An evaluation of methods for partitioning 
eddy covariance-measured net ecosystem 
exchange into photosynthesis and 
respiration
Authors: Paul C. Stoy, Gabriel G. Katul, Mario B.S. 
Siqueira, Jehn-Yih Juang, Kimberly A. Novick, 
Joshua M. Uebelherr, and Ram Oren 
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# 
141, ISSUE# 1, (December 2006), DOI# 10.1016/j.agrformet.2006.09.001.
Stoy PC, Katul GG, Siqueira MBS, Juang J-Y, Novick KA, Oren R (2006) An evaluation of 
methods for partitioning eddy covariance-measured net ecosystem exchange into photosynthesis 
and respiration. Agricultural and Forest Meteorology 141: 2-18. DOI: 10.1016/
j.agrformet.2006.09.001.
Made available through Montana State University’s ScholarWorks 
scholarworks.montana.edu 
An evaluation of models for partitioning eddy covariance-measured
net ecosystem exchange into photosynthesis and respiration
Paul C. Stoy a,b,*, Gabriel G. Katul a,b, Mario B.S. Siqueira a,c, Jehn-Yih Juang a,
Kimberly A. Novick a, Joshua M. Uebelherr a, Ram Oren a
a Nicholas School of the Environment and Earth Sciences, Box 90328, Duke University, Durham, NC 27708-0328, United States
b Department of Civil and Environmental Engineering, Pratt School of Engineering, Duke University, Durham, NC, USA
c Departamento de Engenharia Mecaˆnica, Universidade de Brası´lia, Brazil
Received 5 May 2006; accepted 8 September 2006AbstractWe measured net ecosystem CO2 exchange (NEE) using the eddy covariance (EC) technique for 4 years at adjoining old field
(OF), planted pine (PP) and hardwood forest (HW) ecosystems in the Duke Forest, NC. To compute annual sums of NEE and its
components – gross ecosystem productivity (GEP) and ecosystem respiration (RE) – different ‘flux partitioning’ models (FPMs)
were tested and the resulting C flux estimates were compared against published estimates from C budgeting approaches, inverse
models, physiology-based forward models, chamber respiration measurements, and constraints on assimilation based on sapflux
and evapotranspiration measurements. Our analyses demonstrate that the more complex FPMs, particularly the ‘non-rectangular
hyperbolic method’, consistently produced the most reasonable C flux estimates. Of the FPMs that use nighttime data to estimate
RE, one that parameterized an exponential model over short time periods generated predictions that were closer to expected flux
values. To explore howmuch ‘new information’ was injected into the data by the FPMs, we used formal information theory methods
and computed the Shannon entropy for: (1) the probability density, to assess alterations to the flux measurement distributions, and
(2) the wavelet energy spectra, to assess alterations to the internal autocorrelation within the NEE time series. Based on this joint
analysis, gap-filling had little impact on the IC of daytime data, but gap-filling significantly altered nighttime data in both the
probability and wavelet spectral domains.
# 2006 Elsevier B.V. All rights reserved.
Keywords: Eddy covariance; Information theory; Net ecosystem exchange; Gross ecosystem productivity; Ecosystem respiration; Grass field;
Oak-hickory forest; Pinus taeda1. Introduction
A primary goal of the United States Global Change
Research Program (USGCRP) is to quantify the role of
vegetation in controlling the magnitude and variabilityof the terrestrial carbon sink (Sarmiento and Wofsy,
1999). Eddy covariance (EC) estimates of the net
ecosystem exchange of CO2 (NEE) are ideally suited
for addressing this goal (Baldocchi et al., 2001) with an
important caveat. To evaluate the processes that control
NEE, EC data rely on models that partition NEE into its
components—gross ecosystem productivity (GEP) and
ecosystem respiration (RE). These ‘flux partitioning’
models (FPMs) can be used in turn to ‘gap-fill’ missing
or rejected data, thereby allowing estimates of CO2 flux
over long periods (Falge et al., 2001). Hence, FPMs are
an essential tool for interpreting EC data, yet no study to
date has critically examined how the choice of a given
FPM affects the magnitude and variability of carbon (C)
flux estimates by comparison with independent esti-
mates of C flux for different ecosystems that experience
similar climatic and edaphic conditions.
Here, we used the EC technique tomeasure NEE for 5
years at adjoining old field (OF), planted pine (PP) and
broadleaf-deciduous (‘hardwood’, HW) ecosystems that
represent a typical post-agricultural successional
sequence in the southeastern United States (SE, Oosting,
1942). The study ecosystems offer a unique set-up for a
comparison of FPMs because, although they experience
the same climatic (e.g., geostrophic winds, cloud cover,
precipitation history, air temperature) and edaphic (e.g.,
soil type, rooting depth) conditions, they differ sub-
stantially in frictionvelocity (u*, Fig. 1; Stoy et al., 2006),
which is often used as the primary filter for nighttime EC
measurements (Falge et al., 2001; Gu et al., 2005;
Reichstein et al., 2005). These differences in u* are
attributable to differences in roughness properties that
reflect differences in foliage drag and leaf area density
among the ecosystems (Poggi et al., 2004). Five-year EC
time series from these three ecosystems were used to
generate NEE, GEP, and RE estimates using four FPMs
that differ substantially in the data source employed (i.e.,
daytime versus nighttime data) and complexity of the
modeling procedure. These estimates were compared
against the best available independent data and model
results from a range of other studies that provide
estimates of the magnitude and interannual variability of
net primary productivity (NPP), RE, and gross primary
productivity (GPP). In this way, we ensure that EC-based
estimates are defensible and independently constrained
to the extent that ecosystem flux inter-comparison wasFig. 1. The probability distribution of measured friction velocity (u*) for OF
nighttime and daytime distributions, respectively. u* = 0.2 m s
1, a commo
reference.possible. We then diagnosed the FPMs using both direct
comparison and information theory via the Shannon
entropy to assess which FPM generated the most
reasonable flux estimates across dissimilar vegetative
types. After choosing the optimal FPM, we provide error
estimates on the flux values that are obtained.
With respect to the inter-comparison, it is important to
note that the independent and model-based estimates
themselves are naturally prone to random and systematic
errors and do not represent the unknown ‘true’ flux.
Regardless, alongside EC-based estimates, these techni-
ques represent the state of the art for estimating C flux.
Therefore, estimates generated using these methods
provide ‘expected’ flux values in the logical rather than
the statistical sense. If a given FPM fails to reproduce the
approximate mean and long-term variability of the
independent ormodel-based estimates, or delivers results
that conflict with logical expectations (e.g., if the fast-
growingPPecosystemis foundtobea long-termsourceof
C to the atmosphere), we can conclude that this FPM
should not be used for generating EC-based estimates
across different vegetation types. Below we briefly des-
cribe the methods that have been employed by different
studies to estimate C flux from the study ecosystems.
At OF, Novick et al. (2004) combined EC and leaf-
level physiological measurements with a ‘big-leaf’
assimilation model to estimate NEE, GEP, and RE for
2001 (Table 1), and used the model to predict C fluxes in
the absence of soil moisture (u) limitations. Combined
with other studies, their findings suggest that the
interannual variability of NEE in grassland ecosystems
may be driven by changes inGEP rather thanRE, and that
grassland ecosystems can switch fromC source to C sink
with increasing u content. The interannual variability of
GEPatOF is expected to be large due to reductions in leaf(A), PP (B) and HW (C). The symbols o and x indicate median u* for
n threshold for data filtering, is indicated by vertical dashed lines for
Table 1
Estimates of net ecosystem exchange (NEE), ecosystem respiration (RE) and gross primary productivity (GPP) at the old field (OF) and planted pine
(PP) ecosystems in the Duke Forest, NC
Ecosystem Period NEE GEP RE References
OF 11 April 2001–11 April 2002 97 1202 1299 Novick et al. (2004)
Ecosystem Year NEE GPP RE References
PP 1997 1224a Luo et al. (2001)a, Katul et al. (2001b)b
1998 580 to 650b, 585c, 428d 1250a, 2371f 1920f Falge et al. (2001)c
1999 666c, 605d, 661e, 746h 1808d, 2287e 1203d Lai et al. (2002)d
2000 792e 2486e Scha¨fer et al. (2003)e
2001 355g 2122g 1767g Hamilton et al. (2002)f
2002 410g 2033g 1623g Juang et al. (2006)g
2003 449g 2471g 2022g Hui et al. (2004)h
All flux units are g C m2 year1. Gross ecosystem productivity (GEP) differs from GPP as described in the text.area index (LAI) induced by drought and anthropogenic
management, namely hay removal, and corresponding
changes in canopy conductance. Stoy et al. (2006)
showed that EC-measured evapotranspiration (ET)
increased by nearly 50% between severe drought and
wet years, and used a simple radiation attenuation model
after Campbell and Norman (1998) to model transpira-
tion (T), which varied by over 200% among years. Given
that GEP is controlled in part by T due to the role of the
stomata in coupling canopy H2O losses with CO2 gains,
we expect that GEPwill be highly sensitive to hydrologic
forcing at OF, and thus highly variable given the wide
range of climatic conditions observed over the measure-
ment period.
Estimates of GPP and/or RE at PP are available from
ecosystem carbon budgeting (Hamilton et al., 2002),
canopy conductance-based assimilation models (Scha¨fer
et al., 2003), physiology-based assimilation models (Luo
et al., 2001), and inverse methods using mean CO2 and
mean air temperature (Ta) profiles within the canopy
(Table 1, Lai et al., 2002; Juang et al., 2006). Earlier
studies used EC data to estimate NEE and/or its
components at PP (Table 1, Falge et al., 2001; Katul
et al., 2001b; Law et al., 2002), but did not conduct a
sensitivity analysis of different FPMs as performed here.
We expect that the magnitude and variability of C fluxes
at PP should approximate the independent estimates,
noting that at PP the sensitivity of ET and modeled T to
severe drought and ice storm damage (McCarthy et al.,
2006; Stoy et al., 2006) may inject more variability in
GEP during 2001–2005 than in previous years.
There are no independent estimates of NEE at HWat
the present, but we expect that interannual C flux
variability will be small in a drought-tolerant (Pataki
et al., 1998; Pataki and Oren, 2003) closed canopy
forest that sustained minor ice storm damage and
showed little interannual variability in measured ET ormodeled T from 2001 to 2005 (Stoy et al., 2006). The
difference in chamber measured soil respiration at HW
between mild and severe drought years was ca.
100 g C m2 year1 (6%), and the magnitude of soil
respiration at HW was between 10 and 26% larger than
at PP (Palmroth et al., 2005). Ecosystem respiration is
dominated by forest floor efflux at both ecosystems, but
Mortazavi et al. (2005) used dC13 measurements to
show that soil respiration provides a greater proportion
of RE at HW compared to PP. Thus, to a first order, we
expect RE at HW to be similar to PP given the
similarities in forest floor efflux and the larger
contribution of foliage respiration to ecosystem
respiration at PP. A complication arises due to an
11.9 ha clear-cut that occurred 200 m south of the HW
EC tower on private land in November 2002 (see Fig. 1
in Stoy et al., 2006). We conducted a two-dimensional
footprint analysis to determine the impact of this clear-
cut on annual flux estimates at HW (Appendix A),
which was estimated to be on the order of ca.
200 g C m2 year1 for NEE and RE but less than
50 g C m2 year1 for GEP during 2003 and negligible
during other years.
For the purposes of our study, it is important to
distinguish between GEP (also called gross ecosystem
exchange, GEE) and gross primary productivity (GPP).
GEP does not take into account recycling of CO2 within
the ecosystem, such as within the leaf via re-assimilation
of metabolic (‘dark’) respiration, or within the canopy
volume and below the plane of the EC instrumentation.
GEP is thus analogous, but not identical, to GPP
(Goulden et al., 1997). Note also that all EC-based
methods that use nighttime RE estimates to extrapolate
daytime RE assume that the ratio of photorespiration –
energy expended by the oxygenase function of Rubisco –
to RE is small. The relative role of photorespiration is
likely to be lower in OF which contains a small
proportion of C4 plants (Novick et al., 2004). Our goal
was not to quantify differences betweenGEP andGPP or
to attempt to quantify the photorespiratory term in EC-
measured RE. Rather, our criteria for success in
employing an ideal FPM is its ability to estimate GEP,
RE, andNEEvalues that are consistentwith themean and
interannual variability determined by C flux estimates
that are independent of eddy covariance, or based on
models that assimilate multiple data sources.
We restrict our analysis of FPMs to those that have a
basis in ecosystem physiology, namely those that
parameterize known relationships between RE and
temperature or between GEP (or NEE) and light. These
methodologies employ the basic equations used bymany
ecosystem models, and thus can generate transferable
information (e.g., terms of their parameters) for future
studies of C dynamics both across time at the ecosystems
studied here, and across spacewith respect to ecosystems
with similar characteristics. To this end, we do not
explore statistical methods such as multiple imputation
(Hui et al., 2004) or more complex algorithms such as
neural networks (Hagen et al., 2006), whose ‘black-box’
nature can complicate the analysis of relationships
between driving variables and flux. Commonly used
‘look-up tables’ do not provide a formal mechanism for
generating error estimates – as explored for the case of PP
by Oren et al. (2006) – and are likewise not investigated
here. After choosing the ideal FPM, we provide error
estimates for all flux terms for each ecosystem-year of
measurement.
2. Methods
2.1. Site description
Ecological characteristics of the study sites have
been described extensively (Ellsworth et al., 1995; Oren
et al., 1998; Lai and Katul, 2000; Pataki and Oren, 2003;Table 2
Summary of ancillary measurements and sensor locations
Variable Instrument Company
Ta/RH/D HMP 35C/45C Campbell Scientific,
PAR LI-190SA Li-Cor, Lincoln, NE
Rn Q7
a, CNR1 REBS, Seattle, WA;
u/u* CSAT3 triaxial sonic anemometer Campbell Scientific,
LAI LAI-2000, litter baskets Li-Cor, Lincoln, NE
NEE/LE LI 7500b Li-Cor, Lincoln, NE
Negative height values denote distance belowground.
a Q7 radiometers were employed before 1 January 2004 and the CNR1 r
b Coupled to CSAT3. Corrections for the effects of air density on flux mNovick et al., 2004). Briefly, OF is dominated by the
grass Festuca arundinacea Shreb., with contributions
from other C3 and C4 grasses and various herbs, and is
mowed at least once annually for forage (Novick et al.,
2004). PP is a 22-year-old (in 2005) planted Pinus taeda
L. forest (Ellsworth et al., 1995; Oren et al., 1998; Katul
et al., 1999). HW is an 80–100-year-old mixed
deciduous forest dominated by late-successional
shade-tolerant Quercus (oak) and Carya (hickory)
species with fewer shade-intolerant individuals of
Liquidambar styraciflua L. and Liriodendron tulipifera
L. positioned in the upper canopy (Pataki and Oren,
2003; Palmroth et al., 2005).
Measurement details are listed in Table 2. Flux
measurements were made with open path LI-7500
IRGAs (Li-Cor, Lincoln, NE) coupled with CSAT-3
sonic anemometers (Campbell Scientific, Logan, UT).
A closed path system with a LI-6400 was employed at
PP before 1 May 2001. Corrections to fluxes measured
by the closed path system that resulted from an
experiment with both open and closed path systems are
described in Oren et al. (2006).
2.2. Measurement period
The 2001–2005 measurement period was notable for
an extended drought that began in the latter part of the
growing season (April–September) in 2001 (hereafter
‘mild drought’) and continued through most of the 2002
growing season (‘severe drought’, Table 3). Mean Ta
was higher during the severe drought than during the
other growing seasons measured here. A severe ice
storm event in December 2002 sharply reduced leaf area
index (LAI) at PP (McCarthy et al., 2006), but resulted
in little observed damage to canopy structure and
canopy conductance at OF and HW (Stoy et al., 2006).
The 2003 growing season was wetter than average (‘wet
year’) and the 2004 growing season experiencedHeight (m)
OF PP HW
Logan, UT 2.0 12 23
4.8 22.2 41.8
Kipp and Zonen, Delft, The Netherlands 4.8 22.2 41.8
Logan, UT 2.8 20.2 39.8
2.8 20.2 39.8
adiometers thereafter.
easurements were calculated after Webb et al. (1980).
Table 3
Interannual variation in the sum (S) of precipitation (P) and photosynthetically active radiation (PAR) and mean air temperature (Ta) during the
April–September growing season (gs) across the 2001–2005 measurement period
Year
P
P (mm gs1) Hydrologic signature
P
PAR (mol photons m2 gs1) Mean Ta (8C)
2001 529 (1s) Mild late-season drought 7274 20.3
2002 371 (2s) Severe drought 7135 21.9
2003 790 (+1s) Wet 6472 20.4
2004 661 Normal 6692 21.0
2005 359 (2s) Severe late-season drought 6669 21.2
Numbers in parentheses are the standard deviation (s) of the measurements from the long-term (110 years) mean. Monthly climatic and hydrologic
variability for the 2001–2004 period is detailed in Stoy et al. (2006).precipitation (P) commensurate with the long-term (110
years) mean (‘average year’). P during the 2005
growing season was extremely low as in 2002, but
the effects of this drought were constrained to the latter
months as in 2001. The cloudier conditions during the
2003–2005 growing seasons resulted in lower photo-
synthetically active radiation (PAR) than during the
previous years (Table 3). Macroclimatic conditions
were identical among ecosystems because they are
adjacent; flux towers lie within 750 m of each other.
Some micrometeorological drivers, including vapor
pressure deficit (D), soil temperature (Ts) and u, varied
slightly among ecosystems and across seasons due to
inter-ecosystem differences in LAI and evapotranspira-
tion (Stoy et al., 2006).
2.3. Flux partitioning methods
We explored four methods of varying complexity to
partition NEE into its RE and GEP components. We
restricted our analysis to methods that parameterize
known relationships between meteorological driving
variables and flux to maximize information transfer
with modeling efforts as mentioned. Two methods, one
based on the annual Q10 model (AQ10) and another on
short-term exponential fits (STE), use measured night-
time fluxes (defined here as periods in which the solar
zenith angle exceeds 908) to model RE as a function of
temperature. The other two methods, one based on the
rectangular hyperbolic fit (RH) and the other on a non-
rectangular hyperbolic fit (NRH), use the intercept of
the relationship between PAR and daytime NEE (NEEd)
to model RE. GEP was then calculated by definition:
GEP ¼ NEE RE (1)
We use the micrometeorological convention where
fluxes from biosphere to atmosphere are denoted as
positive. The NRH was used to gap-fill missing NEEd
measurements for the AQ10 and STE; these methods as
employed here only gap-fill nighttime RE measure-ments, a far larger source of missing EC data. All model
parameters were determined using non-linear least
squares via an unconstrained Gauss–Newton algorithm
(MATLAB, Natick, MA). Optimization via the Leven-
burg–Marquardt algorithm was also investigated but did
not result in substantial differences in annual flux
estimates. We note that the error distribution of EC
measurements may be better approximated as double
exponential (Laplacian), rather than normal (Gaussian),
and thus least absolute regression may be preferred for
estimating model parameters (Hollinger and Richard-
son, 2005; Richardson and Hollinger, 2005; Richardson
et al., 2006). We employed least-squares optimization
here for comparisons with previous studies.
2.4. Annual Q10 (AQ10)
One of the simplest methods for gap-filling missing
nighttime data is to model RE measurements collected
under nighttime conditions of sufficient turbulence
(e.g., u* > 0.2 m s
1, Fig. 1) as a function of a physical
driver, usually Ta or Ts, or to use a slightly more
complex model that might include, for example, u.
Morgenstern et al. (2004) found 29 studies between
1993 and 2001 that employed different methods to
relate RE to T. Here, we use the commonly employed
Q10 model:
REAQ10 ¼ R10QðTa10Þ=1010 (2)
where R10 is base RE at a reference temperature, here
10 8C, and Q10 describes the exponential temperature
response of RE. At both PP and HW, the frequency
characteristics of RE estimated from Eq. (2) better
matched the frequency characteristics of observed RE
when Ta rather than Ts was used as the driving variable
(Stoy et al., 2005). The parameter values of R10 and Q10
were determined for each ecosystem at the annual time
scale. Seasonal variations in u* are likely to result in a
data set that is biased towards colder periods when
parameterizing Eq. (2) (Gu et al., 2005; Reichstein
et al., 2005). This is in part attributable to the fact that
summer-time nocturnal runs experience smaller mean
u* than their winter-counterpart at these ecosystems (see
Juang et al., 2006 or Stoy et al., 2006 for time series of
u* measurements at the study sites).
2.5. Short-term exponential (STE)
Reichstein et al. (2005) noted that the parameters in
Eq. (2) may vary along the annual cycle (cf. Janssens
and Pilegaard, 2003; Palmroth et al., 2005) and
suggested a strategy by which the parameter values
of the base respiration and the temperature response can
be estimated over the time scales at which they might
vary using the ‘short-term exponential’ (STE) method.
They used the Arrhenius (1889) equation after Lloyd
and Taylor (1994):
RESTE ¼ R10;STE exp

E0

1
283:15 T0 
1
Ta  T0

(3)
with a constant T0 parameter (227.13 K, Lloyd and
Taylor, 1994) and a 15-day moving window to deter-
mine variability in the temperature sensitivity parameter
(E0), then a 4-day window to estimate the base respira-
tion parameter (R10,STE). More methodological details
are discussed in Reichstein et al. (2005).
2.6. Rectangular hyperbola (RH)
Lee et al. (1999) noted that the relationship between
accepted nighttime flux data and T was statistically
significant but considerably scattered. To investigate the
seasonal dynamics of RE in a way that was less subject
to the constraints of nighttime EC data quality, they
used the intercept parameter (g) of the rectangular
hyperbolic model (RH, i.e., the Michaelis–Menten
model), which for NEE can be written as (Ruimy et al.,
1995):
NEERH ¼  ab PAR
a PARþ bþ g (4)
where a is the mean apparent ecosystem quantum yield,
b is GEP at light saturation, and signs follow the
micrometeorological convention. Lee et al. (1999) used
a 15-day window to estimate the parameters in Eq. (4),
and then regressed g, as a measure of RE, against Ts. We
determined the parameters of Eq. (4) for monthly
periods because some gaps in the 15 site-year measure-ment period were greater than 15 days, and used
monthly g as an estimate of RE.
2.7. Non-rectangular hyperbola (NRH)
Gilmanov et al. (2003) proposed a method that
employs a non-rectangular hyperbolic fit (NRH):
NEENRH ¼  1
2h
ða PAR þ b

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ða PARþ bÞ2  4abh PAR
q
Þ þ g (5)
where a, b and g have the same definition as in Eq. (4),
and the parameter h describes model curvature as the
NEE/PAR relationship approaches saturation. Eq. (4) is
a special case of Eq. (5) when h = 1. A major metho-
dological difference between the NRH and the RH
methods is that in the NRH, parameters were estimated
for daily periods according to the original use of Eqs. (4)
and (5) (Gilmanov et al., 2003), and missing data were
gap-filled using Eq. (5) with mean monthly parameter
values in the former. Because an unconstrained non-
linear optimization was performed, daily parameter sets
were rejected if parameter values were not realistic (i.e.,
if a, g or h were less than zero and if h was greater than
unity), or if the root mean squared error between model
and measurements exceeded 0.001 mg C m2 s1.
Parameter sets were also rejected in instances where
the Gauss–Newton algorithm did not converge to a
solution within the specified tolerance (here 106)
within the maximum admissible number of iterations
(here 4000). In these cases, the four-parameter NRH
model was a poor descriptor of the data for that day of
the measurement record.
2.8. Thresholds
FPMparameters, and thus annualNEEestimates,may
be sensitive to extreme data values. An added advantage
of the NRH is that it can recursively provide inference
into realistic flux data thresholds that otherwise might
have to be selected subjectively, because extreme values
may result in non-convergence of the curve-fitting
algorithm on a daily basis. We explored the population
of data points from daily data sets when the Gauss–
Newton algorithm converged to an optimum and
determined the minimum admissible NEE measurement
to be 0.7, 0.85, and 0.8 mg C m2 s1 for OF, PP,
and HW, respectively, and the maximum NEE threshold
to be 0.5 mg C m2 s1 for all ecosystems. Measure-
ments that exceeded these thresholds were removed. The
maximum NEE threshold is larger than the maximum
soil respiratory fluxes reported in Palmroth et al. (2005)
for PP and HW, as expected for whole ecosystem
respiration.
To minimize subjectivity in determining a u*
threshold to filter nighttime data (Barford et al.,
2001), we used the atmospheric stability threshold
after Novick et al. (2004), which requires near-neutral
atmospheric stability for nighttime data acceptance.
There are three reasons for adopting this threshold:
(1) Day-time and night-time footprint sizes become
comparable (though not identical). This is an
important consideration if footprint size may exceed
ecosystem dimensions, as is the case here, or if there
is a large spatial component in ecosystem flux
activity, as discussed for the case of PP by Oren
et al. (2006). (2) The flux-transporting eddies under
near-neutral conditions are dominated by ramp-like
structures that appear insensitive to rapid transients
such as passage of clouds during nocturnal conditions
(Cava et al., 2004). (3) Storage fluxes are likely to be
small under near-neutral conditions when compared
to the fluxes above the canopy (Juang et al., 2006).
We note that u* thresholds are employed in the
original usage of the STE. These thresholds are
determined over 3-month periods for each ecosystem
based on the original algorithm described in Reich-
stein et al. (2005). Briefly, flux data are sorted into
temperature and u* classes and the threshold for each
class is taken to be the u* value for which mean flux
exceeded 95% of the mean flux at higher values of u*.
Because the intent of this method is similar to one of
the goals of the atmospheric stability method – to
make flux estimates independent of turbulent statis-
tics – we adopted the atmospheric stability threshold
for the STE as well.
2.9. Information content
In addition to the comparison among FPMs, we
explored the information content (IC) of raw and gap-
filled flux data in both the probability and frequency
domains using the Shannon entropy and a wavelet
variant thereof. We undertook this analysis to investi-
gate how the process of gap-filling generates differences
in the IC relative to that contained in raw data (Katul
et al., 2001a). The Shannon-entropy (ENTS; Shannon,
1948) is a formal method for assessing the IC and is
calculated as
ENTS ¼ 
XN
i
piðxˆÞ lnð piðxˆÞÞ; (6)where pi is the discrete probability density function
(here estimated from a histogram defined over N bins)
of an arbitrary time series having zero mean and unit
variance, i.e., if x denotes the original time series, then
xˆ ¼ ðx x¯Þ=sx where the over bar denotes time aver-
aging and s ¼ ððx x¯Þ2Þ
1=2
is the standard deviation.
Maximum entropy (or minimum IC) is attained when
pi = 1/N, which represents random (white) noise. We
further normalized our entropy estimates by this max-
imum theoretical value such that when pi = 1/N,
ENTS = 1. This normalization provides entropy esti-
mates that are independent of N provided N is suffi-
ciently large (here 200).
ENTS integrates the IC over time through the pdf and
hence is primarily sensitive to the frequency of
occurrence of excursions in NEE measurements from
their mean state. To quantify whether gap-filling results
in a change in the frequency characteristics of NEE (or
alternatively their auto-correlation), we computed the
‘wavelet entropy’ (ENTW) of normalized raw and gap-
filled flux time series after computing orthonormal
wavelet coefficients using the Haar wavelet basis. In
this application, p (Eq. (6)) represents the spectral
energy of the Haar wavelet coefficients computed over a
sequential doubling of frequencies i and
N = log2(87648)  1 ffi 15, where 87,648 is the number
of 1/2 h periods in the 5-year measurement records.
Note that because of the strong locality of the Haar
wavelet basis in the temporal domain, p can be readily
computed for both gap-filled and gap-infected time
series. More details on the application of the wavelet
transformation for flux analysis can be found elsewhere
(Katul et al., 2001b; Braswell et al., 2005; Stoy et al.,
2005).
3. Results and discussion
3.1. Flux partitioning methods
The salient results of the FPM analyses (Fig. 2) are
that (1) flux estimation is highly sensitive to the method
employed, and (2) the simpler methodologies return
unrealistically high interannualmeanfluxmagnitude (the
AQ10 and RH) or variation in flux (AQ10). For example,
mean NEERH at HW is less than 700 g C m2 year1
(Fig. 2A), a stronger C sequestration rate than predicted
by all methods for all ecosystems, even the fast-growing
PP. In contrast, theAQ10 predicts that OF is a C source of
400 g C m2 year1 on average, a situation that would
not be sustainable over the long term. Also, the
interannual variability of NEEAQ10 is higher by over
Table 4
Estimates of net ecosystem exchange (NEE), ecosystem respiration
(RE), and gross ecosystem productivity (GEP) in old field (OF), pine
plantation (PP) and hardwood forest (HW) ecosystems from (A) the
short-time exponential method (STE) of Reichstein et al. (2005) and
(B) the non-rectangular hyperbolic (NRH) method of Gilmanov et al.
(2003)
Ecosystem Year NEE GEP RE
(A) STE
OF 2001 200 1720 1920
2002 200 1270 1470
2003 320 2170 2490
2004 200 1770 1980
2005 170 1310 1480
PP 2001 360 2410 2050
2002 20 2440 2460
2003 220 2080 1860
2004 520 2190 1670
2005 770 2690 1920
HW 2001 230 2240 2000
2002 100 2220 2120
2003 240 1980 1740
2004 350 2000 1650
2005 470 1850 1380
(B) NRH
OF 2001 40 1270 1230
2002 20 970 980
2003 30 1520 1500
2004 30 1360 1390
2005 60 1050 1110
PP 2001 610 1950 1340
2002 270 1880 1610
2003 230 1950 1730
2004 420 2180 1760
2005 740 2580 1840
HW 2001 510 1710 1200
2002 390 1710 1320
2003 170 (400) 1610 (1650) 1440 (1250)
2004 430 1750 1310
2005 490 1720 1230
Numbers in parentheses represent adjusted estimates of 2003 HW
fluxes after accounting for clear-cut effects using the footprint thresh-
olding methodology described in Appendix A. Units are
g C m2 year1. NEE estimates using the NRH for the 2001–2004
period for PP were given in Oren et al. (2006). They are reproduced
here for completeness.
Fig. 2. The mean (A) and standard deviation (B) of annual NEE at old
field (OF), planted pine (PP) and hardwood forest (HW) ecosystems
for the 2001–2005 measurement period. Negative values represent
flux from atmosphere to biosphere. A comprehensive spatial varia-
bility study combined with a flux error analysis by Oren et al. (2006)
suggested that NEE error at PP is on the order of 100 g C m2 year1,
as indicated by the dashed line.50 g C m2 year1 for all ecosystems, in part because it
results in unreasonable values for certain ecosystem-
years (Appendix B). For example, observe the strong C
source predicted for PP in 2002, and the positive NEE
estimates at HW during the drought years of 2001 and
2002 but negative estimates during the drought year of
2005 (seeAppendixB).We, therefore, concentrate on the
contrast between the STE and the NRH, both of which
generated more realistic estimates of NEE, RE, and GEP
(Fig. 2, Table 4). (Flux estimates for each site year
generated by the AQ10 and RH methods are listed in
Table B1 for completeness.) We note that minor
differences in the output of two different techniques is
expected andmay fall within the range of error quantified
in the case of NEE at PP to be over 100 g C m2 year1
for some years (Oren et al., 2006), as indicated by the
dashed line in Fig. 2. Therefore, we interpret the results
from periods in which estimates differed from expecta-
tions by at least that amount. We begin with the STE
method and discuss only the major findings of the
comparisons.
The STE estimated that OF lost nearly
1100 g C m2 year1 (Table 4A) to the atmosphere
over the measurement period, a value that is difficult to
reconcile with the amount of C removed as biomass
during the annual mow, which represented an additional
loss of ca. 100, 100, 400, 200 and 100 g C m2 year1
for 2001–2005, respectively. Despite anomalous
NEESTE values at OF, annual GEPSTE was stronglyrelated to ET (r2 = 0.78, Stoy et al., 2006), as expected
due to the coupling of these two fluxes via stomatal
function.
NEESTE at PP followed an interannual pattern that
was broadly consistent with the drought sensitivity of P.
taeda and the P record during the 5-year study period
(Oren et al., 1998; Pataki and Oren, 2003; Stoy et al.,
2005; Stoy et al., 2006), and the ice storm-induced
reduction in LAI (McCarthy et al., 2006). GEPSTE
ranged from ca. 2000 to 2700 g C m2 year1
(Table 4A) largely within the range predicted by the
independent estimates (1800–2500 g C m2 year1,
Table 1). However, NEESTE was positive in 2002 and
RESTE during 2003 was 600 g C m
2 year1 less than
during severe drought (2002) despite the substantial
production of needle and branch litter by the ice storm,
which is an unexpected result.
GEPSTE differed by less than 20% among years at
HW, as expected at a late successional hardwood forest
comprised largely of drought-tolerant species (Pataki
et al., 1998; Oren and Pataki, 2001; Pataki and Oren,
2003; Palmroth et al., 2005). However, RESTE differed
by nearly 35% among years; the 2001 and 2002 RESTE
estimates were over 600 g C m2 year1 greater than
2005 despite similar drought conditions. Consequently,
the magnitude of NEESTE during 2005 was more than
twice that in 2001 and nearly five times the 2002
estimate. In summary, whereas the STE largely agreed
with expectations of ecosystem flux given prior
knowledge, there were a number of instances where
results may be considered unreasonable.
At OF, mean annual NEENRH was not significantly
different fromzero as determined by a two-tailed unequal
variance t-test ( p = 0.69, t = 0.43; Ruxton, 2006).
However, adding biomass removed by the mow to
NEENRH resulted in a distribution that is significantly
different from zero ( p = 0.24, t = 3.5) with a mean of ca.
200 g C m2 year1. This is the imbalance in the C
budget that cannot be explained by combining EC and
harvest estimates, and represents either measurement
uncertainty or an annual loss of soil C to the atmosphere
or to groundwater via dissolved organic carbon (DOC)
dissolution.At PP, growing nearby on the same soil,DOC
represents a trivial ecosystem C loss compared to other
fluxes (ca. 10 g C m2 year1, Scha¨fer et al., 2003).
Other soil C losses canbe inferred by comparisonwith PP
because both ecosystems shared common management
practices and ecosystem composition across a large
portion of their current extents. Preliminary estimates of
belowground C show similar values (4500 g C m2, K.
Johnsen, unpublished) at both sites. Thus, it is unlikely
that soil C loss explains the imbalance, which appears to
be attributable to measurement uncertainty. Whereas a
lack of closure of 200 g C m2 year1 appears large
compared to the near zero NEE estimates, this term is
within the range of error of GEPNRH and RENRH as
discussed below. The lack of closure that results from
using NRH estimates is approximately half of that which
arises when using the STE.
Annual GEPNRH at OF was strongly related to ET
(r2 = 0.84), but averaged over 400 g C m2 year1 lessthan GEPSTE due to lower RE estimates. The 2001
GEP and RE estimates using the NRH and STE were
on the order of 1300 and 1200 and 1700 and
1900 g C m2 year1, respectively. The GEPNRH and
RENRH estimates for 2001 are closer to those of Novick
et al. (2004), who used a combination of EC and
porometry measurements in conjunction with a big leaf
model to estimate GEP and RE values of 1200 and
1300 g C m2 year1, respectively.
At PP, the magnitude of NEENRH was largest in the
years with late-season droughts, 2001 and 2005.
Although at first glance this appears puzzling, the
finding is reasonable when considering that in both
years growing conditions were ideal until the latter part
of the growing season when the drought intensified;
Palmroth et al. (2005) demonstrated that in both stands
low u greatly reduced soil respiration. Thus, high NEE
in these years is due to high GEP early in the season
followed by low RE later in the season. GEPNRH and
RENRH for PP were slightly larger in magnitude than
estimates for an earlier period (1998–1999) obtained
with an inverse modeling approach (Lai et al., 2002,
Tables 1 and 4B), consistent with the fact that the PP
canopy closed only at the end of 1999. GEPNRH and
RENRH for 2001–2003 at PP were approximately equal
to or slightly lower than estimates from a similar
inversion approach that accounted for local thermal
stratification (Juang et al., 2006). GEP is expected to be
lower than GPP due to the C ‘recycling’ issues
mentioned in Section 1. GEPNRH was equal to or
slightly lower than GPP from budgeting approaches
(Hamilton et al., 2002) and estimates based on sap flux
constraints (Scha¨fer et al., 2003), with a similar degree
of interannual variability (Tables 1 and 4). NEENRH at
PP was lowest in 2003 (225 g C m2 year1), as
expected after ice storm canopy damage, but its
magnitude was over 200 g C m2 year1 less than
estimates from Juang et al. (2006).
Excluding 2003, interannual variability of NEENRH,
GEPNRH, and RENRH at HW were low, on the order of
50–100 g C m2 year1. A two-dimensional footprint
analysis suggests that fluxes during 2003 were impacted
by the December 2002 private land clear-cut (see
Appendix A). After accounting for the clear-cut effects,
HW had the lowest NEE variability among the methods
for all three ecosystems, as expected based on results
from previous sap flux and soil respiration studies
(Pataki and Oren, 2003; Palmroth et al., 2005). This low
interannual flux variability at HW is consistent with
recent findings from a HW site atWalker Branch, TN by
Hanson et al. (2004), who also found NEE variability on
the order of 100 g C m2 year1 for 1993–2000. These
Fig. 3. Monthly net ecosystem exchange (NEE), gross ecosystem productivity (GEP) and ecosystem respiration (RE) at the old field (OF), planted
pine (PP) and hardwood forest (HW) ecosystems in Duke Forest. NRH refers to estimates using the non-rectangular hyperbolic method of Gilmanov
et al. (2003) and STE to estimates using the short-time exponential method of Reichstein et al. (2005). Drought and wet conditions are indicated by
solid and dashed bars, respectively. The severe ice storm in December 2002 is indicated by an asterisk. Flux units are g C m2 month1.values were gap-filled using the ‘look-up table’ method,
incorporating a temperature response curve for RE
similar to the AQ10; we showed earlier that such
approach tends to amplify interannual variability. In
summary, annual flux estimates using the NRH largely
matched independent estimates with the possible
exception of the low 2003 NEE estimate at PP. We
proceed to compare monthly estimates from the STE
and NRH approaches.
Monthly NEE, GEP, and RE estimates from the NRH
and STE generally followed a logical seasonal pattern
(Fig. 3) with the most notable deviation being the
extremely high RESTE estimates during late summer in
2003 at OF. This period was marked with vigorous grass
growth (Stoy et al., 2006), but a monthly respiratory flux
of 500 g C m2 month1 seems unlikely and exceeds
many of the annual RE estimates for grassland
ecosystems worldwide as summarized in Novick
et al. (2004). The dramatic decrease in GEP at OF
during the peak of the 2002 drought is also apparent in
both methods. Grasses were largely dead during this
period. Interestingly, both methods predict that PP is a C
sink during the winter between 2004 and 2005 as icestorm recovery was nearing completion, but wintertime
NEE was slightly positive or near zero on a monthly
basis during other years. HW exhibits later leaf-on
phenology than the other two ecosystems (one of which
is coniferous), and this effect can be observed in the
GEP dynamics of both methods. Finally, the STE
estimates show a decline in peak monthly GEP and RE
through time at PP and HW (Fig. 3); the magnitude of
maximum monthly component fluxes is higher in 2001
and 2002 than in 2003 and 2004 excluding a spike in
RESTE at HW in late 2004. This result is difficult to
reconcile with the fact that the 2001 and 2002 period
experienced drought; the pattern is inconsistent with
estimates based on the NRH approach.
3.2. Information content
The IC of gap-filled NEEd was similar to the IC in the
gap-infested time series in both the Shannon entropy
(ENTS) and wavelet entropy (ENTW) sense; raw and
gap-filled NEEd data differed by a maximum of 2 and
7% for ENTS and ENTW, respectively (Table 5, note that
ENT as defined here exists between 0 and 1). Gap-filled
Table 5
The Shannon entropy (ENTS) and wavelet entropy (ENTW) values for
daytime net ecosystem exchange (NEEd) and ecosystem respiration
(RE) for the 2001–2005measurement period at the old field (OF), pine
plantation (PP) and hardwood forest (HW) study ecosystems
NEEd Ecosystem Raw data Gap-filled data
ENTS OF 0.72 0.71
PP 0.81 0.80
HW 0.80 0.78
ENTW OF 0.27 0.33
PP 0.41 0.39
HW 0.33 0.26
RE Ecosystem Raw data Gap-filled STE Gap-filled NRH
ENTS OF 0.66 0.74 0.51
PP 0.73 0.74 0.54
HW 0.60 0.73 0.46
ENTW OF 0.55 0.29 0.32
PP 0.28 0.28 0.27
HW 0.38 0.41 0.39
Raw data refers to eddy covariance data filtered using the atmospheric
stability thresholding approach of Novick et al. (2004). NRH refers to
nighttime data gap-filled using the non-rectangular hyperbola after
Gilmanov et al. (2003) and STE refers to the short-term exponential
approach of Reichstein et al. (2005). MissingNEEd datawas gap-filled
using the NRH. All entropy estimates are normalized by the maximum
theoretical value.data closely matched the frequency characteristics of
the total NEE time series for all ecosystems (Fig. 4) with
the exception of the higher frequencies, those on the
time scale of hours on the left hand side of the abscissa
in Fig. 4. This result was expected given the known
spatial distribution of ecosystem activity (Oren et al.,
2006). The FPMs did not seek to capture the scale of
variability due to hourly shifts in footprint direction and
extent because we focused on examining the longer-
term estimates more relevant to the ecosystem C cycling
questions referred to in the introduction. Therefore,
gap-filling does not inject new information to the
complete NEE time series in the probability domain,
and in the spectral domains at time scales longer than
approximately 1 day. Despite success in replicating the
time domain and frequency characteristics of measured
NEE data, the special case of nighttime NEE data (i.e.,
RE) is a focus of the present study and warrants more
investigation.
The ENTS of NRH-gap-filled RE was on average
16% lower than that of raw nighttime flux data, and the
ENTS of STE-gap-filled data averaged 23% higher than
RENRH (Table 5). From this analysis, it is clear that gap-
filling using the NRH adds ‘organization’ to the pdf of
nighttime NEE, and that RESTE is more random than itsNRH counterpart. This result follows from the fact that
the NRH models RE as a constant that varies monthly
and the STE models RE as an exponential function of
Ta, itself highly variable on diurnal and seasonal time
scales.
In contrast to the ENTS, both the NRH and STE
decreased the ENTW of RE on average (Table 5). In
other words, gap-filling RE adds order to (or filters out
randomness of) the time series in the spectral domain.
This spectral imbalance is unavoidable unless far more
nighttime runs are accepted, which would reduce data
quality. This result may not be surprising given the
highly stochastic nature of nighttime EC flux measure-
ments, as indicated by the relatively high degree of
spectral energy on the hourly time scale (Fig. 4). At OF
and HW the power spectra of nighttime data increase at
high frequencies, and exhibit the characteristics of high
frequency noise when compared to hourly flux data
from the entire NEE time series. However, the direct
comparison of flux magnitudes indicates that the
magnitude of annual RE was similar to independent
estimates, especially when using the NRH. For these
reasons and those discussed above, the NRH is the
preferable FPM for gap-filling NEE data in the case of
these ecosystems. This finding agrees with the
comparison of EC and inverse model estimates at PP
(Juang et al., 2006) in which RE estimates based on the
intercept of the light response curve compared well at
the annual basis with RE estimates from a Eulerian
version of a constrained source optimization inverse
model (Lai et al., 2002).
3.3. Flux error
We undertook an analysis of error (e) in flux
estimates generated by the NRH to provide a complete
description of flux findings for the 2001–2005 period at
the Duke Forest AmeriFlux ecosystems, noting that the
largest component of flux e may be FPM selection itself
(Hagen et al., 2006), which in part motivated this study.
The choice of the NRH after comparison with
independent and model estimates is intended to
decrease this uncertainty in model selection.
A full analysis of random, instrument, and spatial e
was provided for PP NEE data by Oren et al. (2006). We
expand on this analysis by providing e estimates for
annual NEE, GEP, and RE for all ecosystem-years of
measurement using a variation of the approach of
Goulden et al. (1997) as discussed in greater detail in
Stoy et al. (2006) and Oren et al. (2006). Briefly, e in
NEE and RE due to gap-filling (‘sampling uncertainty’)
was estimated by randomly selecting NRH parameters
Table 6
Error estimates for net ecosystem exchange (NEE), gross ecosystem
productivity (GEP) and ecosystem respiration (RE) generated using
the non-rectangular hyperbolic method (NRH) of Gilmanov et al.
(2003)
Ecosystem Year NEE GEP RE
OF 2001 68 274 (22) 265 (21)
2002 48 290 (30) 286 (29)
2003 49 241 (16) 236 (16)
2004 42 253 (19) 249 (18)
2005 48 177 (17) 170 (15)
PP 2001 92 [108] 188 (11) 165 (13)
2002 100 [111] 235 (16) 214 (19)
2003 96 [95] 183 (11) 156 (14)
2004 84 [69] 167 (9) 146 (11)
2005 53 204 (11) 198 (12)
HW 2001 108 255 (15) 229 (19)
2002 113 309 (18) 287 (21)
2003 84 224 (14) 206 (16)
2004 90 243 (14) 225 (17)
2005 90 231 (13) 211 (17)
Numbers in square brackets are fromOren et al. (2006) and the percent
error of GEP and RE is listed in parentheses. Flux units are
g C m2 year1.
Fig. 4. Orthonormal wavelet spectra using the Haar wavelet basis for raw and gap-filled normalized (Norm.) net ecosystem exchange (NEE) and
ecosystem respiration (RE) data for the Duke Forest old field (OF), pine plantation (PP) and hardwood forest (HW) ecosystems. STE and NRH refer
to the short-term exponential gap-filling method of Reichstein et al. (2005) and the non-rectangular hyperbolic gap-filling method of Gilmanov et al.
(2003), respectively.based on the variance of monthly parameter values
using a Monte Carlo simulation with 100 realizations. e
in the EC system (‘uniform systematic e’) was estimated
using the variance of nighttime ET measurements made
under non-turbulent conditions in the absence of
radiative perturbations (Cava et al., 2004), a situation
that should result in zero flux. These two e sources were
combined to provide an estimate of total error; unlike
Oren et al. (2006), the spatial component of EC e is not
considered here. We used the variance calculated from
the beta parameter of the Laplacian distribution in this
analysis (Richardson and Hollinger, 2005); thus, NEE e
terms (eNEE) are in general slightly smaller than those
reported in Oren et al. (2006) (Table 6).
RE is a component of NEE and the covariance
between eNEE and eRE must be taken into account when
estimating eGEP error. Calculated this way, eGEP
averaged 16% of GEP and eRE averaged 17% of RE,
noting that the magnitude of the former is generally
greater because the forested ecosystems represented a
strong C sink. These relative error estimates are broadly
consistent with but slightly greater than estimates from
other flux studies (Meyers, 2001; Wilson and Meyers,
2001) and represent a conservative e estimate for NRH-
generated GEP and RE values.
4. Summary and conclusion
This study is the first to explore the effect of
vegetation cover, and thereby different momentum
roughness properties and u* characteristics (Fig. 1), on
C flux estimates generated by a suite of FPMs with the
goal of determining which FPM is most robust across
ecosystem type given prior flux estimates. We demon-
strated that the choice of FPM can influence under-
standing of long-term variation in C flux – including the
ability to estimate whether an ecosystem acts as a C
source or sink in a given year – and can change the
apparent ecosystem response to climatic events such as
droughts and ice storms (Fig. 2, Table 4). Results
showed that simpler FPMs (namely, the AQ10 and RH)
produced estimates of C flux magnitude or variability
that were not substantiated by independent measure-
ments. The ‘non-rectangular hyperbolic method’ of
Gilmanov et al. (2003) generated NEE, GEP, and RE
estimates that best matched the magnitude and
interannual variability of independent estimates across
ecosystem types. Thus, a complex method that employs
daytime EC data to estimate RE was the most robust in
the case of these three ecosystems. These results agree
with Juang et al. (2006), who found that nighttime RE
measurements from the intercept of the light response
curve agreed with estimates from a constrained source
optimization on CO2 profile data from PP. However,
across EC research sites, simpler methodologies based
on nighttime data are more commonly used (e.g., Falge
et al., 2001). If the use of nighttime data for RE
estimation is preferred, the short-time exponential
method of Reichstein et al. (2005) provided flux
estimates that were nearer to the expected magnitudes,
although in some cases interannual flux variability was
unexpectedly large.
The information content (IC) of NEEd was not
altered by gap-filling in both the probability and wavelet
domains, but nighttime gap-filling altered both the
probabilistic and spectral structure of the RE time
series. In particular, the wavelet analysis suggests that
RE gap-filling added a high degree of ‘organization’
originating from the surrogate variables or assumptions
used in the gap-filling and did not replicate the
variability of the time series, which exhibited some
characteristics of noise at high frequencies. Despite
these limitations, EC-based RE estimates roughly
matched the magnitude of independent data estimates
over longer-term (e.g., annual) time scales when anappropriate FPM (namely, the NRH) was used. Based
on this analysis, we suggest that more focus be placed
on methods that use daytime NEE data for RE (and
thereby GEP) estimation when interpreting EC data
when possible.
Acknowledgements
Support for this studywas providedby theDepartment
of Energy (DOE) through the FACE-FACTS and
Terrestrial Carbon Processes (TCP) programs, by the
National Institute of Global Environmental Change
(NIGEC) through the Southeast Regional Center at the
University of Alabama, Tuscaloosa (DOE cooperative
agreement DE-FC030-90ER61010) and by the SERC-
NIGEC RCIAP Research Program. We would like to
thank H.R. McCarthy for helpful comments on the
manuscript andK.Wesson, C.-T. Lai, Y. Parashkevov, H.
McCarthy, H.-S Kim, A.C. Oishi, K. Scha¨fer, B. Poulter,
E. Ward, J. Pippen, R. LaMorte, R. Nettles, K. Lewin, G.
Hon, A. Mace and J. Nagy for logistical assistance in the
Duke Forest. Kurt Johnsen of the USDA Forest Service
Southern Forest Research Station provided estimates of
bulk soil C.
Appendix A. Estimating the contribution of the
clear-cut to C exchange at HW
A clear-cut on private land within the extent of HW
200 m south of the tower occurred in late November
2002 (see Fig. 1 in Stoy et al., 2006). We restrict the
analysis of its effects on flux measurements beginning
January 2003, as December 2002 was dominated by the
ice storm and cold conditions that resulted in low
observed RE. We used the two-dimensional representa-
tion of the footprint model of Hsieh et al. (2000) after
Detto et al. (2006) to quantify the fraction of the HW
flux footprint encompassed by the private land clear-cut
for each 30-min measurement period. We then
conducted a sensitivity analysis by filtering data for
which the percent of the footprint in the clear-cut
exceeded certain thresholds. In this way, we estimated
the NEE, GEP, and RE that would have resulted had the
clear-cut not occurred.
Measurements taken when the wind comes from the
south or southwest are more likely to experience
footprint contamination due to the clear-cut, but these
conditions commonly occur during the growing season
and are usually associated with a favorable climate. To
test for any effects of selectively removing data, we also
applied the footprint filtering procedure on 2001 and
2002 NEE data as a control, and expected that the
Fig. A1. The impacts of selectively filtering the clear-cut contribution to the flux footprint on annual net ecosystem exchange (NEE, A), gross
ecosystem productivity (GEP, B) and ecosystem respiration (RE, C) at the hardwood forest ecosystem (HW). The analytical footprint model of Hsieh
et al. (2000) extended to two dimensions by Detto et al. (2006) is used. ‘Maximum contribution of footprint to clear-cut’ is the fraction of footprint
area within the dimensions of the clear-cut.magnitude of NEE would decrease when measurements
collected under relatively ideal climactic conditions
were selectively removed.
Fig. A1 shows the impact of filtering fluxes on NEE,
GEP, and RE estimates (using the NRHM). The right
hand side of Fig. A1 displays flux estimates without
filtering (i.e., the area of the clear-cut is allowed to
comprise over 30% of the footprint) and the left hand
side represents a strong filter (i.e., measurements are
filtered if the clear-cut comprises only 1% of the
footprint). Strengthening the filter to remove fluxes with
<1% potential clear-cut contamination had the poten-
tial to retain too few points with which to obtain stable
estimates of annual flux.
NEE during 2001, 2002, 2004 and 2005 showed a
similar response to selective filtering; the magnitude of
NEE estimates decreases, although not monotonically,
as the strength of filter is increased (from right to left on
the abscissa). The reduction in NEE was expected
because favorable climate is primarily associated with
south and southwesterly prevailing winds from the
direction of the clear-cut as mentioned. NEE estimates
for 2003 differ from other years in that their magnitude
increases from 150 to 300 g C m2 year1 as the
strength of the clear-cut filter increases. The increase
in NEE magnitude was attributable to a ca.100 g C m2 year1 decrease in the magnitude of
GEP (Fig. A1B) and a 250 g C m2 year1 decrease
in RE (Fig. A1C). After applying the clear-cut filter,
GEP estimates in 2001 and 2003 are similar and
200 g C m2 year1 smaller in magnitude than 2002
and 2004. RE estimates are also similar and within
100 g C m2 year1 for all years. Taking into account
the fact that the magnitude of NEE estimates in other
years decreased between 25 and 175 g C m2 year1
from the weakest to strongest filter, the best estimate
of 2003 NEE at HW is on the order of
400 g C m2 year1 (Table 4). This accounts for
the observed NEE of 150 g C m2 year1, adds
150 g C m2 year1 from filtering out clear-cut
effects, and 100 g C m2 year1 to account for the
fact that ideal climatic conditions were selectively
filtered. Accordingly, adjusted 2003 HW RE and
GEP estimates are on the order of 1250 and
1650 g C m2 year1, respectively (Table 4).
Appendix B. Flux estimates from the AQ10 and
RH methods
C flux estimates from the AQ10 and RH for each
ecosystem and measurement year are presented in
Table B1.
Table B1
Estimates of net ecosystem exchange (NEE), ecosystem respiration
(RE), and gross ecosystem productivity (GEP) at the old field (OF),
pine plantation (PP) and hardwood forest (HW) ecosystems
Ecosystem Year AQ10 RH
NEE
OF 2001 570 150
2002 670 60
2003 220 220
2004 70 140
2005 320 150
PP 2001 240 530
2002 210 420
2003 170 290
2004 610 580
2005 840 740
HW 2001 170 680
2002 190 680
2003 160 740
2004 370 690
2005 390 800
GEP
OF 2001 1690 1290
2002 1490 920
2003 1820 1330
2004 1860 1160
2005 1530 1000
PP 2001 2910 1900
2002 3170 1680
2003 2080 1860
2004 2180 2050
2005 2510 2560
HW 2001 2160 1610
2002 2360 1470
2003 2000 1430
2004 2140 1540
2005 1970 1750
RE
OF 2001 2270 1430
2002 2160 980
2003 2040 1110
2004 1930 1020
2005 1850 1150
PP 2001 2670 1370
2002 3390 1260
2003 1910 1560
2004 1580 1470
2005 1670 1810
HW 2001 2320 930
2002 2550 790
2003 1840 690
2004 1770 850
2005 1590 950
‘AQ10’ uses the Q10 respiration equation and RH uses a rectangular
hyperbolic fit. Units are g C m2 year1.References
Arrhenius, S., 1889. U¨ber die Reaktionsgeschwindigkeit bei der
Inversion von Rohrzucker durch Sauren. Z. Phys. Chem. 4,
226–248.
Baldocchi, D., Falge, E., Gu, L.H., Olson, R., Hollinger, D., Running,
S., Anthoni, P., Bernhofer, C., Davis, K., Evans, R., Fuentes, J.,
Goldstein, A., Katul, G., Law, B., Lee, X.H., Malhi, Y., Meyers, T.,
Munger, W., Oechel, W., Paw, U.K.T., Pilegaard, K., Schmid, H.P.,
Valentini, R., Verma, S., Vesala, T., Wilson, K., Wofsy, S., 2001.
FLUXNET: a new tool to study the temporal and spatial variability
of ecosystem-scale carbon dioxide, water vapor, and energy flux
densities. Brit. Am. Meteorol. Soc. 82, 2415–2434.
Barford, C.C., Wofsy, S.C., Goulden, M.L., Munger, J.W., Pyle, E.H.,
Urbanski, S.P., Hutyra, L., Saleska, S.R., Fitzjarrald, D., Moore,
K., 2001. Factors controlling long- and short-term sequestration of
atmospheric CO2 in a mid-latitude forest. Science 294, 1688–
1691.
Braswell, B.H., Sacks, W.J., Linder, E., Schimel, D.S., 2005. Estimat-
ing diurnal to annual ecosystem parameters by synthesis of a
carbon flux model with eddy covariance net ecosystem exchange
observations. Global Change Biol. 11, 335–355.
Campbell, G.S., Norman, J.M., 1998. An Introduction to Environ-
mental Biophysics. Springer, New York.
Cava, D., Giostra, U., Siqueira, M.B.S., Katul, G.G., 2004. Organized
motion and radiative perturbations in the nocturnal canopy sub-
layer above an even-aged pine forest. Bound. LayerMeteorol. 112,
129–157.
Detto, M., Montaldo, N., Albertson, J.D., Mancini, M., Katul, G.G.,
2006. Soil moisture and vegetation controls on evapotranspiration
in a heterogeneous Mediterranean ecosystem on Sardinia, Italy.
Water Resour. Res. 42, W08419, doi:10.1029/2005WR004693.
Ellsworth, D.S., Oren, R., Huang, C., Phillips, N., Hendrey, G.R.,
1995. Leaf and canopy responses to elevated CO2 in a pine forest
under free-air CO2 enrichment. Oecologia 104, 139–146.
Falge, E., Baldocchi, D., Olson, R., Anthoni, P., Aubinet, M., Bern-
hofer, C., Burba, G., Ceulemans, R., 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
defensible annual sums of net ecosystem exchange. Agric. For.
Meteorol. 107, 43–69.
Gilmanov, T.G., Verma, S.B., Sims, P.L., Meyers, T.P., Bradford, J.A.,
Burba, G.G., Suyker, A.E., 2003. Gross primary production and
light response parameters of four southern plains ecosystems
estimated using long-term CO2-flux tower measurements. Global
Biogeochem. Cy. 17, 1071 doi:10.1029/2002GB002023.
Goulden, M.L., Daube, B.C., Fan, S.-M., Sutton, D.J., Bazzaz, A.,
Munger, J.W., Wofsy, S.C., 1997. Physiological responses of a
black spruce forest to weather. J. Geophys. Res. Atmos. 102,
28987–28996.
Gu, L.H., Falge, E.M., Boden, T., Baldocchi, D.D., Black, T.A.,
Saleska, S.R., Suni, T., Verma, S.B., Vesala, T., Wofsy, S.C.,
Xu, L.K., 2005. Objective threshold determination for nighttime
eddy flux filtering. Agric. For. Meteorol. 128, 179–197.
Hagen, S.C., Braswell, B.H., Linder, E., Frolking, S., Richardson,
A.D., Hollinger, D.Y., 2006. Statistical uncertainty of eddy flux-
based estimates of gross ecosystem carbon exchange at Howland
Forest. Maine. J. Geophys. Res. Atmos. 111, D08S03 doi:10.1029/
2005JD006154.
Hamilton, J.G., De Lucia, E.H., George, K., Naidu, S.L., Finzi, A.C.,
Schlesinger, W.H., 2002. Forest carbon balance under elevated
CO2. Oecologia 131, 250–260.
Hanson, P.J., Amthor, J.S., Wullschleger, S.D., Wilson, K.B., Grant,
R.F., Hartley, A., Hui, D., Hunt Jr., E.R., Johnson, D.W., Kimball,
J.S., King, A.W., Luo, Y., McNulty, S.G., Sun, G., Thornton, P.E.,
Wang, S., Williams, M., Baldocchi, D.D., Cushman, R.M., 2004.
Oak forest carbon and water simulations: model intercomparisons
and evaluations against independent data. Ecol. Monogr. 74, 443–
489.
Hollinger, D.Y., Richardson, A.D., 2005. Uncertainty in eddy covar-
iance measurements and its application to physiological models.
Tree Physiol. 25, 873–885.
Hsieh, C.I., Katul, G., Chi, T., 2000. An approximate analytical model
for footprint estimation of scalar fluxes in thermally stratified
atmospheric flows. Adv. Water Resour. 23, 765–772.
Hui, D.F., Wan, S.Q., Su, B., Katul, G., Monson, R., Luo, Y.Q., 2004.
Gap-filling missing data in eddy covariance measurements using
multiple imputation (MI) for annual estimations. Agric. For.
Meteorol. 121, 93–111.
Janssens, I.A., Pilegaard, K., 2003. Large seasonal changes in Q(10)
of soil respiration in a beech forest. Global Change Biol. 9, 911–
918.
Juang, J.-Y., Katul, G.G., Siqueira, M.B.S., Stoy, P.C., Palmroth, S.,
McCarthy, H.R., Kim, H.-S., Oren, R., 2006. Modeling nighttime
ecosystem respiration from measured CO2 concentration and air
temperature profiles using inverse methods. J. Geophys. Res.
Atmos. 111, D08S05 doi:10.1029/2005JD005976.
Katul, G.G., Hsieh, C.-I., Bowling, D., 1999. Spatial variability of
turbulent fluxes in the roughness sublayer of an even-aged pine
forest. Bound. Layer Meteorol. 93, 1–28.
Katul, G.G., Lai, C.-T., Albertson, J.D., Vidakovic, B., Scha¨fer,
K.V.R., Hsieh, C.I., Oren, R., 2001a. Quantifying the complexity
in mapping energy inputs and hydrologic state variables into land-
surface fluxes. Geophys. Res. Lett. 28, 3305–3307.
Katul, G.G., Lai, C.-T., Scha¨fer, K.V.R., Vidakovic, B., Albertson,
J.D., Ellsworth, D.S., Oren, R., 2001b. Multiscale analysis of
vegetation surface fluxes: from seconds to years. Adv. Water
Resour. 24, 1119–1132.
Lai, C.-T., Katul, G.G., 2000. The dynamic role of root-water uptake
in coupling potential to actual transpiration. Adv. Water Resour.
23, 427–439.
Lai, C.T., Katul, G., Butnor, J., Ellsworth, D., Oren, R., 2002.
Modelling night-time ecosystem respiration by a constrained
source optimization method. Global Change Biol. 8, 124–141.
Law, B.E., Falge, E., Gu, L., Baldocchi, D.D., Bakwin, P., Berbigier,
P., Davis, K., Dolman, A.J., Falk, M., Fuentes, J.D., Goldstein, A.,
Granier, A., Grelle, A., Hollinger, D., Janssens, I.A., Jarvis, P.,
Jensen, N.O., Katul, G., Mahli, Y., Matteucci, G., Meyers, T.,
Monson, R., Munger, W., Oechel, W., Olson, R., Pilegaard, K.,
Paw, K.T., Thorgeirsson, H., Valentini, R., Verma, S., Vesala, T.,
Wilson, K., Wofsy, S., 2002. Environmental controls over carbon
dioxide and water vapor exchange of terrestrial vegetation. Agric.
For. Meteorol. 113, 97–120.
Lee, X., Fuentes, J.D., Staebler, R.M., Neumann, H.H., 1999. Long-
term observation of the atmospheric exchange of CO2 with a
temperate deciduous forest in southern Ontario, Canada. J. Geo-
phys. Res. Atmos. 104, 15975–15984.
Lloyd, J., Taylor, J.A., 1994. On the temperature dependence of soil
respiration. Funct. Ecol. 8, 315–323.
Luo, Y.Q., Medlyn, B.E., Hui, D.F., Ellsworth, D., Reynolds, J., Katul,
G.G., 2001. Gross primary productivity in Duke Forest: modelingsynthesis of CO2 experiment and eddy-flux data. Ecol. Appl. 11,
239–252.
McCarthy, H.R., Oren, R., Johnsen, K.H., Pritchard, S.G., Davis,
M.A., Maier, C., Kim, H.-S., 2006. Ice storms and management
practices interact to affect current carbon sequestration in forests
with potential mitigation under future CO2 atmosphere. J.
Geophys. Res. Atmos 111, D15103, doi:10.1029/2005JD
006428.
Meyers, T.P., 2001. A comparison of summertime water and CO2
fluxes over rangeland for well watered and drought conditions.
Agric. For. Meteorol. 106, 205–214.
Morgenstern, K., Black, T.A., Humphreys, E.R., Griffis, T.J., Drewitt,
G.B., Cai, T.B., Nesic, Z., Spittlehouse, D.L., Livingstone, N.J.,
2004. Sensitivity and uncertainty of the carbon balance of a Pacific
Northwest Douglas-fir forest during an El Nino La Nina cycle.
Agric. For. Meteorol. 123, 201–219.
Mortazavi, B., Chanton, J.P., Prater, J.L., Oishi, A.C., Oren, R., Katul,
G.G., 2005. Temporal variability in 13C of respired CO2 in a pine
and a hardwood forest subject to similar climatic conditions.
Oecologia 142, 57–69.
Novick, K.A., Stoy, P.C., Katul, G.G., Ellsworth, D.S., Siqueira,
M.B.S., Juang, J., Oren, R., 2004. Carbon dioxide and water vapor
exchange in a warm temperate grassland. Oecologia 138, 259–
274.
Oosting, H.J., 1942. An ecological analysis of the plant communities
of Piedmont, North Carolina. Am. Midl. Nat. 28, 1–126.
Oren, R., Ewers, B.E., Todd, P., Phillips, N., Katul, G., 1998. Water
balance delineates the soil layer in which moisture affects canopy
conductance. Ecol. Appl. 8, 990–1002.
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., Hsieh, C.-I., Stoy, P.C., Albertson, J., McCarthy, H.R.,
Harrell, P., Katul, G.G., 2006. Estimating the uncertainty in annual
net ecosystem carbon exchange: spatial variation in turbulent
fluxes and sampling errors in eddy-covariance measurements.
Global Change Biol. 12, 883–896.
Palmroth, S., Maier, C.A., McCarthy, H.R., Oishi, A.C., Kim, H.-S.,
Johnsen, K.H., Katul, G.G., Oren, R., 2005. Contrasting responses
to drought of the forest floor CO2 efflux in a loblolly pine
plantation and a nearby oak-hickory forest. Global Change Biol.
11, 421–434.
Pataki, D.E., Oren, R., 2003. Species differences in stomatal control of
water loss at the canopy scale in a mature bottomland deciduous
forest. Adv. Water Resour. 26, 1267–1278.
Pataki, D.E., Oren, R., Katul, G.G., Sigmon, J., 1998. Canopy con-
ductance of Pinus taeda, Liquidambar styraciflua and Quercus
phellos under varying atmospheric and soil water conditions. Tree
Physiol. 18, 307–315.
Poggi, D., Katul, G.G., Albertson, J.D., 2004. Momentum transfer and
turbulent kinetic energy budgets within a dense model canopy.
Bound. Layer Meteorol. 111, 589–614.
Reichstein, M., Falge, E., Baldocchi, D., Papale, D., Aubinet, M.,
Berbigier, P., Bernhofer, C., Buchmann, N., Gilmanov, T.G.,
Granier, A., Gru¨nwald, T., Havra´nkova´, K., Ilvesniemi, H., Janous,
D., Knohl, A., Laurila, T., Lohila, A., Loustau, D., Matteucci, G.,
Meyers, T., Miglietta, F., Ourcival, J.-M., Pumpanen, J., Rambal,
S., Rotenberg, E., Sanz, M., Tenhunen, J., Seufert, G., Vaccari, F.,
Vesala, T., Yakier, D., Valentini, R., 2005. On the separation of net
ecosystem exchange into assimilation and ecosystem respiration:
review and improved algorithm. Global Change Biol. 11, 1424–
1439.
Richardson, A.D., Hollinger, D.Y., 2005. Statistical modeling of
ecosystem respiration using eddy covariance data: maximum
likelihood parameter estimation, and Monte Carlo simulation of
model and parameter uncertainty, applied to three simple models.
Agric. For. Meteorol. 131, 191–208.
Richardson, A.D., Hollinger, D.Y., Burba, G.G., Davis, K.J., Flana-
gan, L.B., Katul, G.G., Munger, J.W., Ricciuto, D.M., Stoy, P.C.,
Suyker, A.E., Verma, S.B., Wofsy, S.C., 2006. A multi-site
analysis of uncertainty in tower-based measurements of carbon
and energy fluxes. Agric. For. Meteorol. 136, 1–18.
Ruimy, A., Jarvis, P.G., Baldocchi, D.D., Saugier, B., 1995. CO2 fluxes
over plant canopies and solar radiation: a review. Adv. Ecol. Res.
26, 1–68.
Ruxton, G.D., 2006. The unequal variance t-test is an underused
alternative to Student’s t-test and the Mann–Whitney U test.
Behav. Ecol. 17, 688–690.
Sarmiento, J., Wofsy, S. (co-chairs), 1999. AUS carbon cycle science
plan: a report of the carbon and climate working group, US Global
Change Research Program, Washington, DC.
Scha¨fer, K.V.R., Oren, R., Ellsworth, D.S., Lai, C.T., Herrick, J.D.,
Finzi, A.C., Richter, D.D., Katul, G.G., 2003. Exposure toan enriched CO2 atmosphere alters carbon assimilation and
allocation in a pine forest ecosystem. Global Change Biol. 9,
1378–1400.
Shannon, C.E., 1948. A mathematical theory of communication.
AT&T Tech. J. 27 379–423 and 623–656.
Stoy, P.C., Katul, G.G., Siqueira, M.B.S., Juang, J.-Y., McCarthy,
H.R., Kim, H.-S., Oishi, A.C., Oren, R., 2005. Variability in net
ecosystem exchange from hourly to interannual time scales at
adjacent pine and hardwood forests: a wavelet analysis. Tree
Physiol. 25, 887–902.
Stoy, P.C., Katul, G.G., Siqueira, M.B.S., Juang, J.-Y., McCarthy,
H.R., Oishi, A.C., Uebelherr, J.M., Kim, H.-S., Oren, R., 2006.
Separating the effects of climate and vegetation on evapotran-
spiration along a successional chronosequence in the southeastern
US. Global Change Biol. 12, 2115–2135.
Webb, E.K., Pearman, G.I., Leuning, R., 1980. Correction of flux
measurements for density effects due to heat and water vapour
transfer. Q. J. Roy. Meteor. Soc. 106, 85–100.
Wilson, K.B., Meyers, T.P., 2001. The spatial variability of energy and
carbon dioxide fluxes at the floor of a deciduous forest. Bound.
Layer Meteorol. 98, 443–473.