration in a nat-ural forest and in growth chamber, MS The-sis, Duke University, Durham, NC, USA.

[60] Topp G.C., Davis J.L., Measurement of soilwater content using time domain reflectom-etry (TDR): a field evaluation, Soil Sci. Soc.Am. J. 49 (1985) 19-24.

[61] Topp G.C., Davis J.L., Annon A.P., Electro-magnetic determination of soil water content:measurements in coaxial transmission lines,Water Resour. Res. 16 (1980) 547-582.

[62] Vertessy R.A., Benyon R.G., O’Sullivan S.K.,Gribben P.R., Relationships between stemdiameter, sapwood area, leaf area and tran-spiration in a young mountain ash forest, TreePhysiol. 15 (1995) 559-567.

[63] Vogt K.A., Grier C.C., Gower S.T., SprugelT.G., Vogt D.J., Overestimation of net rootproduction: A real or imaginary problem?Ecology 67 (1986) 577-579.

[64] Wedler M., Heindl B., Hahn S., Köstner B.,Bernhofer Ch., Tenhunen J.D., Model-basedestimates of water loss from "patches" of theunderstory mosaic of the Hartheim Scots pineplantation, Theor. Appl. Climatol. 53 (1996)135-144.

[65] Werk K.S., Oren R., Schulze E.D., Zimmer-mann R., Meyer J., Performance of Piceaabies (L.) Karst. Stands at different stages ofdecline. III. Canopy transpiration of greentrees, Oecologia 76 (1988) 519-524.

[66] Wendroth O., Katul G.G., Parlange M.B.,Puente C.E., Nielsen D.R., A nonlinear fil-tering approach for determining hydraulicconductivity functions in field soils, Soil Sci-ence 156 (1993) 293-301.