Estimating components of forest 
evapotranspiration: A footprint approach for 
scaling sap flux measurements
Authors: A. Christopher Oishi, Ram Oren, and Paul 
C. Stoy
NOTICE: this is the author’s version of a work that was accepted for publication in Agricultural and 
Forest Meteorology. Changes resulting from the publishing process, such as peer review, editing, 
corrections, structural formatting, and other quality control mechanisms may not be reflected in 
this document. Changes may have been made to this work since it was submitted for publication. 
A definitive version was subsequently published in Agricultural and Forest Meteorology, VOL# 
148, ISSUE# 11, (October 2018), DOI# 10.1016/j.agrformet.2008.06.013.
Oishi, A. Christopher, Ram Oren, and Paul C. Stoy. “Estimating Components of Forest 
Evapotranspiration: A Footprint Approach for Scaling Sap Flux Measurements.” Agricultural and 
Forest Meteorology 148, no. 11 (October 2008): 1719–1732. doi:10.1016/j.agrformet.2008.06.013.
Made available through Montana State University’s ScholarWorks 
scholarworks.montana.edu 
Estimating components of forest evapotranspiration:
A footprint approach for scaling sap flux measurements
A. Christopher Oishi a,*, Ram Oren a, Paul C. Stoy a,b
aNicholas School of the Environment and Earth Sciences, Duke University, Box 90328, Durham, NC 27701, USA
b School of GeoSciences, Department of Atmospheric and Environmental Science, University of Edinburgh, Edinburgh EH9 3JN, UK
a b s t r a c t
Forest evapotranspiration (ET) estimates that include scaled sap flux measurements often
underestimate eddy covariance (EC)-measured latent heat flux (LE). We investigated poten-
tial causes for this bias using 4 years of coupled sap flux and LE measurements from a
mature oak-hickory forest in North Carolina, USA. We focused on accuracy in sap flux
estimates from heat dissipation probes by investigating nocturnal water uptake, radial
pattern in flux rates, and sensor-to-stand scaling. We also produced empirical functions
describing canopy interception losses (measured as the difference between precipitation
and throughfall) and soil evaporation (based on wintertime eddy covariance fluxes minus
wintertime water losses through bark), and added these components to the scaled sap flux
to estimate stand evapotranspiration (ETS). We show that scaling based on areas in which
the leaf area index of predominant species deviates from that of the EC footprint can lead to
either higher or lower estimate of ETS than LE (i.e. there is no bias).We found that accounting
for nocturnal water uptake increased the estimate of growing season transpiration by an
average of 22%, with inter-annual standard deviation of 4%. Annual ETS estimate that
included sap flux corrected for nocturnal flux and scaled to the EC footprint were similar to
LE estimates (633  26 versus 604  19 mm, respectively). At monthly or shorter time scales,
ETS was higher than LE at periods of low flux, similar at periods of moderate flux, and lower
at periods of high flux, indicating potential shortcomings of both methods. Nevertheless,
this study demonstrates that accounting for the effects of nocturnal flux on the baseline
signal was essential for eliminating much of the bias between EC-based and component-
based estimates of ET, but the agreement between these estimates is greatly affected by the
scaling procedure.1. Introduction
Stand-level water vapor fluxes are now monitored across
many ecosystems with eddy covariance systems, providing
continuous, long-term measurements of latent heat flux (LE);
however, this approach does not quantify the individual
components of evapotranspiration: interception of precipita-
tion during rain events (IC), evaporation from the soil andforest floor (ES), and transpiration (EC). (See Table 1 for a full list
of abbreviations.) Quantifying these components is anessential
step in assessing andmodeling the processes controlling these
physiological and ecosystem fluxes. A common approach for
estimating EC is the scaling of sap flux density (JS) measured
with the popular thermal dissipation probes (Granier, 1987;
Oren et al., 1998b). Correct applications of these probes can
provide reliable estimates of species-specific transpiration at
Table 1 – List of abbreviations with definitions and units
Abbreviation Definition Units
AB Basal area of trees per unit ground area cm
2 m2
AFj Integrated area beneath fitted curve of radial sap flux profile (see Eq. (4)) cm
2
AS Sapwood area of trees per unit ground area cm
2 m2
ASi AS for species i cm
2 m2
ASih ASi for one-hectare plot cm
2 m2
ASj Sapwood area for individual tree cm
2 m2
cj Distance from center of tree to centroid of fitted curve of radial sap flux
profile (see Eq. (4))
cm
D Vapor pressure deficit kPa
DZ Day-length-normalized vapor pressure deficit kPa
DBH Tree diameter at breast height cm
EC Canopy transpiration mm time
1
ECi EC for species i mm time
1
ECih ECi for one-hectare plot mm time
1
ES Soil surface evaporation mm time
1
ET Evapotranspiration mm time1
ETS Evapotranspiration, estimated from sap flux-scaled budget mm time
1
IC Canopy interception mm time
1
JS Sap flux density g H2O m
2 s1
JSi Sap flux density for species i g H2O m
2 s1
LAI Leaf area index m2 m2
LE Latent heat fluxa mm time1
P Precipitation mm time1
PT Throughfall (P  IC) mm time1
PAR Photosynthetically active radiation mmol m2 s1
RH Relative humidity %
SLA Specific leaf area cm2 g1
TA Air temperature 8C
TB Bark thickness mm
TSW Sapwood thickness cm
Vj Volume of a rotated geometric solid (see Eq. (4)) cm
3
DT Temperature difference between heated and unheated sap flux probes mV
DTmax Maximum daily DT mV
u Volumetric soil moisture content m3 m3
a LE is commonly expressed in terms of Wm2 but can be converted to units of mm by considering the latent heat of vaporization and the
density of air for studies of the water balance.the stand level (Clearwater et al., 1998; Ford et al., 2007; Lu et al.,
2004;Orenet al., 1998b; Phillips et al., 1996;Williamsetal., 2004).
Although some studies have found good agreement
between component-based estimates of evapotranspiration
(ETS), including sap flux-based EC, and LE (Arneth et al., 1996;
Granier, 1987; Granier et al., 2000; Ko¨stner et al., 1992), others
have found that thermal dissipation probes may under-
estimate high flux rates, generally leading to ETS that is lower
than LE (Bovard et al., 2005; Hogg et al., 1997; Scha¨fer et al.,
2002;Wilson et al., 2001). This discrepancymay be the result of
three methodological challenges: (1) improper processing of
the sap flux sensor output, including failure to account for
non-zero nocturnal water uptake, (2) failing to scale sap flux
measurements to a similar footprint as the LE measurements,
and (3) failing to accurately quantify all components of
evapotranspiration when comparing ETS to LE.
Some recent research has focused on (1), demonstrating
that accurate stand-level hydrologic budgetsmust account for
nocturnal sap flux, used either to recharge storage (Daley and
Phillips, 2006; Ko¨stner et al., 1992; Meinzer et al., 2001; Phillips
et al., 1996) or to provide for water loss from leaves
maintaining finite stomatal conductance at night (Daley and
Phillips, 2006; Dawson et al., 2007; Oren et al., 1999). There is
some evidence that nocturnal sap flow observed in data fromheat pulse velocity sensorsmay have beenmissed in data from
thermal dissipation sensors (Hogg et al., 1997), possibly because
of incorrect signal processing (Luet al., 2004). Inorder toprocess
data from thermal dissipation sensors, Lu et al. (2004) pointed
out that the baseline connecting points where zero flux occurs
(DTmax, see Eq. (3) in Section 2) must be dynamic, reflecting
changes in sapwood moisture content, and might not be
reached every night. Mischaracterizing this baseline not only
results in missed nocturnal water uptake, but translates to a
large underestimate of daytime transpiration.
The second methodological challenge, scaling to the
appropriate footprint, requires that probes are installed
properly in sapwood (Clearwater et al., 1998), whole-tree
transpiration estimates account for radial variability in flow
for a sufficient number of sample trees (Ford et al., 2004;
Phillips et al., 1996), and species-specific sapwood area within
the reference footprint is characterized accurately (Wulls-
chleger et al., 2001). Finally, because other evaporative fluxes
may contribute to nearly half of evapotranspiration (Table 2),
any comparison of ETS and LE requires reliable estimates of ES
and IC, yet ES is not often measured directly and is difficult to
model (Wilson et al., 2000).
Here, we investigate the relative contributions of the first
two methodological challenges to the discrepancy between
Table 2 – Components of forest evapotranspiration from published studies using thermal dissipation probes in
comparable regional deciduous forests and from this study
Site description Year P IC ES EC ETS ET LOC EC/ET Reference
Annual sums
Eddy covariance-generated
estimates, same study area
2002 1092 440 610 0.72 Stoy et al. (2006)
2003 1346 410 580 0.71
2004 992 460 640 0.72
2005 934 460a 640 0.72
Upland oak-dominated
broadleaf forest,
Oak Ridge, TN
1998 1225 104 86 230 420 547, 502b 127, 82 0.42, 0.46 Wilson et al. (2001)
1999 1152 105 91 269 465 605, 642b 140, 177 0.44, 0.42
Upland oak-dominated
broadleaf forest,
Oak Ridge, TN
2000 766 325 Wullschleger
and Hanson (2006)2001 539 309
2002 730 255
2003 968 315
This study 2002 1092 189 84 336 610 577 33 0.58 This study
2003 1346 236 102 329 668 618 50 0.53
2004 992 181 108 346 635 618 18 0.56
2005 934 157 119 343 619 605 15 0.57
Growing season
Different plots within
same study area
1997 626 88 264 Pataki and
Oren (2003)
Upland hardwood stand,
Duke Forest, NC
1993 642 90 278 Pataki and
Oren (2003)
Upland oak-dominated
broadleaf forest,
Oak Ridge, TN
1996 267 Wullschleger
et al. (2001)
This study 2002 610 80 68 306 453 505 52 0.67 This study
2003 859 123 69 299 491 531 40 0.61
2004 720 145 72 311 529 525 4 0.59
2005 426 34 88 311 433 517 84 0.72
Annual and growing season values of precipitation (P), canopy interception (IC), soil evaporation (ES), canopy transpiration (EC), sap-flux-based
canopy evapotranspiration (ETS), ET estimated through other means (eddy covariance as LE unless noted), lack of closure (LOC) between ETS
and LE, and the proportion of EC to ET. All values are mm year
1, except EC/ET which is unitless.
a EC modeled as component of LE.
b ET estimated through catchment water balance.component-based and eddy covariance-based ET estimates,
after carefully quantifying the evaporation components of
ETS. We developed estimates of evapotranspiration in a
mature oak-hickory forest in the southeastern U.S. over a 4-
year period (2002–2005), which included both a severe drought
year and a very wet year. Tree-level sap flux was monitored
with thermal dissipation probes, corrected for nighttime
fluxes, and scaled to the stand level, accounting for radial
patterns, tree size, and species distribution within the eddy
covariance footprint and subplots therein, allowing us to
separate the effects of signal processing fromscaling. Sapflux-
scaled transpiration was combined with measured and
modeled evaporative losses (IC and ES, respectively), thus
accounting for all components of evapotranspiration (ETS), to
permit a proper comparison with LE measurements.2. Materials and methods
2.1. Setting
The study was conducted at the Duke Forest Ameriflux Hard-
wood site, Orange County, North Carolina (36858041.43000N,79805039.08700W). The forest stand is comprised of mixed hard-
wood species with a maximum age of ca. 80–100 years. Mean
canopy height is 25 m with emergent crown tops extending
above 35m. The stand is dominated by hickories (Carya
tomentosa (Poir.) Nutt., C. glabra (P. Mill). Sweet.), yellow poplar
(Liriodendron tulipifera L.), sweetgum (Liquidambar styraciflua L.),
and oaks (Quercus alba L., Q. michauxii Nutt., Q. phellos L.). Other
species that contribute to the mid- and under-story include
Carpinus caroliniana Walt., Ostrya virginiana (P. Mill.) K. Koch.,
Ulmus sp., Cornus florida L., and Cercis canadensis L. Coniferous
species including Pinus taeda L. and Juniperus virginiana L. make
upaminorcomponentoftheover-andunderstory, respectively.
Long-term (115-year)meanannual precipitation for the area
is 1146 (166)mm,with 630 (133)mmoccurring betweenApril
and September (www.ncdc.noaa.gov/). The soil is an Iredell
gravely loam and topography is relatively flat with <4% slope.
Theupper 35 cm is a clay loamwith a porosity of 0.54 m3 m3. A
clay pan with low hydraulic conductivity limits the majority of
the rootingzone toapproximately35 cm(Orenetal., 1998a). Soil
depth canbeasdeepas 2 m (Richter, personal communication),
which overlays bedrock. The site has been the subject of a
previous investigation on the transpiration of several canopy
and sub-canopy species (Pataki and Oren, 2003).
Fig. 1 – Estimated leaf area index (LAI, m2 mS2) for the ca.
6.25 ha overlapping most of the eddy covariance-based
flux footprint at the Duke Ameriflux Hardwood Forest near
Durham, NC. Black circles represent the boundaries of the
wetter (to the west) and drier sap flux plots. The blue
diamond delineates the area of the one-hectare plot. Gray
squares represent the location of litter traps. Isometric
lines represent the probability distribution of the peak of
the source-weight function of acceptable eddy covariance
flux measurements estimated using the semi-empirical
footprint model of Hsieh et al. (2000). The peak of the
source weight function lies within the white line 50% of
the 2002–2005 measurement period, and within the green
line 95% of the period. In November 2002, a clearcut was
created to the south of the tower, outside of the study area.
Fluxes originating from this area were excluded from flux
estimates as described by Stoy et al. (2006).2.2. Monitoring design and biometric measurements
An area of approximately 6.25 ha around the AmeriFlux tower
(represented by LAI shading in Fig. 1) was identified for this
study because it included most of the dynamic flux footprintTable 3 – Allometric relationships from data collected to estima
(DBH, cm) using either an exponential function (a T expbTDBH)
Bark thickness
n Function a
Diffuse porous
L. tulipifera 19 Exp 9.687
L. styraciflua 19 Exp 3.580
Ring porous
All Carya 11 Linear 0.105
Q. alba and Q. michauxii
Q. phellos
Combined Quercus 22 Linear 0.199
Relationships for sapwood area are estimated using a power function (a(Stoy et al., 2006), estimated using the semi-empiricalmodel of
Hsieh et al. (2000).Within this area, two 25 m radius plotswere
established for the sap flux study. These two circular plots,
henceforth the ‘sap flux plots’, were chosen to represent a wet
(to the west) and a dry micro-site. Species and diameter at
breast height (1.45 m aboveground; DBH), down to aminimum
diameter of 40 mm, were recorded in the two sap flux plots,
and in an entire hectare surrounding the tower (henceforth
the ‘hectare plot’). Bark thickness (TB) was measured on
several trees and estimated for each individual using the best
fit (linear or exponential) with DBH for each species or genus
(Table 3). Cross-sectional sapwood area for individual trees
(ASj) was estimated from tree cores of sapwood depth (TSW)
taken at the site, using the equation:
ASj ¼ p
DBH
2
 TB
 2
 p DBH
2
 TB  TSW
 2
(1)
where DBH, TB, and TSW are in cm. A generalized estimation of
ASj for each species was developed using:
ASj ¼ aDBHb (2)
where a and b are empirical parameters (Table 3).
Leaf litter was collected in 48 baskets, each with an area of
0.5 m2. Eight baskets were positioned in a circular arrange-
ment, 15 m from the tower in primary and secondary compass
directions. Beyond this 15 m circle, in the S, SW, and W
directions, seven baskets were placed at 30 m intervals along
transects (Fig. 1), sampling the area most commonly within
the tower’s footprint (Geron et al., 1997; Stoy et al., 2006). Each
sap flux plot also contained 10 baskets, with the second basket
along the westward transect doubling as 1 of the 10 in the
western plot. Leaves were collected as often as every 2 weeks
when litterfall was heaviest, and sorted by species. One-sided
surface area of 20-leaf sub-samples of each species was
measured using a Digital Image Analysis System (DIAS,
Decagon Devices, Inc., Pullman, WA, USA), and the weight
of each leafwas obtained after drying (70 8C for 48 h). A specific
leaf area (SLA) of each species was estimated using a linear
regression of leaf surface area versus mass with a zero-
intercept ( p < 0.001, Table 4). Leaf area index (LAI) was
estimated by multiplying SLA for each species by the total
mass of leaves for that species after similar drying (Table 4).te bark thickness (mm) based on diameter at breast height
or linear function (aT DBH + b)
Sapwood area regression
b r2 n a b r2
0.019 0.783 15 0.382 2.010 0.986
0.021 0.549 16 0.246 2.202 0.971
3.070 0.526 15 1.499 1.669 0.944
11 0.612 1.737 0.987
9 0.284 1.932 0.948
1.322 0.826
 DBHb).
Table 4 – Basal area (AB) and sapwood area (AS) in cm
2 mS2 ground area for the dry and wet sap flux plots, the hectare plot
surrounding the eddy covariance tower, and for the kriged area representing the footprint for measured latent heat flux
(LE)
Dry plot Wet plot Hectare plot LE footprint SLA
AB AS AB AS AB AS AS LAI
Diffuse porous
L. tulipifera 9.80 4.97 3.22 1.63 6.29 3.18 2.14 0.83 161.8
L. styraciflua 5.38 3.52 6.89 4.79 4.53 2.96 3.43 0.69 102.4
Mixed species 4.33 2.52 3.25 1.87 4.17 2.35 2.44 1.78 173.3
Ring porous
All Carya 8.03 4.44 4.76 2.78 6.20 3.41 2.91 2.14 131.9
All Quercus 4.59 1.34 14.61 4.30 4.67 1.39 3.21 1.27
Q. alba 4.59 1.34 1.08 0.34 1.73 0.52 0.57 112.2
Q. michauxii 0 0 6.75 2.08 1.95 0.59 0.59 128.9
Q. phellos 0 0 6.77 1.88 0.99 0.27 0.11 101.9
Mixed species 0.17 0.07 3.48 1.14 1.28 0.43 0.45 0.33
Plot total 32.35 16.86 36.21 16.51 27.13 13.70 11.37 7.03
Leaf area index (LAI, m2 m2) derived from specific leaf area (SLA, cm2 g1) was used to scale AS across the LE footprint (see Fig. 1).
Table 5 – Diameter at breast height (DBH, cm) and
maximum sap flux sensor depth (mm) for trees sampled
at the wet and dry sites
Dry site DBH Depth Wet site DBH Depth
L. tulipifera 65.4 40–60 L. tulipifera 59.8 40–60
44.4 40–60 38.7 20–40
42.0 40–60 37.7 20–40
26.4 20–40 36.0 20–40
16.1 20–40a 26.6 20–40
L. styraciflua 48.0 40–60 L. styraciflua 55.6 20–40
47.4 40–60 42.8 20–40
35.0 40–60 37.5 20–40
24.3 20–40 34.4 20–40
19.7 20–40 21.1 20–40
C. tomentosa 58.4 40–60 Q. michauxii 54.4 20–40a
54.4 40–60 47.6 20–40a
25.1 20–40 30.2 20–40a
20.0 20–40 20.1 0–20
12.7 0–20 16.1 0–20
Q. alba 57.7 20–40a Q. phellos 63.6 20–40
43.1 20–40a 53.5 20–40
30.8 0–20 44.0 20–40a
16.4 0–20 43.2 20–40a
13.7 0–20 43.1 20–40a
a Deepest sensor required correction due to contact with non-
conductive tissue (Clearwater et al., 1998).2.3. Environmental measurements
Air temperature (TA) and relative humidity (RH) were
measured at two-thirds canopy height using HMP35C Ta/RH
probes (Campbell Scientific, Logan, UT, USA) and were used to
calculate vapor pressure deficit (D). Photosynthetically active
radiation (PAR) and net radiation were measured above the
canopy at 42 m (see Stoy et al., 2006). Precipitation (P) was
measured daily with a rain gauge and partitioned over half-
hourly values using data from tipping buckets (TR-525USW,
Texas Electronics, Dallas, TX, USA) positioned at the Duke
FACE site, <1 km away. Throughfall (PT) was measured with 6
rain gauges on the forest floor, manually collected once or
twice per week.
Soilmoisture (u, m3 m3) wasmeasuredwith 12 ThetaProbe
sensors (Delta-T Devices, Cambridge, UK), four in each of the
wet and dry sap flux plots and four next to the eddy covariance
tower; at each location two were installed at 5–10 cm depth
and two at 20–25 cm. Data were filtered for unrealistic spikes
after rain events. Missing data, due to power outages in one of
the plots or sensor failure, were gap-filled using the best linear
regressions with other working sensors. The regressions were
comprised of data on both sides of the gap, equal to the length
of the gap in each direction. Periods where u reached
saturation (0.54 m3 m3) or the hygroscopic minimum
(0.125 m3 m3) were identified and the recorded u values for
each sensor were rescaled to match these values (Scha¨fer
et al., 2002).
2.4. Sap flux measurement
Granier-type, heat dissipation sensors were used to monitor JS
(Granier, 1987). Each pair of sensors was 20 mm in length and
the heated element received a constant power of 0.2 W. Five L.
tulipifera and L. styraciflua were equipped in each plot. In
addition, five C. tomentosa and Q. albawere equipped in the dry
sap flux plot, and fiveQ. michauxii andQ. phellos in the wet plot.
These species were selected because they comprised the
majority of sapwood area at the site, and because fluxesmonitored on several other species in this and a nearby site
were similar for a given xylem type (i.e. within ring- or diffuse-
porous groups; Oren and Pataki, 2001; Pataki and Oren, 2003;
Wullschleger et al., 2001). To quantify the radial profile of sap
flux density, sensors were installed at 20-mm depth intervals
based on the expected sapwood depth. Tree DBH and sensor
depths are listed in Table 5.
Sap flux sensors measure the temperature differential (DT)
between the paired heated and unheated probes. DT (recorded
inmV) for each sensor pair wasmeasured at 30 s intervals and
30 min averages were stored on a CR23X datalogger (Campbell
Scientific, Logan, UT, USA). To convert these data into water
flux, the following equation is used:
JS ¼ 119
DTmax
DT
 1
 1:23
(3)
whereDTmax is themaximumtemperaturedifferential atwhich
sap flux is zero (Granier, 1987). In generating sap flux estimates,
we accounted for sensor contactwith poorly conductive xylem;
sapflux is underestimated if a portionof the sensor is in contact
with heartwood (Lu et al., 2004). Although corrections were
made to account for flux underestimation by sensors so posi-
tioned (Clearwater et al., 1998), the exact proportion of a parti-
cular sensor’s length that extends into non-conductive
sapwood cannot be determined without a destructive harvest;
with other ongoing studies at the site, determination through
such harvest could not be made. Inaccurate estimates of inac-
tive sapwood in contactwith sensors can lead to large under-or
over-estimates of sap flux after corrections are implemented.
Thus, data were discarded if corrected fluxes were outside two
standard deviations from the mean of similar sensors (species
and depth) and replacement sensors were installed in new
positions on the same tree. In all, 8 of the 83 sensors were
partially incontactwithheartwood;data fromtwosensorswere
considered unreasonable resulting in sensor replacement.
Table 5 details the sensors that underwent corrections based
on Clearwater et al. (1998) and were considered acceptable.
To account for potential nocturnal fluxes due to both
transpiration and recharge, we selected the highest daily DT to
representDTmax if two conditions are satisfied simultaneously:
(a) the average, minimum 2-h D is <0.05 kPa, thus assuring
that water loss to the atmosphere is negligible, and (b) the
standard deviation of the four highestDT values is<0.5%of the
mean of these values; such stable measurement of maximum
DT ensures that recharge of water above the sensor height is
completed or negligible. In our sap flux time series, zero-flux
nighttime conditions were often not met for several con-
secutive days.
We developed a modified method for scaling tree-level
transpiration that accommodates changes in JS with depth.
Sap flux density in the outer 20 mm did not vary with tree
diameter for any of the species (minimum p > 0.60), which
allowed to combine these data into a time series of mean JSi
(where subscript i represents an individual species) in the
outer xylem. Measured daily JSi values from deeper sensors in
each tree were normalized by themean JSi of all outer sensors.
These normalized values were fit to a Gaussian function,
y ¼ expð0:5ðx a=bÞ2Þ, where y is the normalized flux and x is
the relative depth of the sensor’s center point in the sapwood,
normalized between 0 at the cambium and 1 at the sapwood–
heartwood interface (SigmaPlot 2002, version 8.02, SPSS Inc.).
For species in which the peak of the curve did not occur at the
edge of the sapwood–cambium interface, a maximum rate of
normalized sap flux (i.e. 1) was assumed between the position
of the peak and the cambium. Integrated whole-tree JS was
estimated using Pappus’s second theorem for calculating the
volume of a rotated geometric solid:
Vj ¼ 2pcjAFj (4)
where AFj is the area beneath the fitted curve for an individual
tree, cj is the distance from the center of that tree to thecentroid of the curve, and Vj is a volume that represents the
effective amount of highly conductive sapwood. We can con-
sider this volume with respect to time (cm3 s1) in terms of a
velocity (cm s1)multiplied by an area (cm2)where the velocity
is JS (cm
3 H2O per cm
2 s1, or cm s1) and the area is ASj. Thus,
multiplying Vj by the mean, outer-xylem JSi for that species
yields whole-tree transpiration.
Occasional sensor failure and power outage in a particular
plot produced missing data. Data before and after each gap
was fitted to a power functionwith all functioning sensors and
gapfilled using the best fit against a functioning sensor. The
best fitting sensor was identified based on r2, closeness to
linearity (i.e. exponential parameter closest to 1), and the
distribution of residuals. In all, 40% of growing season data
was gap-filled.
2.5. Stand-level transpiration
Using allometric relationships, AS was estimated for the area
covered by the two sap flux plots and the hectare plot (Table 3).
Large differences were observed among the three estimates of
AS, either for a particular species or in total (Table 4), with
likely effect on stand-level transpiration estimates. To allow
comparison of stand-level component-based estimate of
evapotranspiration with LE from the larger area representing
the eddy covariance footprint, it was necessary to expand the
spatial scale of our sap flux study plots.
We first established a linear relationship between species-
specific LAI data from the 10 litter baskets in the hectare plot
and the total AS within an optimized distance (based on r
2)
from each basket. Combining all species from the Carya and
Quercus genera produced the best fits, AS = 3.287  LAI + 0.166
(r2 = 0.914; p < 0.0001) and AS = 1.428  LAI + 0.363 (r2 = 0.902;
p < 0.0001), respectively, where LAI is in m2 m2 and AS is in
cm2 m2of ground area. The relationship for L. tulipifera was
AS = 5.028  LAI + 0.157 (r2 = 0.758; p = 0.0011). No suitable
relationship was found for L. styraciflua and for the less
abundant and sub-canopy species, so the mean AS (3.43 and
2.89 cm2 m2, respectively) was applied over the entire site.
Species-specific LAI at each of the 29 transect trap locations
was converted to a spatial map for the entire stand using
simple kriging (ArcGIS 9, ESRI, Redlands, CA; Fig. 1). Using the
relationships or averages,weused LAI to estimateAS across the
entire kriged area. Inter-annualmeanmaximum LAI at the site
was 6.3 (0.4) m2 m2. The two sap flux plots and the hectare
plot were positioned in an area with LAI similar to the EC
footprint (6.8 0.3 m2 m2), yet thecontributionof eachspecies
or genera varied among some of these areas. The LE footprint
included areas ranging in LAI by as much as1.7 m2 m2 from
the mean (Fig. 1).
For the two sap flux plots and the hectare plot we summed
whole-tree transpiration to estimate EC. Across the larger
domain, representing the eddy covariance flux footprint, EC
was based on scaling with LAI-based AS estimate:
EC ¼
X
i
ECih 
ASi
ASih
 
(5)
where ECih is ECi for the hectare plot, ASih is sapwood area for
species i in the hectare plot, and ASi is sapwood area for the
Table 6 – Parameters for equations used to estimates
components of evaporation
a y0 r
2
Canopy throughfall
Gauge 1 0.8675 0.5272 0.90
Gauge 2 0.7886 0.4780 0.94
Gauge 3 0.9000 0.6791 0.94
Gauge 4 0.9040 0.4229 0.92
Gauge 5 0.8469 0.4582 0.86
Gauge 6 0.9366 0.4670 0.94
a b r2
Soil evaporation
2002 0.0066 1.4320 0.09
2003–2005 0.0123 1.3003 0.14
Canopy transpiration
Winter (LAI < 1) 0.3114 0.3305 0.24
Canopy interception (IC) was estimated as precipitation (P) minus
measured throughfall (PT). Missing PT data were estimated for each
throughfall gauge using the linear function PT = aP + y0. Soil evapora-
tion (ES) used a power function: ES = aD
b, where D is vapor pressure
deficit. Winter canopy transpiration (EC) was estimated using a
power function: EC ¼ aDbZ where DZ is day-length-normalized vapor
pressure deficit. All regressions are significant at p < 0.0001.entire stand. The hectare plot was used as a basis for scaling
because trees in this plot were amore complete representation
of the species and range of size classes found in the larger eddy
covariance footprint than the trees in the smaller sap-fluxplots.
In scaling, themean sap flux of the threemonitoredQuercus
species was employed for unmonitored Quercus species (Q.
coccinea andQ. prinus, together comprising 3%ofAS). Sap flux of
C. tomentosawas used for unmonitored Carya species (C. glabra
and C. ovata; 15% ofAS). Unmonitored diffuse-porous and ring-
porous genera were estimated to contribute 20% of standAS in
the eddy covariance footprint. This sapwood was partitioned
between the two xylem types based on their proportions in the
hectare plot (Table 4). The average sap flux of Quercus and
Carya was employed to estimate transpiration of the other
ring-porous genera, and that of L. styraciflua and L. tulipifera of
the other diffuse-porous genera. Using the sap flux of either of
the latter species alone affected stand transpiration (EC) during
the growing season by an average of 2.3 (0.08) mm, or less
than 1% of total growing season transpiration, demonstrating
that the EC estimate is reasonably robust to the choice of
representative species.
2.6. Evaporation losses
Latent heat flux (LE) measured with eddy covariance
(expressed in mm H2O) should balance against the compo-
nents of evapotranspiration such that:
LE ¼ IC þ ES þ EC; (6)
where IC is canopy interception and ES is evaporation from the
forest floor and soil surface.
LE was measured using the eddy covariance method
comprising of a triaxial sonic anemometer (CSAT3, Campbell
Scientific, Logan, UT, USA) and an open-path infrared gas
analyzer (IRGA, LI-7500, Li-Cor, Lincoln, NE, USA) positioned
39.8 m above the forest floor. Vertical wind velocity, tempera-
ture, and scalar concentrations of H2O were sampled at 10 Hz
and averaged for half-hour periods. For processing, density
corrections, and analyses of the seasonal and dynamics of
components of the energy balance including its closure, see
Stoy et al. (2006). The path between transducers in the sonic
anemometer or optical length in the open path IRGA may be
blocked during and immediately following rain events, and
correctly identifying these data ‘gaps’ is required to ensure
that long-term sums are correct (Falge et al., 2001).
IC was estimated by subtracting PT from P measured
between collection periods. An exploratory investigation on
the proportion of P reaching the forest floor as stemflow was
conducted over a 2-month period with varying LAI. The
exploratory studywas conducted on six trees representing the
most abundant species and a range of sizes. The rate of
stemflow, normalized by tree circumference, was unrelated to
tree size (p > 0.3), consistent with Granier et al. (2000). When
scaled to the stand, stemflowwas estimated to contribute<1%
of annual precipitation and was excluded from further
consideration. To convert weekly and bi-weekly IC measure-
ment to continuous, half-hourly values, PT accumulated
between measurements was distributed based on a normal-
ized time series of P. For dates of missing PT measurements,estimates for each throughfall gaugeweremade using a linear
regression with P (Table 6). To avoid mischaracterizing
interception associated with multiple, small rain events as a
single, large rain event, data for these regressionswere filtered
to include collection periods in which only one precipitation
event occurred.
ES was not measured directly. The decoupling coefficient
(Jarvis andMcNaughton, 1986) approaches zero inwinter (Stoy
et al., 2006), indicating a strong coupling between surface
conductance and evaporative demand. Thus, ES was esti-
mated using the wintertime (DOY 300–75) relationship
between D and LE from the eddy covariance system. We
excludeddata from the first 3 days after precipitation events to
avoid double-counting IC and discounted the small amounts of
water loss through the bark surface (available from scaled sap
flux measurements) to avoid double-counting EC. We found
significant (p < 0.001) differences between the power function
in 2002 and the subsequent years (Table 6). This differencewas
possibly due to inter-annual variation in surface water
availability, generated by consecutive growing season
droughts in 2001 and 2002. Peak values in ES showed
maximum cross-correlation with peak D values at a 3-h time
lag, which was incorporated into the regression to eliminate a
pattern in the residuals. We tested these estimates of ES by
comparing themwithnighttime LE during the growing season,
using non-gap-filled data and again avoiding periods after
precipitation. We found no significant difference between
estimated ES and measured nighttime LE (p > 0.6).3. Results and discussion
We focus first on methodological aspects of sap flux
measurements, then analyze our procedure for scaling sap
Fig. 2 – Environmental variables measured at hardwood
stand at Duke Forest, 2002–2005. (a) Above-canopy
photosynthetically active radiation (PAR), (b) daily mean
mid-canopy vapor pressure deficit (D) and air temperature
(TA), (c) weekly totals of precipitation (P), and (d) volumetric
soil moisture (u) for the wet and dry sap flux plots.flux measurements to the canopy level and conclude by
evaluating the contribution of individual components to the
closure of stand-level evaporation balance. We note that high
variability in intra- and inter-annual weather (Fig. 2) presents
an opportunity to use our sap flux processing and scaling
methodology over a wide range of environmental conditions.
3.1. Revised methodology for sap flux signal processing
The revised approach for converting sap flux data by selecting
DTmax only during nights with stable DT and D  0 kPa
accounts for both the seasonal shifts of DTmax, due to the
hydration state of the sapwood, and the combined effects of
nocturnal water loss from leaves and recharge of water above
sensor height. With the revised processing in place, sap flux
was frequently observed throughout nighttime hours. A
representative set of diurnal courses (Fig. 3) illustrates a
fivefold increase in nocturnal JS for all species except L.
styraciflua, which increased by 50%. Revised daytime max-
imum JS estimateswere also higher during this sample period,
showing increases of nearly 20% in L. tulipifera andQuercus spp.
and nearly 35% in C. tomentosa. The smaller increase in
nocturnal JS of L. styraciflua during this period did not lead to a
noticeable change in daytime JS. Later we discuss the effect of
the increase in JS of some of the most prevalent species on
stand EC.
Daley and Phillips (2006) used sap flux sensors at two
heights on the stem along with leaf-level gas exchangemeasurements to detect and partition nocturnal fluxes into
recharge and conductance in three deciduous species. In their
study, shade-intolerant, early-successional paper birch (Betula
paprifera) exhibited the highest nocturnal fluxes, which were
almost exclusively due to transpiration. Nocturnal fluxes of
red oak (Quercus rubra) and red maple (Acer rubrum), more
shade-tolerant species, were used almost entirely to re-supply
water to the trunk. In our study, early-successional species
showed lower nighttime JS than late-successional species
(early-successional L. tulipifera and L. styraciflua JS of 13.1  0.2
and 10.1  4.4 g H2O m2 sapwood area per night, respectively,
and late-successional C. tomentosa and Quercus spp. 21.6  5.8
and 21.7  4.7 g H2O m2 sapwood area per night, respec-
tively). Our design did not permit species-specific partitioning
of these nocturnal fluxes into recharge versus transpiration;
however, Dawson et al. (2007) observed nocturnal transpira-
tion across awide range of woody species.We did find that the
large absolute differences in nocturnal JS translated to similar
proportional changes in estimated total growing season ECi,
amounting to 11% in early-successional species and14% in
late-successional species.
Species using nocturnal water uptake to supply transpira-
tion more than recharge show a rapid rise in sap flux with
increasing D (Oren and Pataki, 2001). We found such a trend
based on mean nighttime JS and D after days without rainfall
(data not shown). Following rain events, nocturnal sap flux
exhibitedmuchmore erratic responses toD. And although the
majority of afternoons following rains were characterized by
low D and low JS, themajority of these nights had high JS when
compared to the expected flux based on the sensitivity to D as
observed on dry days. These large nocturnal fluxes following
drought-breaking rains represent recharge of stored water
progressively depleted over entire drying cycles.
During a particular drying cycle, the amount of water
recharging trees at night has been shown to increase with soil
moisture depletion (Phillips et al., 1996), causing recharge to
account for an increasing proportion of daily transpiration
(Oren et al., 1998b). In this study, average nocturnal flux was
significantly higher (p < 0.001) when u < 0.20 m3 m3, a value
shown to limit stomatal conductance and transpiration in this
stand (Pataki and Oren, 2003). Thus, as soil drying intensifies
during a cycle, more water is taken up each night. Our study
does not permit a species-specific evaluation of whether the
increasednocturnal fluxwith soil drying represents increasing
amount of water drawn from storage each day and recharged
each night, or increasing nocturnal water loss from leaves
driven by increasing D with the progression of drying cycles.
However,we show later that, on average for the standandover
the 4-year study, nightlywater uptakewas used to both supply
water lost from leaves and recharge the storage. Considering
that the forest is composed nearly equally of shade-tolerant
and shade-intolerant species, this finding is consistent with
that of Daley and Phillips (2006).
3.2. Scaling sap flux measurements to the eddy
covariance footprint
Sap flux density can be highly variable among individuals of a
given species, necessitating a large number of replicates to
attain an accurate estimate of the mean flux (Oren et al.,
Fig. 3 – Sap flux density by species or genus (JSi) for 3 days during the 2005 growing season with nighttime D > 0.05 kPa.
Black circles show data converted using a method that establishes a baseline value under the assumption that fluxes drop
to zero every night. Open circles show data converted with the revised method in which the DTmax baseline allows for
nighttime flux. Error bars represent 1 S.E.1998b). This is difficult to achieve in species-rich forests,
where increasing replicate numbers can be achieved only by
setting more plots spaced further apart, each requiring power
and a full complement of environmental sensors to capture
the spatial variability in conditions. Kumagai et al. (2005)
recommended monitoring a minimum of six trees to account
for randomvariation.Wewere able to position our dataloggers
such that five individuals of each species in each plot were
monitored. Furthermore we found that neither L. tulipifera nor
L. styraciflua, the two species sampled in both the wet and dry
plots, showed plot-level differences in daily JSi (p > 0.1),
allowing to pool the individuals of each species (thus
producing n = 10). Similarly, the three monitored Quercus
species showed similar daily JSi (p > 0.1) allowing to pool
the individuals of this genus (n = 15). This left only C. tomentosa
(n = 5) with less than the minimum recommended sample
size.
Radial patterns in flux were assessed based on sensors
installed at different depths. Radial sap flux trends for ring-
porous and diffuse-porous species were consistent with some
but not all studies (Phillips et al., 1996;Wullschleger andNorby,
2001).The JSipatterninthesapwoodofdiffuse-porousspecies,L.tulipifera and L. styraciflua, as well as ring-porous C. tomentosa,
wasbestdescribedasGaussian (Fig. 4).Nevertheless, basedona
synthesis of studies on radial patterns in flux (Phillips et al.,
1996), sapwood between the cambium and the peak of the
Gaussian curvewas assumed to transpire at themaximum rate
(represented by the dashed line in Fig. 4). Regressions using
sensors’ relative depth in sapwood, rather than absolute depth
insapwood,hadhigher r2 valuesandshowedsimilarpatterns in
trees of different diameters. None of the three Quercus species
showed a radial pattern in sap flux and were assumed to have
uniformflowthroughout thesapwood, similar toresults fromQ.
alba (Phillipsetal.,1996).Ring-porousspecieswiththinsapwood
areespeciallypronetoerrors in JSestimatesifthesensorsextend
into heartwood (Clearwater et al., 1998; Wullschleger and
Hanson, 2006), can show a sharp decrease over small intervals
within the sapwood (see Phillips et al., 1996), and may support
flow even beyond the visually determined sapwood (Poyatos
et al., 2007). These factors conspire to produce a large degree of
variation among individuals of ring-porous species, necessitat-
ing a higher number of replicates to attain a similar degree of
accuracy than is required for diffuse-porous and non-porous
species (Oren et al., 1998b).
Fig. 4 – Radial profiles of sap flux (JSi) based on relative sapwood depth (beginning in the cambium interface) and normalized
mean sap flux. Data were fitted to a Gaussian equation. The equation for normalized daily sap flux for L. tulipifera was
expð0:5ðx 0:055=0:568Þ2Þ (adjusted r2 = 0.37; p = 0.003), L. styraciflua was expð0:5ðx 0:222=0:343Þ2Þ (adjusted r2 = 0.21;
p = 0.047), and C. tomentosa was expð0:5ðx 0:078=0:396Þ2Þ (adjusted r2 = 0.48; p = 0.029), where x is the relative sapwood
depth. Dashed lines in the top two panels indicate sections where the peak of the curve did not occur at the sapwood–
cambium interface and where a maximum rate of normalized sap flux (i.e. 1.0) was assumed. The horizontal dashed line in
the Quercus spp. panel indicates that no radial pattern was observed in any species and a uniform sap flux was assumed
throughout the sapwood.Further complication in scaling may occur when the
proportion of sap flux measured in inner sensors varies with
environmental conditions and seasons, requiring sufficient
data to quantify the changing patterns. In Pinus taeda this ratio
decreased with soil water availability (Phillips et al., 1996), and
was near unity inwinter, reaching aminimum inmid-growing
season (Scha¨fer et al., 2002). Here, despite large inter-annual
variability in growing season soil moisture, the relationships
between radial depth and fluxwere similar under drought and
non-drought conditions within a year, and did not change
among years (p > 0.1). Thus, transpiration for each tree within
our sap flux plots and the hectare plot was estimated based on
Eq. (4), using the mean species- or genus-specific flux in the
outer xylem, the radial sap flux patterns (Fig. 4), and sapwood
area estimated based on allometric equations (Table 3).
Species-specific values of transpiration (ECi, where i
represents an individual species; Fig. 5a and b) for the hectare
plot, obtained by summing individual tree transpiration, were
normalized by ASi in that plot then multiplied by the EC
footprint ASi (Table 4). Values of ECi were combined to produce
EC (Fig. 5c). Growing season (April–October) EC was very
consistent among years, comprising approximately 65% of
growing season ETS (Table 2), despite large differences in the
amount and timing of precipitation (Fig. 2c). Over the growing
season, Quercus spp. accounted for 38% (2% among the 4
years) of total EC. The rest of EC was contributed by Carya spp.,
L. styraciflua, and L. tulipifera at 19 (2), 16 (1), and 11 (1) %,
respectively. Other species, which included most understoryand some overstory trees, accounted for the remaining 16
(<1) % of EC. The order of contribution was poorly related to
the order of the species or genus ASi (Table 4), reflecting the
differences observed in JSi, as observed in another study in a
similar forest (Wullschleger et al., 2001).
Pataki and Oren (2003) measured sap flux in a different plot
in the same stand in 1997, and found lower growing season EC
(264 mm), similar to an estimate at a nearby, upland broadleaf
stand (278 mm, Oren and Pataki, 2001). Basing their scaling on
the findings of Phillips et al. (1996), these previous studies did
not account for differential radial flow patterns, which would
tend to overestimate EC (Ford et al., 2004). However, their plots
were positioned in areas with lower sapwood area density
than this study, which should somewhat compensate. Indeed,
estimates from the previous studies are very similar to our
estimates before we accounted for nighttime fluxes
(279  11 mm). Thus, we conclude that the previous studies
underestimated EC because they failed to account for the
effects of nocturnal fluxes in data processing. Other estimates
for similar sites show similar annual EC as well as the
proportion of P used as EC (Table 2).
Sap fluxmeasurements continued through thewinter after
the loss of leaves and showed low, but detectable, fluxes that
may be attributed to water loss from the bark surfaces
(Kozlowski, 1943; Oren and Pataki, 2001; Weaver and Mogen-
sen, 1919). The half-hourly fluctuations in DT were often of a
similar magnitude to the diel fluctuations, making it difficult
to identify a reliableDTmax formanywinter days. Therefore, EC
Fig. 5 – Stand-level fluxes over the four study years. (a) and
(b) show sap flux-scaled canopy transpiration by species
(ECi), (c) shows total sap flux-scaled canopy transpiration
(EC) and eddy covariance-measured latent heat flux (LE),
(d) estimates for the remaining components of stand
evapotranspiration (soil evaporation, ES) plus canopy
interception (IC, plotted as a 3-day moving average)
compared to the difference LE S EC, (e) leaf area index
(LAI, m2 mS2).was modeled for winter months by combining reliable data
from all trees using a power function with day-length-
normalized vapor pressure deficit (DZ; Table 6). This function
was used to estimate EC during winter months (December–
March, classified by LAI < 1) for the entire stand, which
averaged 20 (0.8) mm year1, about 6% of annual EC. Mean
winter ECwas 0.17 (0.01)mm d1, higher than that previously
reported in this area for Acer rubrum and Q. alba (0.07 mm d1,
Pataki and Oren, 2003), most like due to accounting for
nocturnal sap flux. Wintertime EC was allocated among
species based on their proportion of growing season EC.
The large footprint of LE measurements above tall forests
can present challenges for scaling sap flux as a component of
ETS at a comparable scale. As a reminder, we scaled sap flux to
three areas of the stand based on AS obtained from the
inventory in (1) the two sap flux plots, (2) the hectare plot
around the tower, and (3) from LAI in the approximately
250 m  250 m area representing most of the eddy covariancefootprint (Fig. 1). Estimated annual EC for the eddy covariance
footprint was 338 (7) mm (Table 2), 8% lower than for the two
sap flux plots due to an overrepresentation of Quercus in the
wet sap flux plot (Table 4), and 17% higher than the hectare
plot due to an under-representation of this genus in the area
immediately surrounding the tower. Thus, despite the
similarity of average LAI among these sample areas, non-
uniform species distribution in the forest (affecting the scaling
ASi) combined with species differences in sap flux produced
substantially different estimates of EC. A proper comparison of
ETS with LE requires scaling sap flux to EC that accounts for
species or functional groups, rather than only the bulk canopy
properties in the sap flux and eddy covariance footprints.
3.3. Evaporative losses
As expected, sap flux-scaled daily EC and LE followed similar
trends (Fig. 5d). However, because annual EC comprised only
54 (3) % of LE, other components of total evapotranspiration
required accurate quantification to make the conclusions
regarding correction for nocturnal flux and scaling mean-
ingful.
Annual estimates of IC are presented in Table 2. Our mean
growing season estimate of 96 (49) mm was approximately
14% of growing season P, similar to the 14% reported in
previous studies in this area (Pataki and Oren, 2003). While
thesemean values for the site agree well with previous results
(Table 2), the standard deviation of annual IC (based on
variation in throughfall among gauges) was nearly 60 mm, or
about 9% of ETS. This variability represents the spatial
heterogeneity of interception, translating to uncertainty in
estimated ETS.
The estimates of ES (Table 2) were constructed based on a
subset of the wintertime eddy covariance-measured LE. These
half-hourly estimates based on the wintertime relationships
were close to nighttime LE values during the growing season,
suggesting that the relationship was useful for D values
outside the range used in its derivation. We note that the
estimate of ES is thus not entirely independent of LE with
which we ultimately compare the component-sum evapo-
transpiration (ETS = EC + ES + IC). Mean annual ES estimate was
103 (14) mm. For a different estimate of ES in this stand (Stoy
et al., 2006), combined modeled growing season ES based on
radiation penetration through the canopy with wintertime
measured LE, arriving at an annual value of 176  7 mm. Our
estimates of ES are more similar to LE measured with an eddy
covariance system at 2 m above the forest floor in another
southeastern deciduous forest (88 mm) where LE above the
forest was similar to ours (Table 2; Wilson et al., 2001). Few
other studies of broadleaf forests in this region have
incorporated estimates of forest floor evaporation and this
component of ETS remains the source of some uncertainty.
We assessed the agreement between estimates of various
components of ET in time scales ranging from daily to inter-
annually. Two methods for estimating forest evaporation (i.e.
excluding transpiration), LE  EC and ES + IC, are compared in
Fig. 5d. At the daily time scale, ES + ICwas typically higher than
LE  EC during and immediately after rain events, but was
frequently lower during periods of high radiation loads. On a
monthly basis, ETS showed good agreement with LE but,
Fig. 6 – Comparison of monthly ecosystem evapotranspiration from eddy covariance-measured latent heat flux (LE) and
from sap flux-scaled hydrologic budget (ETS). ETS includes EC estimates that ignore nighttime sap flux (a) and those that
account for nighttime sap flux (b).consistent with the daily comparisons above, monthly ETS
was somewhat higher during periods of low to intermediate
radiation loads and lower during periods of very high radiation
loads (Fig. 6). Routines used to gapfill eddy covariance-
measured LE (Falge et al., 2001; Stoy et al., 2006) may not
completely account for potentially high evaporation rates
from wet canopy and forest floor following rain events,
because relationships derived from data obtained when
surfaces are dry would underestimate evaporation following
rain events when surface conductance is high. This effect is
magnified during periods with low radiation loads, because
sensors remain wet for longer periods producing higher
proportion of unacceptable eddy covariance data. This is
reflected in a significant linear increase in the number of
gapfilled data points with decreasing monthly net radiation
(linear regression: p = 0.0015). Although underestimate of ES
following rainswill similarly bias evaporation estimates based
on both methods, only LE-based evaporation estimate
includes underestimated IC as well. The component-based IC
estimate uses throughfall measurements that, although are
spatially very variable, are largely immune to technical
problems that cause a bias under particular conditions.
Periods of high radiation loads are restricted to themonths
in which solar zenith angle is low. During these periods, but
excluding times in which the canopy is wet, ETS is often lower
than LE (Bovard et al., 2005; Oren et al., 1998b; Scha¨fer et al.,
2002; Wilson et al., 2001). This may be the result of under-
estimating stand-level EC. EC may be underestimated for two
reasons: (1) the signalmay be saturating under high flux rates,
as has been commonly observed (Bovard et al., 2005; Hogg
et al., 1997; Wilson et al., 2001) and (2) the contribution of the
sub-canopy to EC may be higher during periods in which
radiation penetrates deeper in the canopy. The contribution of
small understory individuals (<40 mm in diameter) and
herbaceous vegetation was not estimated in this study, but
can be large (Gholz and Clark, 2002; Vincke et al., 2005). In
support of (2), Granier et al. (2000) found a linear relationshipbetween EC and LE (i.e., no sign of saturation) in a study in
which equal attentionwas given tomonitoring large and small
individuals. The importance of the sub-canopy to stand
transpiration has been shown in many studies. Transpiration
rates of canopy and sub-canopy trees compensate as stands
develop, leading to a conservative forest transpiration (Phillips
and Oren, 2001) as has been shown spatially among stands of
different degrees of canopy closure (Roberts, 1983). Thus, we
conclude that underestimation of stand-level EC is often the
result of inadequate representation of the sub-canopy
components in scaling, rather than instrument deficiencies.
Nocturnal sap flux scaled to nocturnal EC (occurring as
recharge or water loss from leaves when PAR = 0) averaged
0.19 (0.11) mm d1 over the growing season. The ratio of
night/day EC, 0.17  0.19, is within the range of 0–0.25 for
deciduous trees (Dawson et al., 2007). Assuming for simplicity
that nocturnal EC is used entirely for recharge, the average
nocturnal recharge rate, or even the highest rate of
0.6 mm d1, fall well within the 1.0 mm d1 estimated
based on a relationship between recharge and sapwood area
(Goldstein et al., 1998). Nocturnal LE, which includes evapora-
tion in addition to transpiration, was less than half of
nocturnal water uptake (0.08  0.11 mm d1). Thus, the results
suggest that at least half of this nighttime flux is used to re-
supply the trunk with water used earlier in the day, while
some portion of the remainder may be lost as nocturnal
transpiration.
At annual time scale, estimates of ETS and LE showed good
agreement (see Table 2, Fig. 7). ETS was lower than LE in each
year before accounting for nighttime sap flux in estimates of
EC, averaging 6 (3) %, reversing to +5 (3) % after sap flux
data were processed based on the new approach. Accounting
for nocturnal flux had a more striking effect during the
growing season—with the difference decreasing from 16
(2)% to 4 (3)%. Thus, although accounting for the effect of
nocturnal fluxes did not resolve the discrepancy between ETS
and LE in each day and each month – possibly due to
Fig. 7 – Annual ET budgets for 2002–2005 including sap
flux-based EC and the remaining evaporation components,
and eddy covariance based estimates (LE). Lack of closure
represents the relative difference between the two
methods (using LE as the base).underestimation of EC from small trees and other understory
species during high radiation periods – the approach resulted
in a substantial 18 (2) % increase in the estimate of annual EC
(61  9 mm year1), leading to both annual and seasonal
similarity of ETS and LE. The increase in estimated growing
season EC of 22 (4)% based on the newmethod for accounting
for nocturnal fluxes was intermediate compared to increases
produced by other methods: a 12% increase in a Populus
trichocarpa  P. deltoides plantation (Kim et al., 2008) and 30%
increase in boreal Picea abies (L.) Karst. stands (Ward et al.,
2008).
3.4. Implications
The analysis showed that mischaracterization of the footprint
area is not the likely source of the reported consistent lower
estimates of ETS than LE. Scaling to EC based on the hectare
plot, ETS would have been only 3 (3) % lower than LE, while
scaling based on the two sap flux plots, would have resulted in
ETS that was 10 (3) % higher than LE. Thus, when scaled
properly, and after accounting for the major contributing
fluxes, seasonal and annual estimates of evapotranspiration
that include sap flux-scaled EC were in good agreement with
those based on eddy covariance.
Accounting for nocturnal sap flux in trees caused by the
recharge of water to upper trunks and branches, as well as
nocturnal water loss, is a vital step for accurately estimating
EC. Failure to account for nocturnal fluxes is the most likely
explanation for the previously observed bias towards lowerestimates of component-based evapotranspiration. Never-
theless, comparisons of daily and monthly estimates of
evapotranspiration indicate that the eddy covariance
approach tends to underestimate during low radiation-high
surface wetness periods, while the component sum tends to
underestimate during periods of high radiation. Thus, the
agreement between the methods at coarser temporal resolu-
tion is somewhat achieved by compensating errors, making it
case specific.
Acknowledgements
This Researchwas supported by theOffice of Science (BER) U.S.
Department of Energy, Grant No. DE-FG02_00ER63015. We
would like to thank H. McCarthy, H.-S. Kim, B. Poulter, K.
Scha¨fer, E. Ward, M. Siqueira, J.-Y. Juang, K. Novick and G.
Katul for assistance with data collection and analysis and D.
Richter for rooting depth data.r e f e r e n c e s
Arneth, A., Kelliher, F.M., Bauer, G., Hollinger, D.Y., Byers, J.N.,
Hunt, J.E., McSeveny, T.M., Ziegler, W., Vygodskaya, N.N.,
Milukova, I., Sogachov, A., Varlagin, A., Schulze, E.D., 1996.
Environmental regulation of xylem sap flow and total
conductance of Larix gmelinii trees in eastern Siberia. Tree
Physiology 16 (1–2), 247–255.
Bovard, B.D., Curtis, P.S., Vogel, C.S., Su, H.-B., Schmid, H.P.,
2005. Environmental controls on sap flow in a northern
hardwood forest. Tree Physiology 25, 31–38.
Clearwater, M.J., Meinzer, F.C., Andrade, J.L., Goldstein, G.,
Holbrook, N.M., 1998. Potential errors in measurement of
nonuniform sap flow using heat dissipation probes. Tree
Physiology 19, 681–687.
Daley, M.J., Phillips, N.G., 2006. Interspecific variation in
nighttime transpiration and stomatal conductance in a
mixed New England deciduous forest. Tree Physiology 26,
411–419.
Dawson, T.E., Burgess, S.S.O., Tu, K.P., Oliveira, R.S., Santiago,
L.S., Fisher, J.B., Simonin, K.A., Ambrose, A.R., 2007.
Nighttime transpiration in woody plants from contrasting
ecosystems. Tree Physiology 27 (4), 561–575.
Falge, E., Baldocchi, D., Olson, R., Anthoni, P., Aubinet, M.,
Bernhofer, C., Burba, G., Ceulemans, G., Clement, R.,
Dolman, H., Granier, A., Gross, P., Grunwald, T., Hollinger,
D., Jensen, N.O., Katul, G., Keronen, P., Kowalski, A., Lai,
C.T., Law, B.E., Meyers, T., Moncrieff, J., Moors, E., Munger,
J.W., Pilegaard, K., Rannik, U., Rebmann, C., Suyker, A.,
Tenhunen, J., Tu, K., Verma, S., Vesala, T., Wilson, K.,
Wofsy, S., 2001. Gap filling strategies for long term energy
flux data sets. Agricultural and Forest Meteorology 107 (1),
71–77.
Ford, C.R., Goranson, C.E., Mitchell, R.J., Will, R.E., Teskey, R.O.,
2004. Dirunal and seasonal variability in the radial
distribution of sap flow: predicting total stem flow in Pinus
taeda trees. Tree Physiology 24, 951–960.
Ford, C.R., Hubbard, R.M., Kloeppel, B.D., Vose, J.M., 2007. A
comparison of sap flux-based evapotranspiration estimates
with catchment-scale water balance. Agricultural and
Forest Meteorology 145 (3–4), 176–185.
Geron, C.D., Nie, D., Arnts, R.R., Sharkey, T.D., Singsaas, E.L.,
Vanderveer, P.J., Guenther, A., Sickles II, J.E., Kleindienst,
T.E., 1997. Biogenic isoprene emission: model evaluation in
a southeastern United States bottomland deciduous forest.
Journal of Geophysical Research 102 (D15), 18889–18901.
Gholz, H.L., Clark, K.L., 2002. Energy exchange across a
chronosequence of slash pine forests in Florida.
Agricultural and Forest Meteorology 112 (2), 87–102.
Goldstein, G., Andrade, J.L., Meinzer, F.C., Holbrook, N.M.,
Cavelier, J., Jackson, P., Celis, A., 1998. Stem water storage
and diurnal patterns of water use in tropical forest canopy
trees. Plant Cell and Environment 21 (4), 397–406.
Granier, A., 1987. Sap flow measurements in Douglas fir tree
trunks by means of a new thermal method. Annales des
Sciences Forestie`res 44, 1–14.
Granier, A., Biron, P., Lemoine, D., 2000. Water balance,
transpiration and canopy conductance in two beech stands.
Agricultural and Forest Meteorology 100 (4), 291–308.
Hogg, E.H., Black, T.A., den Hartog, G., Neumann, H.H.,
Zimmermann, R., Hurdle, P.A., Blanken, P.D., Nesic, Z.,
Yang, P.C., Staebler, R.M., McDonald, K.C., Oren, R., 1997. A
comparison of sap flow and eddy fluxes of water vapor from
a boreal deciduous forest. Journal of Geophysical Research
102, 28929–28937.
Hsieh, C.-I., Katul, G.G., Chi, T.-W., 2000. An approximate
analytical model for footprint estimation of scalar fluxes in
thermally stratified atmospheric flows. Advances in Water
Resources 23, 765–772.
Jarvis, P.G., McNaughton, K.G., 1986. Stomatal control of
transpiration—scaling up from leaf to region. Advances in
Ecological Research 15, 1–49.
Kim, H.-S., Oren, R., Hinckley, T.M., 2008. Actual and potential
transpiration and carbon assimilation in an irrigated poplar
plantation. Tree Physiology 28, 559–577.
Ko¨stner, B.M.M., Schulze, E.-D., Kelliher, F.M., Hollinger, D.Y.,
Byers, J.N., Hunt, J.E., McSeveny, T.M., Meserth, R., Weir,
P.L., 1992. Transpiration and canopy conductance in a
pristine broad-leaved forest of Nothofagus: an analysis of
xylem sap flow and eddy correlation measurements.
Oecologia 91, 350–359.
Kozlowski, T.T., 1943. Transpiration rates of some forest tree
species during the dormant season. Plant Physiology 18 (2),
252–260.
Kumagai, T., Aoki, S., Nagasawa, H., Mabuchi, T., Kubota, K.,
Inoue, S., Utsumi, Y., Otsuki, K., 2005. Effects of tree-to-tree
and radial variations on sap flow estimates of transpiration
in Japanese cedar. Agricultural and Forest Meteorology 135
(1–4), 110–116.
Lu, P., Urban, L., Zhao, P., 2004. Granier’s thermal dissipation
probe (TDP) method for measuring sap flow in trees: theory
and practice. Acta Botanica Sinica 46, 631–646.
Meinzer, F.C., Clearwater, M.J., Goldstein, G., 2001. Water
transport in trees: current perspectives, new insights and
some controversies. Environmental and Experimental
Botany 45, 239–262.
Oren, R., Pataki, D.E., 2001. Transpiration in response to
variation in microclimate and soil moisture in southeastern
deciduous forests. Oecologia 127, 549–559.
Oren, R., Ewers, B.E., Todd, P., Phillips, N., Katul, G.G., 1998a.
Water balance delineates the soil layer in which moisture
affects canopy conductance. Ecological Applications 8 (4),
990–1002.
Oren, R., Phillips, N., Katul, G.G., Ewers, B.E., Pataki, D.E., 1998b.
Scaling xylem sap flux and soil water balance and
calculating variance: a method for partitioning water flux in
forests. Annales des Sciences Forestie`res 55, 191–216.
Oren, R., Phillips, N., Ewers, B.E., Pataki, D.E., Megonigal, J.P.,
1999. Sap-flux-scaled transpiration responses to light, vaporpressure deficit, and leaf area reduction in a flooded
Taxodium distichum forest. Tree Physiology 19, 337–347.
Pataki, D.E., Oren, R., 2003. Species differences in stomatal
control of water loss at the canopy scale in a mature
bottomland deciduous forest. Advances in Water Resources
26, 1267–1278.
Phillips, N., Oren, R., 2001. Intra- and inter-annual variation in
transpiration of a pine forest. Ecological Applications 11 (2),
385–396.
Phillips, N., Oren, R., Zimmermann, R., 1996. Radial patterns of
xylem sap flow in non-, diffuse- and ring-porous tree
species. Plant, Cell and Environment 19, 983–990.
Poyatos, R., Cermak, J., Llorens, P., 2007. Variation in the radial
patterns of sap flux density in pubescent oak (Quercus
pubescens) and its implications for tree and stand
transpirationmeasurements. Tree Physiology 27 (4), 537–548.
Roberts, J., 1983. Forest transpiration—a conservative
hydrological process. Journal of Hydrology 66 (1–4), 133–141.
Scha¨fer, K.V.R., Oren, R., Lai, C.-T., Katul, G.G., 2002. Hydrologic
balance in an intact temperate forest ecosystem under
ambient and elevated atmospheric CO2 concentration.
Global Change Biology 8, 895–911.
Stoy, P.C., Katul, G.G., Siqueira, M.B.S., Juang, J.-Y., Novick, K.A.,
McCarthy, H.R., Oishi, A.C., Uebelherr, J.M., Kim, H.-S., Oren,
R., 2006. Separating the effects of climate and vegetation on
evapotranspiration along a successional chronosequence in
the southeastern US. Global Change Biology 12.
Vincke, C., Granier, A., Breda, N., Devillez, F., 2005.
Evapotranspiration of a declining Quercus robur (L.) stand
from 1999 to 2001. II. Daily actual evapotranspiration and
soil water reserve. Annals of Forest Science 62, 615–623.
Ward, E.J., Oren, R., Sigurdsson, B.D., Jarvis, P.G., Linder, S., 2008.
Fertilization effects on mean stomatal conductance are
mediated through changes in the hydraulic attributes of
mature Norway spruce trees. Tree Physiology 28, 579–596.
Weaver, J.E., Mogensen, A., 1919. Relative transpiration of
coniferous and broad-leaved trees in autumn and winter.
The Botanical Gazette 68 (6), 393–424.
Williams, D.G., Cable, W., Hultine, K., Hoedjes, J.C.B., Yepez,
E.A., Simonneaux, V., Er-Raki, S., Boulet, G., de Bruin, H.A.R.,
Chehbouni, A., Hartogensis, O.K., Timouk, F., 2004.
Evapotranspiration components determined by stable
isotope, sap flow and eddy covariance techniques.
Agricultural and Forest Meteorology 125 (3–4), 241–258.
Wilson, K.B., Hanson, P.J., Baldocchi, D.D., 2000. Factors
controlling evaporation and energy partitioning beneath a
deciduous forest over an annual cycle. Agricultural and
Forest Meteorology 102, 83–103.
Wilson, K.B., Hanson, P.J., Mulholland, P.J., Baldocchi, D.D.,
Wullschleger, S.D., 2001. A comparison of methods for
determining forest evapotranspiration and its components:
sap-flow, soil water budget, eddy covariance and catchment
water balance. Agricultural and Forest Meteorology 106,
153–168.
Wullschleger, S.D., Hanson, P.J., 2006. Sensitivity of canopy
transpiration to altered precipitation in an upland oak
forest: evidence from a long-term field manipulation study.
Global Change Biology 12, 97–109.
Wullschleger, S.D., Norby, R.J., 2001. Sap velocity and canopy
transpiration in a sweetgum stand exposed to free-air CO2
enrichment (FACE). New Phytologist 150, 489–498.
Wullschleger, S.D., Hanson, P.J., Todd, D.E., 2001. Transpiration
from a multi-species deciduous forest as estimated by
xylem sap flow techniques. Forest Ecology and
Management 143 (205–213).