Division of Engineering and Applied Science/Department of Earth and Planetary Science, Harvard University, Cambridge, MA 02138, USA
m
,
a g r i c u l t u r a l a n d f o r e s t m e t e o r o l o g y 1 4 8 ( 2 0 0 8 ) 1 8 2 7 ? 1 8 4 7
avai lab le at www.sc iencedi rec t .comDepartment of Forest Resources, University of Minnesota, St. Paul, MN 55108, USA
n Department of Ecology and Evolutionary Biology, University of Colorado, Boulder, CO 80309, USA
o Argonne National Laboratory, Environmental Science Division, Argonne, IL 60439, USA
p Department of Evolution, Ecology, and Organismal Biology, Ohio State University, Columbus, OH 43210, USA
q Smithsonian Environmental Research Center, Edgewater, MD 21037, USA
r Lawrence Berkeley National Laboratory, Environmental Energy Technologies Division, Atmospheric Science Department, Berkeley, CA 94720
USA
s Harvard Forest and Department of Organismic and Evolutionary Biology, Harvard University, Petersham, MA 01366, USA
t Oak Ridge National Laboratory Environmental Sciences Division, Oak Ridge, TN 37831, USA
u Harvard Forest, Harvard University, Petersham, MA 01366, USA
* Corresponding author. Tel.: +1 814 865 9617.
E-mail address: jing@psu.edu (J. Xiao).
1 Present address: Department of Meteorology, Pennsylvania State University, University Park, State College, PA 16802.
0168-1923/$ ? see front matter # 2008 Elsevier B.V. All rights reserved.Estimation of net ecosystem carbon exchange for the
conterminous United States by combining MODIS and
AmeriFlux data
Jingfeng Xiao a,*,1, Qianlai Zhuang b, Dennis D. Baldocchi c, Beverly E. Law d,
Andrew D. Richardson e, Jiquan Chen f, Ram Oren g, Gregory Starr h, Asko Noormets i,
Siyan Ma c, Shashi B. Verma j, Sonia Wharton k, Steven C. Wofsy l, Paul V. Bolstadm,
Sean P. Burns n, David R. Cook o, Peter S. Curtis p, Bert G. Drake q, Matthias Falk k,
Marc L. Fischer r, David R. Foster s, Lianhong Gu t, Julian L. Hadley u, David Y. Hollinger v,
Gabriel G. Katul g, Marcy Litvakw, Timothy A. Martin x, Roser Matamala y, Steve McNulty z,
Tilden P. Meyers A, Russell K. Monson n, J. William Munger B, Walter C. Oechel C,
Kyaw Tha Paw U k, Hans Peter Schmid D,E, Russell L. Scott F, Ge Sun z, Andrew E. Suyker j,
Margaret S. Torn G
a Department of Earth & Atmospheric Sciences, Purdue Climate Change Research Center, Purdue University, West Lafayette, IN 47907, USA
b Department of Earth & Atmospheric Sciences, Department of Agronomy, Purdue Climate Change Research Center, Purdue University, West
Lafayette, IN 47907, USA
c Ecosystem Science Division, Department of Environmental Science, Policy and Management, University of California, Berkeley, CA 94720,
USA
d College of Forestry, Oregon State University, Corvallis, OR 97331, USA
eComplex Systems Research Center Institute for the Study of Earth, Oceans and Space, University of New Hampshire, Durham, NH 03824,
USA
f Department of Environmental Sciences, University of Toledo, Toledo, OH 43606, USA
gNicholas School of the Environment, Duke University, Durham, NC 27708, USA
h Department of Biological Sciences, University of Alabama, Tuscaloosa, AL 35487, USA
i Department of Forestry and Environmental Resources and Southern Global Change Program, North Carolina State University, Raleigh, NC
27695, USA
j School of Natural Resources, University of Nebraska-Lincoln, Lincoln, NE 68583, USA
k Department of Land, Air and Water Resources, University of California, Davis, CA 95616, USA
l
journa l homepage: www.e lsev ier .com/ locate /agr formetdoi:10.1016/j.agrformet.2008.06.015
038
M 8
39
6, U
idg
am
21
05,
an
19,
kel
1. Introduction
Net ecosystem carbon exchange (N
between photosynthetic uptake and release of carbon dioxide sparseness of the CO2 observation network and their biased
Accepted 26 June 2008
Keywords:
Net ecosystem carbon exchange
MODIS
AmeriFlux
NEE
Regression tree
Eddy covariance
tow
a g r i c u l t u r a l a n d f o r e s t m e t e o r o l o g y 1 4 8 ( 2 0 0 8 ) 1 8 2 7 ? 1 8 4 71828(CO2) by respiration from autotrophs (plants) and hetero-
trophs (e.g., microbial decomposition), represents the net
exchange of CO2 between terrestrial ecosystems and the
atmosphere (Law et al., 2006). The quantification of NEE for
regions, continents, or the globe can improve our under-
standing of the feedbacks between the terrestrial biosphere
and the atmosphere in the context of global change and
facilitate climate policy-making.
To date, several techniques have been used to estimate NEE
(Baldocchi et al., 2001). Atmospheric inverse models (e.g., Tans
et al., 1990; Denning et al., 1996; Fan et al., 1998; Gurney et al.,
2002; Deng et al., 2007), biogeochemical models (e.g., Potter
et al., 1993; Running and Hunt, 1993; Field et al., 1995; Zhuang
placement in the marine boundary layers (Tans et al., 1990;
Denning et al., 1996; Fan et al., 1998). Moreover, this approach
does not provide information about which ecosystems are
contributing to the sinks/sources or the processes involved
(Janssens et al., 2003). Most biogeochemical models, however,
are dependent on site level parameterizations, which may
limit the accuracy of model simulations over large areas.
Inventory approaches provide information on NEE at multi-
year to decadal timescales, and therefore do not provide
information on seasonal and interannual variability of NEE
and shorter-term physiological mechanisms (Baldocchi et al.,
2001).
At the site level, eddy covariance flux towers have been
et al., 2003), and inventory approaches (e
Goodale et al., 2002) have been used toEE), the difference
CO2 and provide aggregated information on NEE over large
areas during the past two decades. The accuracy of the
estimates by atmospheric inverse models is limited by the
exchange of carbon dioxide between the terrestrial biosphere and the atmosphere for
regions or continents, flux tower measurements need to be extrapolated to these large
areas. Here we used remotely sensed data from the Moderate Resolution Imaging Spectro-
meter (MODIS) instrument on board the National Aeronautics and Space Administration?s
(NASA) Terra satellite to scale up AmeriFlux NEE measurements to the continental scale. We
first combined MODIS and AmeriFlux data for representative U.S. ecosystems to develop a
predictive NEE model using a modified regression tree approach. The predictive model was
trained and validated using eddy flux NEE data over the periods 2000?2004 and 2005?2006,
respectively. We found that the model predicted NEE well (r = 0.73, p < 0.001). We then
applied the model to the continental scale and estimated NEE for each 1 km  1 km cell
across the conterminous U.S. for each 8-day interval in 2005 using spatially explicit MODIS
data. The model generally captured the expected spatial and seasonal patterns of NEE as
determined from measurements and the literature. Our study demonstrated that our
empirical approach is effective for scaling up eddy flux NEE measurements to the con-
tinental scale and producing wall-to-wall NEE estimates across multiple biomes. Our
estimates may provide an independent dataset from simulations with biogeochemical
models and inverse modeling approaches for examining the spatiotemporal patterns of
NEE and constraining terrestrial carbon budgets over large areas.
# 2008 Elsevier B.V. All rights reserved.Received in revised form
23 June 2008
only represent the caReceived 22 March 2008 exchange (NEE) for a wide range of climate and biome types. However, these measurements
rbon fluxes at the scale of the tower footprint. To quantify the netvUSDA Forest Service, Northeastern Research Station, Durham, NH
wDepartment of Biology, University of New Mexico, Albuquerque, N
x University of Florida, Gainesville, FL 32611, USA
yArgonne National Laboratory, Biosciences Division, Argonne, IL 604
z USDA Forest Service, Southern Research Station, Raleigh, NC 2760
A NOAA/ARL, Atmospheric Turbulence and Diffusion Division, Oak R
B Department of Earth and Planetary Sciences, Harvard University, C
CDepartment of Biology, San Diego State University, San Diego, CA 9
D Department of Geography, Indiana University, Bloomington, IN 474
E Atmospheric Environmental Research, Institute of Meteorology
Kreuzeckbahnstr. 19, 82467 Garmisch-Partenkirchen, Germany
FUSDA-ARS Southwest Watershed Research Center, Tucson, AZ 857
G Lawrence Berkeley National Laboratory, Earth Science Division, Ber
a r t i c l e i n f o
Article history:
a b s t r a c t
Eddy covariance flux.g., Pacala et al., 2001;
infer net exchange of24, USA
7131, USA
, USA
SA
e, TN 37831, USA
bridge, MA 02138, USA
82, USA
USA
d Climate Research, Research Center Karlsruhe (FZK/IMK-IFU),
USA
ey, CA 94720, USA
ers provide continuous measurements of net ecosystem carbonproviding continuous measurements of ecosystem level
exchanges of carbon at half-hourly or hourly time steps since
scaling eddy flux NEE measurements to large areas (Running
a g r i c u l t u r a l a n d f o r e s t m e t e o r o l o g y 1 4 8 ( 2 0 0 8 ) 1 8 2 7 ? 1 8 4 7 1829et al., 1999). There have been several studies developing
methods for integration of flux data with remote sensing data
to quantify NEE over large areas. For example, Yamaji et al.
(2007) linked Moderate Resolution Imaging Spectroradiometer
(MODIS) data to eddy flux NEE data for regional extrapolation
to deciduous broadleaf forests over Japan. Wylie et al. (2007)
estimated NEE for grasslands in the northern Great Plains
using satellite data from the SPOT (Satellite Pour l?Observation
de la Terre) VEGETATION sensor and eddy flux NEE measure-
ments. Papale and Valentini (2003) estimated NEE for
European forests using flux tower data and satellite data
derived from the Advanced Very High Resolution Radiometer
(AVHRR). Despite these efforts, to our knowledge, no study has
scaled eddy flux NEE measurements to the continental scale
and produced spatially explicit estimates of NEE across
multiple biomes.
Here we used remotely sensed data from the National
Aeronautics and Space Administration?s (NASA) Terra
MODIS to scale eddy flux NEE measurements to the
continental scale and produce wall-to-wall NEE estimates
for the conterminous U.S. First, we developed a predictive
NEE model based on site-specific MODIS and AmeriFlux data.
Second, we validated the performance of the model with
AmeriFlux data. Third, we applied the model to estimate NEE
for each 1 km  1 km cell across the conterminous U.S. for
each 8-day period in 2005 using wall-to-wall MODIS data.
Finally, we examined the spatiotemporal patterns of NEE
across the conterminous U.S.
2. Methods
2.1. Piecewise linear regression models
A modified regression tree approach was used to scale tower-
based NEE to the continental scale. Regression tree algorithms
produce rule-based models containing one or more rules, each
of which is a set of conditions associated with a linear
submodel. Regression tree models allow both continuous and
discrete variables as input variables, and account for a non-
linear relationship between predictive and target variables
(Yang et al., 2003). These approaches are also proving not only
more effective than simple techniques including multivariatethe early 1990s (Wofsy et al., 1993; Baldocchi et al., 2001). At
present, over 400 eddy covariance flux towers are operating on
a long-term and continuous basis over the globe (FLUXNET,
2008). This global network encompasses a large range of
climate and biome types (Baldocchi et al., 2001), and provides
the longest, most extensive, and most reliable measurements
of NEE. However, these measurements only represent the
fluxes from the scale of the tower footprint (Running et al.,
1999) up to several square kilometers (Schmid, 1994; Go?ckede
et al., 2008). To quantify the net exchange of CO2 between the
terrestrial biosphere and the atmosphere, we need to scale
these flux tower measurements to regions, continents, or the
globe.
Satellite remote sensing is a potentially valuable tool forlinear regression, but also easier to understand than neural
networks (Huang and Townshend, 2003). We used a modifiedregression tree algorithm implemented in the commercial
software called Cubist. Cubist is a powerful tool for generating
rule-based predictive models. A Cubist model resembles a
piecewise linear model, except that the rules can overlap with
one another (RuleQuest, 2008). Cubist has been used to
estimate percent land cover (Huang and Townshend, 2003),
impervious area (Yang et al., 2003), forest biomass (Salajanu
and Jacobs, 2005), and ecosystem carbon fluxes (Wylie et al.,
2007). Piecewise regression models were selected as the most
appropriate approach for scaling the flux tower data to
ecoregions (Wylie et al., 2007).
We chose Cubist to construct a predictive NEE model based
on AmeriFlux NEE and satellite data. In Cubist, the predictive
accuracy of a rule-based model can be improved by combining
it with an instance-based/nearest-neighbor model that pre-
dicts the target value of a new case using the average predicted
values of the n most similar cases (RuleQuest, 2008). The use of
the composite model can improve the predictive accuracy
relative to the rule-based model alone. Cubist can also
generate committee models made up of several rule-based
models, and each member of the committee model predicts
the target value for a case (RuleQuest, 2008). The member?s
predictions are averaged to give a final prediction.
Cubist uses three statistical measures to measure the
quality of the constructed predictive model, including average
error, relative error, and Pearson product?moment correlation
coefficient. The average error (EA) is calculated as (Yang et al.,
2003):
EA ?
1
N
XN
i?1
jyi  y?ij (1)
where N is the number of samples used to establish the
predictive model, and yi and y?i are the actual and predicted
values of the response variable, respectively. The relative error
(ER) is calculated as (Yang et al., 2003):
ER ?
EA;T
EA;m
(2)
where EA,T is the average error of the constructed model, and
EE,m is the average error that would result from always pre-
dicting the mean value. The Pearson product?moment corre-
lation coefficient is a common measure of the correlation
between two variables. All the three statistical measures pro-
vided by Cubist were used to evaluate the performance of the
predictive model.
2.2. Explanatory variables
NEE is the difference between two carbon fluxes of photo-
synthesis and respiration (Law et al., 1999). It is influenced by a
variety of physical, physiological, atmospheric, hydrologic,
and edaphic variables. At the leaf level, photosynthesis or
gross primary productivity (GPP) is influenced by several
factors, including incoming solar radiation, air temperature,
vapor pressure deficit, soil moisture, and nitrogen availability
(Clark et al., 1999, 2004). At the ecosystem level, GPP is also
influenced by leaf area index (LAI) and canopy phenology.
Ecosystem respiration (Re) includes autotrophic (Ra) and
heterotrophic respiration (Rh). Soil respiration is the largest
Table 1 ? Site descriptions including name, latitude, longitude, vegetation structure, years of data available, and references for each flux site in this study
Site State Latitude Longitude Vegetation structure Vegetation type Year References
Audubon Research Ranch (ARR) AZ 31.59 110.51 Desert grasslands Grasslands 2002?2006
Santa Rita Mesquite (SRM) AZ 31.82 110.87 Mesquite-dominated savanna Savannas 2004?2006 Watts et al. (2007)
Walnut Gulch Kendall Grasslands (WGK) AZ 31.74 109.94 Warm season C4 grassland Grasslands 2004?2006
Sky Oaks Old Stand (SOO) CA 33.37 116.62 Chaparral (Mediterranean-type ecosystems) Shrublands 2004?2006 Lipson et al. (2005)
Sky Oaks Young stand (SOY) CA 33.38 116.62 Chaparral (Mediterranean-type ecosystems) Shrublands 2001?2006 Lipson et al. (2005)
Tonzi Ranch (TR) CA 38.43 120.97 Oak savanna, grazed grassland dominated by
blue oak and grasses
Savannas 2001?2006 Ma et al. (2007)
Vaira Ranch (VR) CA 38.41 120.95 Grazed C3 grassland opening in a region of
oak/grass savanna
Savannas 2001?2006 Xu and Baldocchi
(2004)
Niwot Ridge Forest (NRF) CO 40.03 105.55 Subalpine coniferous forest dominated
by subalpine, Engelmann spruce, and
lodgepole pine
Evergreen forests 2000?2003 Monson et al. (2002)
Kennedy Space Center-Scrub Oak (KSC) FL 28.61 80.67 Scrub-oak palmetto dominated by
schlerophyllous evergreen oaks and the
Saw Palmetto Serenoa repens
Shrublands 2000?2006 Dore et al. (2003)
Austin Cary-Slash Pine (AC) FL 29.74 82.22 Naturally regenerated pine dominated by
Pinus palustris/Pinus ellottii
Evergreen forests 2001?2005 Powell et al. (2005)
Bondville (Bon) IL 40.01 88.29 Annual rotation between corn (C4) and soybeans (C3) Croplands 2001?2006 Hollinger et al. (2005)
FNAL Agricultural site (FAg) IL 41.86 88.22 Soybean/corn Croplands 2005?2006
FNAL Prairie site (FPr) IL 41.84 88.24 Tall grass prairie Grasslands 2004?2006
Morgan Monroe State Forest (MMS) IN 39.32 86.41 Mixed hardwood deciduous forest dominated
by sugar maple, tulip poplar, sassafras, white
oak, and black oak
Deciduous forests 2000?2005 Schmid et al. (2000)
Harvard Forest EMS Tower (HFE) MA 42.54 72.17 Temperate deciduous forest dominated by red oak,
red maple, black birch, white pine,
and hemlock
Deciduous forests 2000?2004 Urbanski et al. (2007)
Harvard Forest Hemlock (HFH) MA 42.54 72.18 Temperate coniferous forest dominated by hemlock Evergreen forests 2004
Little Prospect Hill (LPH) MA 42.54 72.18 Temperate deciduous forest dominated by red oak,
red maple, black birch, white pine, and hemlock
Deciduous forests 2002?2005
Howland Forest (HF) ME 45.20 68.74 Boreal?northern hardwood transitional forest
consisting of hemlock-spruce-fir, aspen-birch, and
hemlock-hardwood mixtures
Evergreen forests 2000?2004 Hollinger et al.
(1999, 2004)
Howland Forest West Tower (HFW) ME 45.21 68.75 Deciduous needle forest, Boreal/northern hardwood
ecoton, old coniferous
Deciduous forests 2000?2004 Hollinger et al.
(1999, 2004)
Sylvania Wilderness Area (SWA) MI 46.24 89.35 Old-growth eastern hemlock/sugar maple/
basswood/yellow birch
Mixed forests 2002?2006 Desai et al. (2005)
Univ. of Mich. Biological Station (UMB) MI 45.56 84.71 Mid-aged conifer and deciduous, northern hardwood,
pine understory, aspen, mostly deciduous,
old-growth hemlock
Mixed forests 2000?2003 Gough et al. (2008)
Missouri Ozark (MO) MO 38.74 92.20 Oak hickory forest Deciduous forests 2004?2006 Gu et al. (2006, 2007)
Goodwin Creek (GC) MS 34.25 89.97 Temperate grassland Grasslands 2002?2006
Fort Peck (FPe) MT 48.31 105.10 Grassland Grasslands 2000?2006
Duke Forest Loblolly Pine (DFP) NC 35.98 79.09 Even-aged loblolly pine forest Evergreen forests 2001?2005 Oren et al. (1998, 2006)
Duke Forest Hardwood (DFH) NC 35.97 79.10 An uneven-aged closed-canopy stand in an oak-
hickory type forest composed of mixed hardwood
species with pine (Pinus taeda) as a minor component
Deciduous forests 2003?2005 Pataki and Oren (2003)
a
g
r
ic
u
l
t
u
r
a
l
a
n
d
f
o
r
e
s
t
m
e
t
e
o
r
o
l
o
g
y
1
4
8
(2
0
0
8
)
1
8
2
7
?
1
8
4
7
1830
North Carolina Loblolly Pine (NCP) NC 35.80 76.67 15-Year-old loblolly pine (Pinus taeda) plantation Evergreen forests 2005?2006 Noormets et al.
(unpublished)
Mead Irrigated Continuous (MIC) NE 41.17 96.48 Continuous maize Croplands 2001?2005 Verma et al. (2005)
Mead Irrigated Rotation (MIR) NE 41.16 96.47 Maize?soybean rotation Croplands 2001?2005 Verma et al. (2005)
Mead Rainfed (MR) NE 41.18 96.44 Maize?soybean rotation Croplands 2001?2005 Verma et al. (2005)
Bartlett Experimental Forest (BEF) NH 44.06 71.29 Temperate northern hardwood forest dominated
by American beech, red maple, paper birch,
and hemlock
Deciduous forests 2004?2005 Jenkins et al. (2007)
Toledo Oak Openings (TOP) OH 41.33 83.51 Oak Savannah dominated by Quercus rubra,
Quercus alba, and Acer rubrum
Savannas 2004?2005 Noormets et al. (2008b)
ARM Oklahoma (ARM) OK 36.61 97.49 Winter wheat, some pasture and summer crops Croplands 2003?2006
Metolius Intermediate (MI) OR 44.45 121.56 Intermediate-aged temperate coniferous forest
dominated by Pinus ponderosa, Purshia tridentate,
Arctostaphylos patula
Evergreen forests 2003?2005 Law et al. (2003)
and Irvine et al. (2007)
Metolius New (MN) OR 44.32 121.61 Young temperate coniferous forest dominated
by Pinus ponderosa and Purshia tridentata
Evergreen forests 2004?2005 Law et al. (2003)
and Irvine et al. (2007)
Brookings (Bro) SD 44.35 96.84 Temperate grassland Grasslands 2004?2006
Freeman Ranch Mesquite Juniper (FRM) TX 29.95 98.00 Grassland in transition to an Ashe juniper-
dominated woodland
Savannas 2004?2006
Wind River Crane Site (WRC) WA 45.82 121.95 Temperate coniferous forest dominated by
Douglas-fir and western hemlock
Evergreen forests 2000?2004 Falk et al. (2008)
Lost Creek (LC) WI 46.08 89.98 Alder-willow deciduous wetland Deciduous forests 2000?2005
Willow Creek (WC) WI 45.81 90.08 Temperate/Boreal forest dominated by white ash,
sugar maple, basswood, green ask, and red oak
Deciduous forests 2000?2006 Cook et al. (2004)
Wisconsin intermediate hardwood (WIH) WI 46.73 91.23 17-Year-old regeneration mixed northern
hardwood with bigtooth aspen (Populus
grandidentata) dominance
Deciduous forests 2003 Noormets et al. (2008a)
Mature Red Pine (MRP) WI 46.74 91.17 65-Year-old red pine (Pinus resinosa) plantation Evergreen forests 2002?2005 Noormets et al. (2007)
The units of latitude and longitude are decimal degrees.
a
g
r
ic
u
l
t
u
r
a
l
a
n
d
f
o
r
e
s
t
m
e
t
e
o
r
o
l
o
g
y
1
4
8
(2
0
0
8
)
1
8
2
7
?
1
8
4
7
1831
a g r i c u l t u r a l a n d f o r e s t m e t e o r o l o g y 1 4 8 ( 2 0 0 8 ) 1 8 2 7 ? 1 8 4 71832component of ecosystem respiration. Because autotrophic and
heterotrophic activity belowground is controlled by rooting
systems and substrate availability, soil respiration is strongly
linked to plant metabolism, photosynthesis and litterfall
(Ryan and Law, 2005). Ra can be empirically modeled as a
function of air temperature and tissue carbon (foliage, stem,
and roots), whereas Rh is often modeled as a function of
substrate availability, soil temperature and soil moisture
(Ryan and Law, 2005). At the stand or regional level, NEE is
significantly affected by disturbances from fire and harvest
(Thornton et al., 2002; Law et al., 2004) and fractional
vegetation cover (DeFries et al., 2002).
Many of these factors influencing NEE can be assessed by
satellite remote sensing. Optical remote sensing systems
measure the surface reflectance, the fraction of solar energy
that is reflected by the Earth?s surface. For a given wavelength,
different vegetation types and/or plant species may have
different reflectance (Schmidt and Skidmore, 2003). The
reflectance of the same vegetation type also depends on
wavelength region, biophysical properties (e.g., biomass, leaf
area, and stand age), soil moisture, and sun-object-sensor
geometry (Ranson et al., 1985; Penuelas et al., 1993). Therefore,
reflectance values from multiple spectral bands can provide
useful information for estimating NEE. Moreover, surface
reflectance can be used to develop vegetation indices and
biophysical parameters that can account for factors influen-
cing NEE, such as the normalized difference vegetation index
(NDVI), the enhanced vegetation index (EVI), the land surface
temperature (LST), the normalized difference water index
(NDWI), the fraction of photosynthetically active radiation
absorbed by vegetation canopies (fPAR), and LAI.
The NDVI captures the contrast between the visible-red
and near-infrared reflectance of vegetation canopies. It is
defined as
NDVI ? rnir  rred
rnir ? rred
(3)
where rred and rnir are the visible-red and near-infrared
reflectance, respectively. NDVI is closely correlated to the
fraction of photosynthetically active radiation (fPAR)
absorbed by vegetation canopies (Asrar et al., 1984; Law
and Waring, 1994) and photosynthetic activity (Xiao and
Moody, 2004). NDVI is also related to vegetation biomass
(Myneni et al., 2001) and fractional vegetation cover (Xiao
and Moody, 2005). However, NDVI has several limitations,
including saturation in a multilayer closed canopy and sen-
sitivity to both atmospheric aerosols and soil background
(Huete et al., 2002; Xiao and Moody, 2005). To account for
these limitations of NDVI, Huete et al. (1997) developed the
improved vegetation index, EVI:
EVI ? 2:5 rnir  rred
rnir ? ?6rred  7:5rblue? ? 1
(4)
where rnir, rred, and rblue are the spectral reflectance at the
near-infrared, red, and blue wavelengths, respectively. Huete
et al. (2002) has developed a global EVI product from MODIS
data for the period from 2000 to present.
The LST derived from MODIS is a measure of the soil
temperature at the surface. The MODIS LST agreed with in situ
measured LST within 1 K in the range 263?322 K (Wan et al.,2002). LST is likely a good indicator of Re as both Ra and RH are
significantly affected by air/surface temperature. Rahman
et al. (2005) demonstrated that satellite-derived LST was
strongly correlated with Re.
As the shortwave infrared (SWIR) spectral band is sensitive
to vegetation water content and soil moisture, a combination
of NIR and SWIR bands have been used to derive water-
sensitive vegetation indices (Ceccato et al., 2002). Gao (1996)
developed the NDWI from satellite data to measure vegetation
liquid water:
NDWI ? rnir  rswir
rnir ? rswir
(5)
where rswir is the reflectance at the SWIR spectral band. The
NDWI was shown to be strongly correlated with leaf water
content (equivalent water thickness (EWT), g H2O m
2) (Jack-
son et al., 2004) and soil moisture (Fensholt and Sandholt,
2003) over time. It was incorporated into the vegetation photo-
synthesis model (VPM) as a water scalar for estimating GPP
(Xiao et al., 2005). Yet, there is still a question as to whether
NDWI provides useful information on canopy water stress
status that affects photosynthesis because of its sensitivity
to the relatively small changes in relative water content
observed in natural vegetation, and inability to discern
changes in canopy biomass from changes in canopy moisture
status (Hunt and Rock, 1989; Gao, 1996).
Satellite data can also provide estimates for LAI and fPAR.
These two variables characterize vegetation canopy function-
ing and energy absorption capacity (Myneni et al., 2002), and
are key parameters in most ecosystem productivity and
biogeochemical models due to their high correlation with
GPP (Sellers et al., 1997).
We therefore selected surface reflectance, EVI, LST, NDWI,
fPAR, and LAI as explanatory variables. All these variables
were derived from MODIS data, which also avoided the
complications and difficulties to merge disparate data sources.
2.3. Data
We obtained the following three types of data: NEE from eddy
covariance flux towers, explanatory variables derived from
MODIS, and a land cover map derived from MODIS.
2.3.1. AmeriFlux data
The AmeriFlux network coordinates regional analysis of
observations from eddy covariance flux towers across North
America, Central America, and South America (Law, 2006). We
obtained the Level 4 NEE product for 42 AmeriFlux sites for the
period 2000?2006 from the AmeriFlux website (http://pub-
lic.ornl.gov/ameriflux/) (Table 1). These sites are distributed
across the conterminous U.S. (Fig. 1), and cover a range of
vegetation types including forests, shrublands, savannas,
grasslands, and croplands (Table 1). Moreover, the distribution
of these sites in the mean annual climate space (Fig. 2)
indicates that they cover typical U.S. climate types. In
addition, they also include some forest sites at different times
since stand replacing disturbance, which are located in
disturbance clusters of sites. We therefore believe that these
sites are fairly representative of U.S. ecosystem and climate
types.
a g r i c u l t u r a l a n d f o r e s t m e t e o r o l o g y 1 4 8 ( 2 0 0 8 ) 1 8 2 7 ? 1 8 4 7 1833The Level 4 product consists of two types of NEE data,
including standardized (NEE_st) and original (NEE_or) NEE
(AmeriFlux, 2007). NEE_st was calculated using CO2 flux
estimated by the eddy covariance method, which includes
summation with CO2 storage in the canopy air space that was
obtained from the discrete approach (single point on the top of
the tower) for all the sites, whereas NEE_or was calculated
using the storage obtained from within canopy CO2 profile
measurements in relatively tall forest canopies or from the
discrete approach. The average data coverage during a year is
only 65% due to system failures or data rejection, and
therefore robust and consistent gap filling methods are
required to provide complete data sets (Falge et al., 2001).
Both NEE_st and NEE_or were filled using the Marginal
Distribution Sampling (MDS) method (Reichstein et al., 2005)
and the Artificial Neural Network (ANN) method (Papale and
Valentini, 2003). The ANN method was generally, if only
slightly, superior to the MDS method (Moffat et al., 2007).
Therefore, we used the gap-filled NEE data based on the ANN
method. For each site, if the percentage of the remaining
missing values for NEE_st was lower than that for NEE_or, we
selected NEE_or; otherwise, we used NEE_st. In this product,
negative sign denotes carbon uptake, and positive sign
denotes carbon release.
The Level 4 product consists of NEE data with four different
time steps, including half-hourly, daily, weekly (8-day), and
Fig. 1 ? Location and spatial distribution of the AmeriFlux sites
land-cover map that was used for the continental-scale estimat
sites.monthly. We used 8-day NEE data (gC m2 day1) to match the
compositing intervals of MODIS data. Moreover, the average
NEE over such a period was shown to largely eliminate
micrometeorological sampling errors, with the remaining
spatial variability representing variation in ecosystem attri-
butes (Oren et al., 2006), here accounted for by data from
MODIS.
2.3.2. MODIS data
MODIS is a key instrument on board the NASA?s Terra and
Aqua satellites. The Terra MODIS and Aqua MODIS view the
entire Earth?s surface every 1?2 days, acquiring data in 36
spectral bands and with the spatial resolution of 250 m, 500 m,
and 1 km. We used the following four MODIS data products:
surface reflectance (MOD09A1; Vermote and Vermeulen,
1999), daytime and nighttime LST (MOD11A2; Wan et al.,
2002), EVI (MOD13A1; Huete et al., 2002), and LAI/fPAR
(MOD15A2; Myneni et al., 2002). Surface reflectance data
consist of reflectance values of seven spectral bands: blue
(459?479 nm), green (545?565 nm), red (620?670 nm), near-
infrared (841?875 nm and 1230?1250 nm), and shortwave
infrared (1628?1652 nm and 2105?2155 nm). Surface reflec-
tance and EVI are at a spatial resolution of 500 m, while LAI,
fPAR, and LAI are at spatial resolution of 1 km. Surface
reflectance, fPAR, and LAI are at a temporal resolution of 8
days, while EVI is at a temporal resolution of 16 days.
used in this study. The base map is the reclassified MODIS
ion of NEE. Symbols indicate the location of the AmeriFlux
step, we averaged the values of each variable using the pixels
with good quality within the area to represent the values at
the flux site. If none of the values within the 3 km  3 km
area was of good quality, we treated the period as missing.
Each 16-day EVI value was split into two 8-day values to
correspond with the compositing interval of other MODIS
data products.
For the continental-scale estimation of NEE, we obtained
continental-scale MODIS data including surface reflectance,
daytime and nighttime LST, and EVI from the Earth Observing
System (EOS) Data Gateway (http://edcimswww.cr.usgs.gov/
pub/imswelcome/). For each variable and for each 8- or 16-day
period, a total of 22 tiles were needed to cover the
conterminous U.S., and these tiles were mosaiced to generate
a g r i c u l t u r a l a n d f o r e s t m e t e o r o l o g y 1 4 8 ( 2 0 0 8 ) 1 8 2 7 ? 1 8 4 71834Fig. 2 ? Distribution of the 42 AmeriFlux sites in mean
annual climate space. Climate parameters are the mean
annual precipitation (x-axis) and mean annual
temperature (y-axis) taken over a 30-year period of record
(1971?2000) from the PRISM database (http://
www.prism.oregonstate.edu/). Gray points indicate the
climate space distribution of landmass within the
conterminous United States. The climate data have beenFor each AmeriFlux site, we obtained the MODIS ASCII
(American Standard Code for Information Interchange)
subsets (Collection 4) consisting of 7 km  7 km regions
centered on the flux tower from the Oak Ridge National
Laboratory?s Distributed Active Archive Center (ORNL DAAC,
2006). We extracted average values for the central
3 km  3 km area within the 7 km  7 km cutouts to better
represent the flux tower footprint (Schmid, 2002; Rahman
et al., 2005). For each variable, we determined the quality of
the value of each pixel within the area using the quality
assurance (QA) flags included in the product. At each time
resampled to 12 km resolution for plotting points in this
figure. Symbols show the location of each AmeriFlux site
in the climate space. The climate data of the sites are from
the AmeriFlux website (http://public.ornl.gov/ameriflux/)
and the PRISM database.
Table 2 ? The seven broader vegetation types used in the stud
Vegetation types UMD classes
Evergreen forests Evergreen needleleaf forests (1)
and evergreen broadleaf forests (2)
Deciduous forests Deciduous needleleaf forests (3)
and deciduous broadleaf forests (4)
Mixed forests Mixed forests (5)
Shrublands Closed shrublands (6) and open shrublands (7)
Savannas Woody savannas (8) and savannas (9)
Grasslands Grasslands (10)
Croplands Croplands (12)a continental-scale image. For each variable, we determined
the quality of the value of each pixel using the QA flags, and
replaced the bad value using a linear interpolation approach
(Zhao et al., 2005). The NDWI was calculated from band 2
(near-infrared) and band 6 (shortwave infrared) of the surface
reflectance product according to Eq. (5).
2.3.3. Land cover
To construct the predictive NEE model, we obtained the land
cover type for each AmeriFlux site based on the site
descriptions (Table 1), and categorized each site into a class
of the University of Maryland land-cover classification system
(UMD). Although the 42 AmeriFlux sites used in this study
cover a variety of vegetation classes of this classification
system, some classes (e.g., deciduous needleleaf forests and
open shrublands) were not covered by any site. We therefore
reclassified all vegetation classes of the UMD classification
system to seven broader classes (Table 2). Specifically, ever-
green needleleaf forests and evergreen broadleaf forests were
merged to evergreen forests, deciduous needleleaf forests and
deciduous broadleaf forests to deciduous forests, closed
shrublands and open shrublands to shrublands, and woody
savannas and savannas to savannas.
To estimate NEE for each 1 km  1 km cell at the con-
tinental scale, we obtained the land cover type for each cell
from the MODIS land cover map with the UMD classification
system (Friedl et al., 2002). Similarly, we reclassified the
vegetation types of the MODIS land cover map to the seven
broader classes (Table 2). The reclassified land-cover map is
shown in Fig. 1.
y and the corresponding UMD vegetation classes
Definition (Belward and Loveland, 1996)
Tree canopy cover >60% and tree height >2 m. Most of
the canopy remains green all year
Tree canopy cover >60% and tree height >2 m. Most of
the canopy is deciduous
Tree canopy cover >60% and tree height >2 m. Mixed
evergreen and deciduous canopy
Shrub canopy cover >10% (10?60% for open shrublands, >60%
for closed shrublands) and height <2 m
Forest canopy cover between 10?60% (30?60% for woody
savannas, 10?30% for savannas) and height >2 m
Herbaceous cover. Woody cover <10%
Temporary crops followed by harvest and a bare soil period
model that includes all the 14 explanatory variables, we also
Rule 2: if land cover in {deciduous forests, savannas},
B2 > 0.34, NDWI < = 0.36, Ld > 18.06, Ln > 11.13, then
NEE ? 5:94 ? 47:2B4  35B1  12:7B2  7B3  3:6NDWI
? 8:4B6 ? 4:4B5  0:4EVI
..
.
Rule 25: if land cover in {deciduous forests, mixed forests,
croplands}, NDWI > 0.02, Ln <= 9.68, then
NEE ? 0:40  37:6B4 ? 15:1B1 ? 8:9B2 ? 0:046Ln ? 0:9B5
? 0:4B3
Rule 26: if land cover in {deciduous forests, mixed forests,
croplands}, NDWI > 0.02, Ln > 9.68, then
NEE ? 2:86 ? 56:5B5  50:5B6 ? 14:9NDWI  2:9B1  0:5B4
 0:5B2
where B1?B6 are surface reflectance bands 1?6, Ld is daytime
e o r o l o g y 1 4 8 ( 2 0 0 8 ) 1 8 2 7 ? 1 8 4 7 1835developed a series of models by dropping one or more variables
at a time using Cubist. To select the best model, we evaluated
the performance of each model based on the average error,
relative error, and correlation coefficient. We chose the model
with the minimal average error and relative error and
maximum correlation coefficient as the best model. We also
evaluated model performance using scatterplots of predicted
versus observed NEE and seasonal variations between the
predicted and observed NEE.
2.5. Continental-scale estimation of NEE
As mentioned earlier, the AmeriFlux sites we selected are
fairly representative of the U.S. ecosystem and climate types.
We believe that the predictive NEE model constructed from the
42 sites can be extrapolated to the conterminous U.S. Thus, we
applied the predictive NEE model to estimate NEE for each
1 km  1 km cell across the conterminous U.S. for each 8-day
period in 2005 using wall-to-wall MODIS data. We then
examined the spatiotemporal patterns of our NEE estimates.
3. Results and discussion
3.1. Model development
The best model contained the following explanatory variables:
surface reflectance bands 1?6, EVI, daytime and nighttime LST,
and NDWI (relative error = 0.64, average error = 0.986, r = 0.73).
This model achieved slightly higher performance than the full
model (relative error = 0.66, average error = 1.01, r = 0.72). The
selected model consisted of five committee models, each of
which was made of a number of rule-based submodels. For
example, the first committee model was made of 26 rule-based
submodels:
Rule 1: if land cover = croplands, daytime LST > 30.07,
EVI > 0.40, then
NEE ? 20:24  430:3B3 ? 431:7B4 ? 80:8B1  108:7B52.4. Model development
We developed a predictive NEE model using Cubist based on the
site-specific MODIS and AmeriFlux NEE data. Our explanatory
variables included surface reflectance (7 bands), daytime and
nighttime LST, EVI, fPAR, and LAI, and our target variable was
NEE. We split the site level data set of AmeriFlux and MODIS
data into a training set (2000?2004) and a test set (2005?2006). If
a site only had NEE observations for the period 2000?2004,
the site was only included in the training set; if a site only had
NEE observations for the period 2005?2006, the site was only
included in the test set; otherwise, the site was included in both
training and test sets. The training and test sets included 40 and
34 AmeriFlux sites, respectively. Each set included sites at
different stages since stand-replacing disturbances. Altogether
we had a total of 4596 and 2257 data points for the training and
test sets, respectively. We trained the model with the training
set, and tested the model with the test set. In addition to the full
a g r i c u l t u r a l a n d f o r e s t m e t 23:4EVI ? 0:22Ld ? 11:4NDWI  27:6B6 ? 4B2LST, and Ln is nighttime LST. The model estimated NEE
reasonably well (r = 0.73, p < 0.001; Fig. 3) considering that
we used multiple years of data from a number of sites
involving a variety of vegetation types across the con-
terminous U.S. The model slightly underestimated positive
NEE values, and overestimated negative NEE values, where
negative values indicate carbon uptake, and positive values
indicate carbon release. In absolute magnitude, the model
slightly underestimated both carbon release and uptake
rates, thus damping the observed amplitude.
The analysis of NEE residuals (Fig. 4) indicated that the
residuals were not randomly distributed. In absolute magni-
tude, low NEE values were generally associated with low
prediction errors, whereas high NEE values were associated
with high prediction errors. This suggests that the uncertain-
ties of carbon flux measurements are directly proportional to
the magnitude of the fluxes (Richardson et al., 2008). InFig. 3 ? Scatterplot of observed 8-day NEE versus predicted
8-day NEE. The solid line is the 1:1 line.
addition, the explanatory variables included in the model
could not completely explain the variance of NEE. For
example, the independent variables used in the model could
not account for the sizes of soil organic carbon pools, thereby
Fig. 4 ? Scatterplot of predicted 8-day NEE versus residuals
(observed - predicted) over the period 2005?2006.
a g r i c u l t u r a l a n d f o r e s t m e t e o r o l o g y 1 4 8 ( 2 0 0 8 ) 1 8 2 7 ? 1 8 4 71836affecting the performance of the model for estimating NEE.
We calculated the average error and relative error across all
AmeriFlux sites for each 8-day period, and then plotted these
two types of error against time (Fig. 5). The average error
showed a strong seasonality. In absolute magnitude, winter
had low average errors (0.6 gC m2 day1), whereas warm
season errors often exceeded 1 gC m2 day1. This was not
surprising as in absolute magnitude, winter months generally
had relatively low NEE, while summer months had relatively
high NEE. Unlike the average error, the relative error did not
exhibit a strong seasonality, indicating that the model
performance did not substantially vary with season.Fig. 5 ? The average error (EA) and relative error (ER)
averaged across all AmeriFlux sites and over the period
2005?2006 for each 8-day period.We also compared our NEE estimates with observed NEE for
each AmeriFlux site (Fig. 6). The NEE estimates captured most
features of observed NEE such as seasonality and interannual
variability over the period 2005?2006. For some sites, episodes of
under- or over-prediction occurred. The model could not
capture exceptionally high and low NEE values that represented
large carbon release and uptake rates, respectively for some
sites, such as Audubon Research Ranch (AZ), FNAL (Fermi
National Accelerator Laboratory) Agricultural site (IL), Goodwin
Creek (MS), and Fort Peck (MT). In absolute magnitude, the
model substantially underestimated those exceptional values.
For example, the model estimates were far below the observed
NEE values that were greater than 2 gC m2 day1 at the
Goodwin Creek site (MS), and were far greater than the observed
NEE values that were below 3 gC m2 day1 at the Audubon
Research Ranch site (AZ). Overall, the model performed better
for deciduous forests, savannas, grasslands and croplands than
for evergreen forests and shrublands.
The disagreement between estimated and observed NEE
values is likely due to the following reasons. First, the MODIS
and tower footprints do not always match with each other. As
mentioned earlier, for each explanatory variable derived from
MODIS data, we used the values averaged within the
3 km  3 km area (i.e., MODIS footprint) surrounding each
flux tower to represent the values of the tower site. For most
sites, vegetation structure within the 3 km  3 km area
surrounding the flux tower is similar to that at the tower,
and therefore the MODIS footprint may match with the tower
footprint. However, some sites are located in complex land
mosaics, and the vegetation structure at the flux tower could
be significantly different from that within the MODIS
footprint. For example, the Tonzi Ranch site (CA) is dominated
by deciduous blue oaks (Quercus douglasii), and the understory
and open grassland are dominated by cool-season C3 annual
species (Ma et al., 2007). The MODIS footprint, however,
consists of a larger fraction of grassland. The phenologies of
blue oaks and grassland are distinct from each other (Ma et al.,
2007), and therefore these two plant species had differential
contributions to the NEE integrated over the MODIS footprint.
In the spring, wet conditions along with warm temperatures
facilitated the fast growth of grass, leading to large carbon
uptake rates within the MODIS footprint. As a result, in
absolute magnitude, our NEE estimates were higher than the
observed values at the tower site. Grasses senesced by the end
of the spring as the rainy season ended (Ma et al., 2007). The
senescence of grasses led to carbon release in the summer,
and thus lowered the carbon uptake rates integrated over the
MODIS footprint. Therefore, in absolute magnitude, our NEE
values were much lower than the observed values at the tower
in the summer.
Second, some sites experienced substantial disturbances
that alter ecosystem carbon fluxes. For example, the Austin
Cary site (FL) suffered from an extreme drought over the
period 1999?2002; a prescribed burn at the site in 2003 then
removed 95% of the understory vegetation; the site was also
hit by three hurricanes in 2004. These disturbances reduced
carbon uptake rates, whereas MODIS data are less sensitive to
changes in understory vegetation in forest ecosystems,
thereby leading to substantial overestimation of carbon
uptake rates.
Third, our model could not sufficiently account for the
factors influencing RH. As mentioned earlier, RH is influenced
by substrate availability, soil temperature, and soil moisture.
LST and NDWI can account for soil temperature and soil
moisture. However, surface reflectance can only partly
account for non-photosynthetic material (e.g., litter). Root
and associated mycorrhizal respiration produce roughly half
of soil respiration, with much of the remainder derived from
decomposition of recently produced root and leaf litter (Ryan
and Law, 2005). Changes in the carbon stored in the soil
generally contribute little to soil respiration, but these
changes, together with shifts in plant carbon allocation,
determine ecosystem carbon storage belowground and its
exchange with the atmosphere (Ryan and Law, 2005). The
eac
re
int
re
a g r i c u l t u r a l a n d f o r e s t m e t e o r o l o g y 1 4 8 ( 2 0 0 8 ) 1 8 2 7 ? 1 8 4 7 1837Fig. 6 ? Observed and predicted 8-day NEE (gC mS2 dayS1) for
with square symbols represents the observed values, and the
For x-axis, the starting dates (month/day) of every two 8-day
Dashed lines were used to separate 2005 from 2006. Site abb
Table 1. The vegetation type for each site is given in parenthes
forests (MF), shrublands (Sh), savannas (Sa), grasslands (Gr), anh AmeriFlux site over the period 2005?2006. The green line
d line with circle symbols represents the predicted values.
ervals were given in parentheses under interval numbers.
viations are used here, and their full names are given in
is: evergreen forests (EF), deciduous forests (DF), mixed
d croplands (Cr).
inability of our model to account for transient carbon pools
contributed to the uncertainties in the NEE estimates
(Richardson et al., 2007).
Finally, we estimated NEE for 8-day intervals, and therefore
our estimates could not capture the variability of NEE within
the intervals. The MODIS LST and EVI products were averaged
from the corresponding daily products over a period of 8 and
16 days, respectively (Huete et al., 2002; Wan et al., 2002). For
each period, only data with good quality were retained for
compositing, and thus the number of days actually used for
compositing is often lower than the total number of days over
the period. The compositing technique for the MODIS surface
reflectance product is based on the minimum-blue criterion
that selects the clearest conditions over the 8-day period
(Vermote and Vermeulen, 1999). Therefore, the 8- or 16-day
values do not always represent the average environmental
conditions and average fluxes over the 8- or 16-day period. For
example, each 16-day EVI composite was an average of daily
EVI over a period of 16 days. The number of acceptable pixels
over a 16-day compositing period is typically less than 10
(often less than 5) due to cloud contaminations and extreme
off-nadir sensor view angles (Huete et al., 2002). Sims et al.
(2005) suggested that midday GPP derived from daily satellite
snapshots of vegetation was highly correlated with 8-day
a g r i c u l t u r a l a n d f o r e s t m e t e o r o l o g y 1 4 8 ( 2 0 0 8 ) 1 8 2 7 ? 1 8 4 71838Fig. 6. (Continued)
a g r i c u l t u r a l a n d f o r e s t m e t e o r o l o g y 1 4 8 ( 2 0 0 8 ) 1 8 2 7 ? 1 8 4 7 1839mean GPP, and the inclusion of cloudy days within 8-day
intervals had less effect on GPP than expected. However,
increased diffuse radiation under cloudy conditions could
result in higher light use efficiency and higher GPP (Gu et al.,
2002). The compositing process may exclude high EVI values
that represented high fPAR, therefore leading to lower carbon
uptake estimates. On the other hand, the compositing process
may also exclude low EVI and LST values, thereby affecting
NEE estimates. Therefore, the exclusion of days with high and
low values could lead to underestimation and overestimation
of NEE values, respectively.
We averaged the estimated and observed 8-day NEE for
each AmeriFlux site and examined the relationship between
the estimated and observed mean 8-day NEE across the
sites (Fig. 7). The model estimated NEE reasonably well at
the site level (r = 0.86, p < 0.001). Overall, in absolute
magnitude, the model underestimated NEE. The perfor-
mance of the model also varied with site. On average, some
Fig. 6. (Contsites were carbon sources, whereas other sites were carbon
sinks. Large overestimation of carbon uptake occurred at
the Toledo Oak Openings site (OH), whereas large under-
estimation of carbon uptake occurred at Mature Red Pine
(WI), Duke Forest Loblolly Pine (NC), Duke Forest Hardwood
(NC), and North Carolina Loblolly Pine (NC). Large over-
estimation of carbon release occurred at Audubon Research
Ranch (AZ), ARM Oklahoma (OK), and Freeman Ranch
Mesquite (TX), whereas large underestimation of carbon
release occurred at Mead Irrigated (NE), Goodwin Creek (MS),
and Austin Cary (FL).
We also averaged our estimated and observed 8-day NEE
over all AmeriFlux sites for each vegetation type (i.e., biome),
and examined the relationship between estimated and
observed NEE across the vegetation types (Fig. 8). The model
predicted NEE at the biome level very well (r = 0.97, p < 0.001).
Again, in absolute magnitude, the model underestimated NEE.
The performance of the model also varied with vegetation
inued).
a g r i c u l t u r a l a n d f o r e s t m e t e o r o l o g y 1 4 8 ( 2 0 0 8 ) 1 8 2 7 ? 1 8 4 71840Fig. 7 ? Scatterplot of observed mean NEE versus predicted
mean NEE across the AmeriFlux sites. Error bars are
standard errors (defined as the standard deviation divided
by the square root of the number of observations) of the
observed and predicted 8-day mean NEE. Abbreviations of
AmeriFlux sites are given in Table 1. For a given site, the
mean NEE values did not necessarily represent the fluxestype. In absolute magnitude, large overestimation occurred for
evergreen forests and shrublands.
Our study demonstrated that MODIS data have great
potential for scaling up eddy flux NEE data to continental
scales across a variety of vegetation types. Unlike GPP
(Heinsch et al., 2006; Yang et al., 2007), NEE is much more
difficult to estimate because the transient carbon pools and
associated heterotrophic respiration are difficult to estimate
(Running et al., 2004; Mahadevan et al., 2008). The perfor-
mance of our model for estimating NEE is encouraging, given
the diversity in ecosystem types, age structures, fire and insect
disturbances, and management practices. In future research,
additional explanatory variables should be selected to better
account for live and dead vegetation carbon pools, and other
factors that influence decomposition of woody detritus and
soil respiration.
3.2. Continental-scale estimation of NEE
We estimated NEE for each 1 km  1 km cell across the
conterminous U.S. for each 8-day interval over the period
from January 2005 to February 2006. Fig. 9 shows examples of
8-day NEE maps that we produced for the conterminous U.S.
from January through December in 2005. For each month, the
second 8-day NEE map was shown here. The predictive model
trained at the AmeriFlux sites generally captured the
expected spatiotemporal patterns of NEE. The majority of
the conterminous U.S. released carbon or were nearly carbon
averaged over the entire 2-year period 2005?2006. The
temporal coverage of NEE data for each site is provided in
Fig. 6.Fig. 8 ? Scatterplot of observed mean NEE versus predicted
mean NEE across biomes: evergreen forests (EF),
deciduous forests (DF), mixed forests (MF), shrublands
(Sh), savannas (Sa), grasslands (Gr), and croplands (Cr).
Error bars are standard errors (defined as the standard
deviation divided by the square root of the number of
observations) of the observed and predicted 8-day mean
NEE.neutral in winter months (December?February) because at
this time of the year the canopies of most ecosystems were
dormant; in summer months (June?August), ecosystems in
the East assimilated carbon from the atmosphere, whereas
many areas in the West released carbon, possibly due to
summer drought effects on NEE. In fall months (September?
November), ecosystems in the East assimilated less carbon
than in the summer months as vegetation began to senesce
and days became shorter. Some ecosystems, particularly
evergreen forests in the Pacific Northwest and California,
assimilated carbon from the atmosphere throughout the
year. Douglas-fir, a major species in the Pacific Northwest and
California, is known to be highly plastic and able to
photosynthesize in winter when temperatures are above
freezing.
We aggregated 8-day NEE estimates for each season in 2005
(Fig. 10). Our NEE estimates exhibited strong seasonal
fluctuations, agreeing with previous studies (e.g., Falge
et al., 2002). Our NEE estimates also varied substantially over
space. In the spring (March?May), many areas in the eastern
half of the conterminous U.S. including the Southeast and the
Gulf Coast assimilated carbon from the atmosphere. The
growing season of these ecosystems started in the mid- to
late-spring, and GPP quickly exceeded Re, leading to net carbon
uptake in the entire season. By contrast, the Upper Great Lakes
region, the northern Great Plains, and the New England region
released carbon. The Upper Great Lakes region and the
northern Great Plains are dominated by croplands with most
crops planted between April and June (corn planted between
April and mid-May; soybeans between mid-May and mid-June;
and sorghum between late May and late June; Shroyer et al.,
a g r i c u l t u r a l a n d f o r e s t m e t e o r o l o g y 1 4 8 ( 2 0 0 8 ) 1 8 2 7 ? 1 8 4 7 18411996). Crops were sparse in the beginning of the growing
season and Re exceeded GPP, thereby leading to carbon
releases. The New England region and the northern portion
of the Upper Great Lakes region are dominated by temperate-
boreal transitional forests, and their relatively late greenup
due to low air temperatures led to carbon releases in the
spring. Many regions in the western half of the conterminous
U.S. also released carbon in the spring because of the sparse
vegetation and the dominance of Re over GPP. The Pacific Coast
assimilated carbon even in the spring because the dominant
evergreen forests in the region assimilated carbon due to mild
temperatures and moist conditions (Anthoni et al., 2002). The
Mediterranean regions in California also assimilated carbon in
the spring. The Mediterranean climate is characterized by
mild winter temperatures concomitant with the rainy season
as opposed to severe summer droughts and heat (Barbour and
Minnich, 2000). These ecosystems assimilated carbon because
of precipitation surplus and relatively warm temperatures in
the spring (Xu and Baldocchi, 2004; Ma et al., 2007).
In the summer months (June?August), the eastern half of
the conterminous U.S. assimilated carbon because GPP far
Fig. 9 ? Examples of predicted 8-day NEE for the conterminous U.
the second 8-day NEE map is shown here. The units are gC mS2 d
values indicate carbon uptake.exceeded Re owing to favorable temperature and soil moisture
conditions. By contrast, a vast majority of the land across the
western counterpart released carbon, including the Great
Basin, the Colorado Plateau, and the western Great Plains. The
2005 summer drought affected these regions (National
Climatic Data Center, 2008) and reduced GPP, whereas the
high temperatures increased Re, leading to net carbon
releases. Some other regions in the West also assimilated
carbon, including the northern Rocky Mountains and the
Pacific Coast. Some Mediterranean ecosystems in California
also released carbon due to precipitation deficits in the
summer.
In the fall (September?November), the Southeast and the
Gulf Coast still assimilated carbon, but net carbon uptake
rates substantially decreased relative to those in the summer.
This is because vegetation began to senesce in these regions in
the fall. The Upper Great Lakes region and the Great Plains
largely released carbon due to the harvesting of crops. The
majority of the land across the West including the Great
Plains, the Great Basin, and the Colorado Plateau released
carbon. The northern Pacific Coast, however, still absorbed
S. from January through December in 2005. For each month,
ayS1. Positive values indicate carbon release, and negative
arc
m
a g r i c u l t u r a l a n d f o r e s t m e t e o r o l o g y 1 4 8 ( 2 0 0 8 ) 1 8 2 7 ? 1 8 4 71842Fig. 10 ? Predicted NEE for each season in 2005: (a) spring (M
November); (d) winter (December?February). The units are gCcarbon. The Mediterranean ecosystems in California contin-
ued releasing carbon as the dry season spanned into the fall.
In the winter (December?February), the vast majority of the
conterminous U.S. released carbon as the canopies of most
ecosystems were dormant at this season of the year. Some
regions in the Pacific Coast, however, assimilated carbon even
in the winter because of the dominance of evergreen forests
and mild temperatures in the regions (Waring and Franklin,
1979). This agreed with the finding of Anthoni et al. (2002) that
old-growth ponderosa pine in Oregon slightly assimilated
carbon in the winter season. For the Mediterranean ecosys-
tems in California, a smaller part of the region released carbon
into the atmosphere relative to the fall as the wet season
started in the winter.
The trajectory of the mean 8-day NEE ( gC m2 day1) for
each vegetation type over the entire conterminous U.S.
throughout 2005 (Fig. 11a) showed that deciduous forests,
croplands, savannas, and mixed forests had large intra-
annual variability in NEE, whereas evergreen forests, grass-
lands, and shrublands exhibited much less variability. The
seasonal patterns of NEE were determined by the seasonal
differences in LAI, physiological capacity, meteorological
conditions, growing season length, soil temperature, moist-
ure status, and management practices (Falge et al., 2002). In
the late fall, winter, and early spring, on average, the U.S.
terrestrial ecosystems released carbon; taken separately,
only evergreen forests and grasslands assimilated carbon
during this time. Among vegetation types exhibiting positive
NEE values, deciduous forests had the highest values,
negative values indicate carbon uptake.h?May); (b) summer (June?August); (c) fall (September?
S2 seasonS1. Positive values indicate carbon release, andfollowed by mixed forests; croplands exhibited intermediate
values; shrublands and savannas exhibited lowest values
while evergreen forests still assimilated carbon. During the
growing season, on average, the U.S. terrestrial ecosystems
strongly assimilated carbon; taken separately, only shrub-
lands released carbon because of high temperatures and low
soil moisture conditions. Among vegetation types assimilat-
ing carbon, the highest absorption rates occurred for
deciduous forests, followed by croplands, savannas, and
mixed forests; intermediate rates occurred for evergreen
forests; the lowest rates occurred for grasslands. Baldocchi
et al. (2001) showed that the net CO2 exchange of temperate
deciduous forests increases by about 5.7 gC m2 day1 for each
additional day that the growing season, defined as the period
over which mean daily CO2 exchange is negative due to net
uptake by ecosystems, is extended. We found that on average,
the CO2 exchange of deciduous and evergreen forests across
the conterminous U.S. increased 3.6 and 1.2 gC m2 day1 for
each additional day that the growing season is extended,
respectively. Our continental-scale estimate for deciduous
forests was 37% lower than that estimated by Baldocchi et al.
(2001) likely because our estimate was based on all the areas
covered by deciduous forests encompassing the full range of
productivity.
The trajectory of the total 8-day NEE (TgC day1) over the
conterminous U.S. (Fig. 11b) showed clear dependence on
vegetation type. The differences in the trajectories of total 8-
day NEE among vegetation types were different from those of
mean 8-day NEE because of the differences in the areas among
a g r i c u l t u r a l a n d f o r e s t m e t e o r o l o g y 1 4 8 ( 2 0 0 8 ) 1 8 2 7 ? 1 8 4 7 1843vegetation types (Fig. 11b). In the late fall, winter, and early
spring, the U.S. terrestrial ecosystems released carbon (1?
2 TgC day1). Taken separately, croplands, deciduous forests,
and mixed forests released carbon, whereas evergreen forests
assimilated carbon; shrublands, savannas, and grasslands,
however, were nearly carbon neutral. During the growing
season, the U.S. terrestrial ecosystems assimilated carbon, with
peak total NEE of 17 TgC day1. All vegetation types except
shrublands assimilated carbon. In absolute magnitudes, the
highest total NEE (10 TgC day1) occurred for croplands; the
intermediate values occurred for deciduous forests, savannas,
and mixed forests; the lowest values occurred for evergreen
forests and grasslands. By contrast, shrublands released
carbon. Total 8-day NEE also showed largest intra-annual
variability for croplands, intermediate variability for deciduous
of the AmeriFlux network will continue to improve our
Fig. 11 ? Predicted mean and total 8-day NEE for each
vegetation type in the conterminous U.S. in 2005. (a) Mean
8-day NEE (gC mS2 dayS1); (b) total 8-day NEE (TgC dayS1).
Inset in plot (b) indicates the area (106 km2) of each
vegetation type: evergreen forests (EF), deciduous forests
(DF), mixed forests (MF), shrublands (Sh), savannas (Sa),
grasslands (Gr), and croplands (Cr). For x-axis, the starting
dates (month/day) of every two 8-day intervals were given
in parentheses under interval numbers.understanding of the impacts of climate variability, distur-
bances, and management practices on terrestrial carbonforests, savannas, and mixed forests, and lowest variability for
evergreen forests, grasslands, and shrublands.
4. Summary and conclusions
We combined MODIS and NEE data from 42 AmeriFlux sites
involving a variety of vegetation types to develop a predictive
NEE model using a modified regression tree approach. The
model was trained and validated using eddy flux NEE data over
the periods 2000?2004 and 2005?2006, respectively. The model
estimated NEE well at the site level. We then applied the model
to estimate NEE for each 1 km  1 km cell across the
conterminous U.S. for each 8-day period in 2005. The model
generally captured the expected spatial and seasonal patterns
of NEE. Our study demonstrated that our empirical approach
along with MODIS data have great potential for scaling up
AmeriFlux NEE measurements to the continental scale. Our
approach can be applied to the entire North America, other
geographical regions including Europe, Southeast Asia, and
South America, or to the globe scale, and to produce
continuous NEE estimates over regions, continents, or the
globe. This approach can also be used to scale other fluxes
including GPP and evapotranspiration to large areas.
Our wall-to-wall NEE estimates across the conterminous
U.S. provided an independent dataset from simulations by
biogeochemical modeling and inverse modeling for examining
the spatiotemporal patterns of NEE and constraining U.S.
terrestrial carbon sinks/sources. Our estimates have advan-
tages over these model simulations by taking advantage of NEE
measurements from a number of AmeriFlux sites involving
representative U.S. ecosystems. Moreover, our scaling-up
approach implicitly considered the effects of climate variability
and extreme climate events. Although our NEE estimates could
not capture the immediate emissions of CO2 due to the burning
of biomass in wildfires, our estimates could partly account for
the carbon fluxes following the disturbances because the
MODIS data we used provide real-time observations of
ecosystems. Compared to inverse modeling techniques, our
approach provided estimates at high spatial (1 km  1 km) and
temporal (8-day) resolutions. In addition, NEE is notoriously
difficult to quantify over large areas (Running et al., 2004), and
the accuracy of simulated NEE for regions and continents by
biogeochemical models is poorly known due to lack of spatially
explicit, independent validation data sets. Our estimates may
also provide an independent validation data set for these model
simulations. We will extend our NEE estimates to the entire
MODIS era (2000-present) for the conterminous U.S., which will
provide a valuable dataset for examining the interannual
variability of the U.S. terrestrial carbon uptake.
The AmeriFlux sites provide valuable measurements of
ecosystem carbon exchange for examining terrestrial carbon
dynamics (Law, 2006, 2007). Our study demonstrated that the
AmeriFlux measurements could be used to examine con-
tinental-scale carbon dynamics, and the continuing operationcycling. Our study also suggested that the current AmeriFlux
network should be augmented by establishing more sites for
Center of the National Institute for Global Environmental
a g r i c u l t u r a l a n d f o r e s t m e t e o r o l o g y 1 4 8 ( 2 0 0 8 ) 1 8 2 7 ? 1 8 4 71844Change under Cooperative Agreements No. DE-FC03-
90ER610100 for funding P.S. Curtis. We thank the principal
investigators of the MODIS data products including A.R.
Huete, Z. Wan, R.B. Myneni, and E.F. Vermote and other
contributors as well as the Oak Ridge National Laboratory
(ORNL), Distributed Active Archive Center (DAAC), and the
Earth Observing System (EOS) Data Gateway for making these
products available. We also thank T.A. Boden at the Carbon
Dioxide Information Analysis Center, ORNL, S.K.S. Vannan at
the ORNL DACC, M. Zhao at the University of Montana, and Z.
Wan at the University of California, Santa Barbara, for helpful
discussion about AmeriFlux data, MODIS ASCII subsets,
MODIS Quality Assurance (QA) flags, and MODIS LST,
respectively. The PRISM climate database was provided by
the PRISM Group, Oregon State University (http://
www.prismclimate.org). Computing support was provided
by the Rosen Center for Advanced Computing, Purdue
University.
r e f e r e n c e s
AmeriFlux, 2007. AmeriFlux Network http://public.ornl.gov/
ameriflux/ (visited on 10/04/07).
Anthoni, P.M., Unsworth, M.H., Law, B.E., Irvine, J., Baldocchi,
D.D., Tuyl, S.V., Moore, D., 2002. Seasonal differences in
carbon and water vapor exchange in young and old-growth
ponderosa pine ecosystems. Agric. For. Meteorol. 111,
203?222.
Asrar, G., Fuchs, M., Kanemasu, E.T., Hatfield, J.L., 1984.
Estimating of absorbed photosynthesis radiation and leaf
area index from spectral reflectance in wheat. Agron. J. 76,
300?306.
Baldocchi, D., Falge, E., Gu, L., Olson, R., Hollinger, D., Running,
S., Anthoni, P., Bernhofer, C., Davis, K., Evans, R., Fuentes, J.,
Goldstein, A., Katul, G., Law, B., Lee, X., 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, watercertain biomes in the UMD classification system (Table 1),
including open shrublands, savannas, grasslands, and crop-
lands. The augmentation should enable the differentiation of
open shrublands from closed shrublands, woody savannas
from savannas, and C3 from C4 plants in scaling-up studies,
thereby improving the estimation of carbon fluxes over large
areas.
Acknowledgements
The work of J. Xiao and Q. Zhuang was partly funded by the
National Science Foundation (NSF) Carbon and Water
Program (EAR-0630319). We thank the U.S. Department of
Energy Biological and Environmental Research, Terrestrial
Carbon Program (Award #DE-FG02-04ER63917) for funding Dr.
B.E. Law to develop the AmeriFlux network protocols and
database design for the network, and the Office of Science,
U.S. Department of Energy, through the Midwestern Regionalvapor, and energy flux densities. Bull. Am. Meteorol. Soc. 82,
2415?2434.Barbour, M., Minnich, B., 2000. California upland forests and
woodlands. In: Barbour, M., Billings, W. (Eds.), North
American Terrestrial Vegetation. 2nd ed. Cambridge
University Press, Cambridge, UK, pp. 161?202.
Belward, A., Loveland, T. 1996. The DIS 1 km land cover data set.
Global Change, The IGBP Newsletter, 27.
Ceccato, P., Gobron, N., Flasse, S., Pinty, B., Tarantola, S., 2002.
Designing a spectral index to estimate vegetation water
content from remote sensing data. Part 1. Theoretical
approach. Remote Sens. Environ. 82, 188?197.
Clark, K.L., Gholz, H.L., Moncrieff, J.B., Cropley, F., Loescher,
H.W., 1999. Environmental controls over net carbon dioxide
from contrasting Florida ecosystems. Ecol. Appl. 9, 936?948.
Clark, K.L., Gholz, H.L., Castro, M.S., 2004. Carbon dynamics
along a chronosequence of slash pine plantations in N.
Florida. Ecol. Appl. 4, 1154?1171.
Cook, B.D., Davis, K.J., Wang, W., Desai, A., Berger, B.W., Teclaw,
R.M., Martin, J.G., Bolstad, P.V., Bakwin, P.S., Yi, C., Heilman,
W., 2004. Carbon exchange and venting anomalies in an
upland deciduous forest in northern Wisconsin, USA. Agric.
For. Meteorol. 126, 271?295.
DeFries, R.S., Houghton, R.A., Hansen, M.C., Field, C.B., Skole, D.,
Townshend, J., 2002. Carbon emissions from tropical
deforestation and regrowth based on satellite observations
for the 1980s and 1990s. PNAS 99, 14256?14261.
Deng, F., Chen, J.M., Ishizawa, M., Yuen, C.-W., Mo, G., Higuchi,
K., Chan, D., Maksyutov, S., 2007. Global monthly CO2 flux
inversion with a focus over North America. Tellus 59B,
179?190.
Denning, A.S., Gollatz, J.G., Zhang, C., Randall, D.A., Berry, J.A.,
Sellers, P.J., Colello, G.D., Dazlich, D.A., 1996. Simulations of
terrestrial carbon metabolism and atmospheric CO2 in a
general circulation model. Part 1. Surface carbon fluxes.
Tellus 48B, 521?542.
Desai, A.R., Bolstad, P.V., Cook, B.D., Davis, K.J., Carey, E.V., 2005.
Comparing net ecosystem exchange of carbon dioxide
between an old-growth and mature forest in the upper
Midwest, USA. Agric. For. Meteorol. 128, 33?55.
Dore, S., Hymus, G.J., Johnson, D.P., Hinkle, C.R., Valentini, R.,
Drake, B.G., 2003. Cross validation of open-top chamber and
eddy covariance measurements of ecosystem CO2 exchange
in a Florida scrub-oak ecosystem. Global Change Biol. 9,
84?95.
Falge, E., Baldocchi, D., Olson, R., Anthoni, P., Aubinet, M.,
Bernhofer, 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.
Falge, E., Baldocchi, D., Tenhunen, J., Aubinet, M., Bakwin, P.,
Berbigier, P., Bernhofer, C., Burba, G., Clement, R., Davis,
K.J., Elbers, J.A., Goldstein, A.H., Grelle, A., Granier, A.,
Gu?mundsson, J., Hollinger, D., Kowalski, A.S., Katul, G.,
Law, B.E., Malhi, Y., Meyers, T., Monson, R.K., Munger, J.W.,
Oechel, W., Paw U, K.T., Pilegaard, K., Rannik, U., Rebmann,
C., Suyker, A., Valentini, R., Wilson, K., Wofsy, S., 2002.
Seasonality of ecosystem respiration and gross primary
production as derived from FLUXNET measurements. Agric.
For. Meteorol. 113, 53?74.
Falk, M., Wharton, S., Schroeder, M., Ustin, S., Paw U, K.T., 2008.
Flux partitioning in an old-growth forest: seasonal and
interannual dynamics. Tree Physiol. 28, 509?520.
Fan, S.M., Gloor, M., Mahlman, J., Pacala, S., Sarmiento, J.,
Takahashi, T., Tans, P., 1998. A large terrestrial carbon sink
in North America implied by atmospheric and oceanic
carbon dioxide data and models. Science 282, 442?446.
a g r i c u l t u r a l a n d f o r e s t m e t e o r o l o g y 1 4 8 ( 2 0 0 8 ) 1 8 2 7 ? 1 8 4 7 1845Fensholt, R., Sandholt, I., 2003. Derivation of a shortwave
infrared water stress index from MODIS near- and
shortwave infrared data in a semiarid environment. Remote
Sens. Environ. 87, 111?121.
Field, C.B., Randerson, J.T., Malmstrom, C.M., 1995. Global net
primary production: combining ecology and remote
sensing. Remote Sens. Environ. 51, 74?88.
FLUXNET, 2008. FLUXNET Project. http://daac.ornl.gov/
FLUXNET (visited on 01/02/08).
Friedl, M.A., McIver, D.K., Hodges, J.C.F., Zhang, X.Y., Muchoney,
D., Strahler, A.H., Woodcock, C.E., Gopal, S., Schneider, A.,
Cooper, A., Baccini, A., Gao, F., Schaaf, C., 2002. Global land
cover mapping from MODIS: algorithms and early results.
Remote Sens. Environ. 83, 287?302.
Gao, B.C., 1996. NDWI?a normalized difference water index for
remote sensing of vegetation liquid water from space.
Remote Sens. Environ. 58, 257?266.
Go?ckede, M., Foken, T., Aubinet, M., Aurela, M., Banza, J.,
Bernhofer, C., Bonnefond, J.M., Brunet, Y., Carrara, A.,
Clement, R., Dellwik, E., Elbers, J., Eugster, W., Fuhrer, J.,
Granier, A., Gru?nwald, T., Heinesch, B., Janssens, I.A.,
Knohl, A., Koeble, R., Laurila, T., Longdoz, B., Manca, G.,
Marek, M., Markkanen, T., Mateus, J., Matteucci, G., Mauder,
M., Migliavacca, M., Minerbi, S., Moncrieff, J., Montagnani,
L., Moors, E., Ourcival, J.-M., Papale, D., Pereira, J., Pilegaard,
K., Pita, G., Rambal, S., Rebmann, C., Rodrigues, A.,
Rotenberg, E., Sanz, M.J., Sedlak, P., Seufert, G., Siebicke, L.,
Soussana, J.F., Valentini, R., Vesala, T., Verbeeck, H., Yakir,
D., 2008. Quality control of CarboEurope flux data. Part 1.
Coupling footprint analyses with flux data quality
assessment to evaluate sites in forest ecosystems.
Biogeosciences 5, 433?450.
Goodale, C.L., Apps, M.J., Birdsey, R.A., Field, C.B., Heath, L.S.,
Houghton, R.A., Jenkins, J.C., Kohlmaier, G.H., Kurz, W.A.,
Liu, S., Nabuurs, G.-J., Nilsson, S., Shvidenko, A.Z., 2002.
Forest carbon sinks in the Northern Hemisphere. Ecol. Appl.
12, 891?899.
Gough, C.M., Vogel, C.S., Schmid, H.P., Su, H.-B., Curtis, P.S.,
2008. Multi-year convergence of biometric and
meteorological estimates of forest carbon storage. Agric.
For. Meteorol. 148, 158?170.
Gu, L., Baldocchi, D., Verma, S.B., Black, T.A., Vesala, T., Falge,
E.M., Dowty, P.R., 2002. Advantages of diffuse radiation for
terrestrial ecosystem productivity. J. Geophys. Res. 107,
4050, doi:10.1029/2001JD001242.
Gu, L., Meyers, T., Pallardy, S.G., Hanson, P.J., Yang, B., Heuer,
M., Hosman, K.P., Liu, Q., Riggs, J.S., Sluss, D., Wullschleger,
S.D., 2007. Influences of biomass heat and biochemical
energy storages on the land surface fluxes and radiative
temperature. J. Geophys. Res. D02107 , doi:10.1029/
2006JD007425.
Gu, L., Meyers, T., Pallardy, S.G., Hanson, P.J., Yang, B., Heuer,
M., Hosman, K.P., Riggs, J.S., Sluss, D., Wullschleger, S.D.,
2006. Direct and indirect effects of atmospheric conditions
and soil moisture on surface energy partitioning revealed by
a prolonged drought at a temperate forest site. J. Geophys.
Res. D16102 , doi:10.1029/2006JD007161.
Gurney, K.R., Law, R.M., Denning, A.S., Rayner, P.J., Baker, D.,
Bousquet, P., Bruhwiler, L., Chen, Y.-H., Ciasi, P., Fan, S.,
Fung, I.Y., Gloor, M., Heimann, M., Higuchi, K., John, J., Maki,
T., Maksyutov, S., Masarie, K., Peylin, P., Prather, M., Pak,
B.C., Randerson, J., Sarmiento, J., Taguchi, S., Takahashi, T.,
Yuen, C.-W., 2002. Towards robust regional estimates of
CO2 sources and sinks using atmospheric transport models.
Nature 415, 626?630.
Heinsch, F.A., Zhao, M., Running, S.W., Kimball, J.S., Nemani,
R.R., Davis, K.J., Bolstad, P.V., Cook, B.D., Desai, A.R.,
Ricciuto, D.M., Law, B.E., Oechel, W.C., Kwon, H.J., Luo, H.,
Wofsy, S.C., Dunn, A.L., Munger, J.W., Baldocchi, D.D., Xu,L., Hollinger, D.Y., Richardson, A.D., Stoy, P.C., Siqueira,
M.B.S., Monson, R.K., Burns, S.P., Flanagan, L.B., 2006.
Evaluation of remote sensing based terrestrial productivity
from MODIS using regional tower eddy flux network
observations. IEEE Trans. Geosci. Remote Sens. 44,
1908?1925.
Hollinger, D.Y., Goltz, S.M., Davidson, E.A., Lee, J.T., Tu, K.,
Valentine, H.T., 1999. Seasonal patterns and environmental
control of carbon dioxide and water vapor exchange in an
ecotonal boreal forest. Global Change Biol. 5, 891?902.
Hollinger, D.Y., Aber, J., Dail, B., Davidson, E.A., Goltz, S.M.,
Hughes, H., Leclerc, M.Y., Lee, J.T., Richardson, A.D.,
Rodrigues, C., Scott, N.A., Achuatavarier, D., Walsh, J., 2004.
Spatial and temporal variability in forest-atmosphere CO2
exchange. Global Change Biol. 10, 1689?1706.
Hollinger, S.E., Bernacchi, C.J., Meyers, T.P., 2005. Carbon budget
of mature no-till ecosystem in North Central Region of the
United States. Agric. For. Meteorol. 130, 59?69.
Huang, C., Townshend, J.R.G., 2003. A stepwise regression tree
for nonlinear approximation: applications to estimating
subpixel land cover. Int. J. Remote Sens. 24, 75?90.
Huete, A.R., Liu, H.Q., Batchily, K., vanLeeuwen, W., 1997. A
comparison of vegetation indices global set of TM images
for EOS-MODIS. Remote Sens. Environ. 59, 440?451.
Huete, A., Didan, K., Miura, T., Rodriguez, E.P., Gao, X., Ferreira,
L.G., 2002. Overview of the radiometric and biophysical
performance of the MODIS vegetation indices. Remote Sens.
Environ. 83, 195?213.
Hunt, E.R., Rock, B.N., 1989. Detection of changes in leaf water
content using near- and middle-infrared reflectances.
Remote Sens. Environ. 30, 43?54.
Irvine, J., Law, B.E., Hibbard, K.A., 2007. Postfire carbon pools and
fluxes in semiarid ponderosa pine in Central Oregon. Global
Change Biol. 13, 1748?1760.
Jackson, T.J., Chen, D., Cosh, M., Li, F., Anderson, M., Walthall,
C., Doriaswamy, P., Hunt, E.R., 2004. Vegetation water
content mapping using Landsat data derived normalized
difference water index for corn and soybeans. Remote Sens.
Environ. 92, 475?482.
Jenkins, J.P., Richardson, A.D., Braswell, B.H., Ollinger, S.V.,
Hollinger, D.Y., Smith, M.-L., 2007. Refining light-use
efficiency calculations for a deciduous forest canopy using
simultaneous tower-based carbon flux and radiometric
measurements. Agric. For. Meteorol. 143, 64?79.
Janssens, I.A., Freibauer, A., Ciais, P., Smith, P., Jabuurs, G.-J.,
et al., 2003. Europe?s terrestrial biosphere absorbs 7 to 12%
of European anthropogenic CO2 emissions. Science 300,
1538?1542.
Law, B.E., 2006. Carbon dynamics in response to climate and
disturbance: recent progress from multiscale measurements
and modeling in AmeriFlux. In: Omasa, K., Nouchi, I., De Kok,
L.J. (Eds.),Plant Responses to Air Pollution and Global Change.
Springer, Tokyo, Japan, pp. 205?213.
Law, B.E., 2007. AmeriFlux network aids global synthesis. EOS
Trans. 88, 286.
Law, B.E., Ryan, M.G., Anthoni, P.M., 1999. Seasonal and annual
respiration of a ponderosa pine ecosystem. Global Change
Biol. 5, 169?182.
Law, B.E., Sun, O.J., Campbell, J., Tuyl, S.V., Thornton, P.E., 2003.
Changes in carbon storage and fluxes in a chronosequence
of ponderosa pine. Global Change Biol. 9, 510?524.
Law, B.E., Turner, D., Campbell, J., Sun, O.J., Van Tuyl, S., Ritts,
W.D., Cohen, W.B., 2004. Disturbance and climate effects on
carbon stocks and fluxes across Western Oregon, USA.
Global Change Biol. 10, 1429?1444.
Law, B.E., Turner, D., Lefsky, M., Campbell, J., Guzy, M., Sun, O.,
Van Tuyl, S., Cohen, W., 2006. Carbon fluxes across regions:
observational constraints at multiple scales. In: Wu, J., Jones,
B., Li, H., Loucks, O. (Eds.), Scaling and Uncertainty Analysis
a g r i c u l t u r a l a n d f o r e s t m e t e o r o l o g y 1 4 8 ( 2 0 0 8 ) 1 8 2 7 ? 1 8 4 71846in Ecology: Methods and Applications. Springer, USA, pp.
167?190.
Law, B.E., Waring, R.H., 1994. Remote sensing of leaf area index
and radiation intercepted by understory vegetation. Ecol.
Appl. 4, 272?279.
Lipson, D.A., Wilson, R.F., Oechel, W.C., 2005. Effects of elevated
atmospheric CO2 on soil microbial biomass, activity, and
diversity in a chaparral ecosystem. Appl. Environ. Microbiol.
71, 8573?8580.
Ma, S., Baldocchi, D.D., Xu, L., Hehn, T., 2007. Inter-annual
variability in carbon dioxide exchange of an oak/grass
savanna and open grassland in California. Agric. For.
Meteorol. 147, 157?171.
Mahadevan, P., Wofsy, S.C., Matross, D.M., Xiao, X., Dunn, A.L.,
Lin, J.C., Gerbig, C., Munger, J.W., Chow, V.Y., Gottlieb, E.W.,
2008. A satellite-based biosphere parameterization for net
ecosystem CO2 exchange: vegetation photosynthesis and
respiration model (VPRM). Global Biogeochem. Cyc. 22,
GB2005.
Monson, R.K., Turnipseed, A.A., Sparks, J.P., Harley, P.C., Scott-
Denton, L.E., Sparks, K., Huxman, T.E., 2002. Carbon
sequestration in a high-elevation, subalpine forest. Global
Change Biol. 8, 459?478.
Moffat, A.M., Papale, D., Reichstein, M., Hollinger, D.Y.,
Richardson, A.D., Barr, A.G., Beckstein, C., Braswell, B.H.,
Churkina, G., Desai, A.R., Falge, E., Gove, J.H., Heimann, M.,
Hui, D., Jarvis, A.J., Kattge, J., Noormets, A., Stauch, V.J.,
2007. Comprehensive comparison of gap-filling techniques
for eddy covariance net carbon fluxes. Agric. For. Meteorol.
147, 209?232.
Myneni, R.B., Dong, J., Tucker, C.J., Kaufmann, R.K., Kauppi, P.E.,
Liski, J., Zhou, L., Alexeyev, V., Hughes, M.K., 2001. A large
carbon sink in the woody biomass of northern forests. PNAS
98, 14784?14789.
Myneni, R.B., Hoffman, S., Knyazikhin, Y., Privette, J.L., Tian, Y.,
Wang, Y., Song, X., Zhang, Y., Smith, G.R., Lotsch, A., Friedl,
M., Morisette, J.T., Votava, P., Nemani, R.R., Running, S.W.,
2002. Global products of vegetation leaf area and fraction
absorbed PAR from year one of MODIS data. Remote Sens.
Environ. 83, 214?231.
National Climatic Data Center, 2008. http://lwf.ncdc.noaa.gov/
oa/climate/research/2005/perspectives.html (visited on 01/
28/08).
Noormets, A., Chen, J., Crow, T.R., 2007. Age-dependent changes
in ecosystem carbon fluxes in managed forests in northern
Wisconsin, USA. Ecosystems 10, 187?203.
Noormets, A., Desai, A.R., Cook, B.D., Euskirchen, E.S., Ricciuto,
D.M., Davis, K.J., Bolstad, P.V., Schmid, H.P., Vogel, C.V.,
Carey, E.V., Su, H.B., Chen, J., 2008a. Moisture sensitivity of
ecosystem respiration: comparison of 14 forest ecosystems
in the Upper Great Lakes Region, USA. Agric. For. Meteorol.
148, 216?230.
Noormets, A., McNulty, S.G., DeForest, J.L., Sun, G., Li, Q., Chen,
J., 2008b. Drought during canopy development has lasting
effect on annual carbon balance in a deciduous temperate
forest. New Phytol. 179, 818?828.
Noormets, A., McNulty, S.G., Gavazzi, M.J., Sun, G., Domec, J.C.,
King, J., Chen, J., unpublished. High drought tolerance of
carbon fluxes in a coastal plain loblolly pine forest. Global
Change Biol.
Oak Ridge National Laboratory Distributed Active Archive
Center (ORNL DAAC), 2006. MODIS subsetted land
products, Collection 4. Available on-line (http://
www.daac.ornl.gov/MODIS/modis.html) from ORNL
DAAC, Oak Ridge, Tennessee, U.S.A. (accessed Month 06,
2007).
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., Hsieh, C.-I., Stoy, P., 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.
Pacala, S.W., Hurtt, G.C., Baker, D., Peylin, P., Houghton, R.A.,
Birdsey, R.A., Heath, L., Sundquist, E.T., Stallard, R.F., Ciais,
P., Moorcroft, P., Caspersen, J.P., Shevliakova, E., Moore, B.,
Kohlmaier, G., Holland, E., Gloor, M., Harmon, M.E., Fan, S.-
M., Sarmiento, J.L., Goodale, C.L., Schimel, D., Field, C.B.,
2001. Consistent land- and atmosphere-based U.S. carbon
sink estimates. Science 292, 2316?2320.
Papale, D., Valentini, A., 2003. A new assessment of European
forests carbon exchange by eddy fluxes and artificial neural
network spatialization. Global Change Biol. 9, 525?535.
Pataki, D.E., Oren, R., 2003. Species difference in stomatal
control of water loss at the canopy scale in a bottomland
deciduous forest. Adv. Water Res. 26, 1267?1278.
Penuelas, J., Gamon, J.A., Griffin, K.L., Field, C.B., 1993. Assessing
community type, plant biomass, pigment composition, and
photosynthetic efficiency of aquatic vegetation from
spectral reflectance. Remote Sens. Environ. 46, 110?118.
Potter, C.S., Randerson, J.T., Field, C.B., Matson, P.A., Vitousek,
P.M., Mooney, H.A., Klosster, S.A., 1993. Terrestrial
ecosystem production?a process model based on global
satellite and surface data. Global Biogeochem. Cyc. 7,
811?841.
Powell, T.L., Starr, G., Clark, K.L., Martin, T.A., Gholz, H.L., 2005.
Ecosystem and understory water and energy exchange for a
mature, naturally regenerated pine flatwoods forest in
north Florida. Can. J. For. Res. 35, 1568?1580.
Rahman, A.F., Sims, D.A., Cordova, V.D., El-Masri, B.Z., 2005.
Potential of MODIS EVI and surface temperature for directly
estimating per-pixel ecosystem C fluxes. Geophys. Res. Lett.
32, L19404, doi:10.1029/2005GL024127.
Ranson, K.J., Daughtry, C.S.T., Biehl, L.L., Bauer, M.E., 1985. Sun-
view angle effects on reflectance factors of corn canopies.
Remote Sens. Environ. 18, 147?161.
Reichstein, M., Falge, E., Baldocchi, D., Papale, D., Aubinet, M.,
Berbigier, P., Bernhofer, C., Buchmann, N., Gilmanov, T.,
Granier, A., Grunwald, T., Havrankova, 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., Yakir, 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., Aber, J.D., Qllinger, S.V.,
Braswell, B.H., 2007. Environmental variation is directly
responsible for short- but not long-term variation in forest-
atmosphere carbon exchange. Global Change Biol. 13,
788?803.
Richardson, A.D., Mahecha, M.D., Falge, E., Kattge, J., Moffat,
A.M., Papale, D., Reichstein, M., Stauch, V.J., Braswell, B.H.,
Churkina, G., Kruijt, B., Hollinger, D.Y., 2008. Statistical
properties of random CO2 flux measurement uncertainty
inferred from model residuals. Agric. For. Meteorol. 148,
38?50.
RuleQuest, 2008. http://www.rulequest.com (visited on 10/18/07).
Running, S.W., Hunt, E.R., 1993. Generalization of a forest
ecosystem process model for other biomes, Biome-BGC, and
an application for global-scale models. Scaling processes
between leaf and landscape levels. In: Ehleringer, J.R., Field,
C.B. (Eds.), Scaling Physiological Processes: Leaf to Globe.
Academic Press, San Diego, pp. 141?158.Running, S.W., Baldocchi, D.D., Turner, D.P., Gower, S.T.,
Bakwin, P.S., Hibbard, K.A., 1999. A global terrestrial
monitoring network integrating tower fluxes, flask
sampling, ecosystem modeling and EOS satellite data.
Remote Sens. Environ. 70, 108?127.
Running, S.W., Nemani, R.R., Heinsch, F.A., Zhao, M., Reeves, M.,
Hashimoto, H., 2004. A continuous satellite-derived
Vermote, E.F., Vermeulen, A., 1999. MODIS Algorithm Technical
Background Document?Atmospheric Correction
Algorithm: Spectral Reflectances (MOD09), Version 4.0
http://modis.gsfc.nasa.gov/data/atbd/atbd_mod08.pdf.
Wan, Z., Zhang, Y., Zhang, Q., Li, Z.-L., 2002. Validation of the
a g r i c u l t u r a l a n d f o r e s t m e t e o r o l o g y 1 4 8 ( 2 0 0 8 ) 1 8 2 7 ? 1 8 4 7 1847measure of global terrestrial primary production.
BioScience 54, 547?560.
Ryan, M.G., Law, B.E., 2005. Interpreting, measuring, and
modeling soil respiration. Biogeochemistry 73, 3?27.
Salajanu, D., Jacobs, D.M., 2005. Assessing biomass and forest
area classifications from MODIS satellite data while
incrementing the number of FIA data panels. In: Pecora 16
??Global Priorities in Land Remote Sensing??, October 23?27,
Sioux Falls, South Dakota, 10 pp.
Schmid, H.P., 1994. Source areas for scalars and scalar fluxes.
Bound.-Layer Meteorol. 67, 293?318.
Schmid, H.P., 2002. Footprint modeling for vegetation
atmospheric exchange studies: a review and perspective.
Agric. For. Meteorol. 113, 159?183.
Schmid, H.P., Grimmond, C.S.B., Cropley, F., Offerle, B., Su, H.B.,
2000. Measurements of CO2 and energy fluxes over a mixed
hardwood forest in the mid-western United States. Agric.
For. Meteorol. 103, 357?374.
Schmidt, K.S., Skidmore, A.K., 2003. Spectral discrimination of
vegetation types in a coastal wetland. Remote Sens.
Environ. 85, 92?108.
Sellers, P.J., Randall, D.A., Betts, A.K., Hall, F.G., Berry, J.A.,
Collatz, G.J., Denning, A.S., Mooney, H.A., Nobre, C.A., Sato,
N., Field, C.B., Henderson-sellers, A., 1997. Modeling the
exchanges of energy, water, and carbon between continents
and the atmosphere. Science 275, 502?509.
Shroyer, J.P., Thompson, C., Brown, R., Ohlenbach, P.D., Fjell,
D.L., Staggenborg, S., Duncan, S., Kilgore, G.L., 1996. Kansas
Crop Planting Guide, vol. L-818. Kansas State University,
Manhattan, KS, pp. 2.
Sims, D.A., Rahman, A.F., Cordova, V.D., Baldocchi, D.D.,
Flanagan, L.B., Goldstein, A.H., Hollinger, D.Y., Misson, L.,
Monson, R.K., Schmid, H.P., Wofsy, S.C., Xu, L., 2005. Midday
values of gross CO2 flux and light use efficiency during
satellite overpasses can be used to directly estimate eight-
day mean flux. Agric. For. Meteorol. 131, 1?12.
Tans, P.P., Fung, I.Y., Takahashi, T., 1990. Observational
constraints on the global atmospheric CO2 budget. Science
247, 1431?1438.
Thornton, P.E., Law, B.E., Gholz, H.L., Clark, K.L., Falge, E.,
Ellsworth, D.S., Goldstein, A.H., Monson, R.K., Hollinger, D.,
Falk, M., Chen, J., Sparks, J.P., 2002. Modeling and measuring
the effects of disturbance history and climate on carbon
and water budgets in evergreen needleleaf forests. Agric.
For. Meteorol. 113, 185?222.
Urbanski, S., Barford, C., Wofsy, S., Kucharik, C., Pyle, E.,
Budney, J., McKain, K., Fitzjarrald, D., Czikowsky, M.,
Munger, J.W., 2007. Factors controlling CO2 exchange on
timescales from hourly to decadal at Harvard Forest. J.
Geophys. Res. 112, G02020, doi:10.1029/2006JG000293.
Verma, S.B., Dobermann, A., Cassman, K.G., Walters, D.T.,
Knops, J.M., Arkebauer, T.J., Suyker, A.E., Burba, G.G., Amos,
B., Yang, H., Ginting, D., Hubbard, K.G., Gitelson, A.A.,
Walter-Shea, E.A., 2005. Annual carbon dioxide exchange in
irrigated and rainfed maize-based agroecosystems. Agric.
For. Meteorol. 131, 77?96.land-surface temperature products retrieved from Terra
Moderate Resolution Imaging Spectroradiometer data.
Remote Sens. Environ. 83, 163?180.
Waring, R.H., Franklin, J.F., 1979. Evergreen coniferous forests of
the Pacific Northwest. Science 204, 1380?1386.
Watts, C.J., Scott, R.L., Garatuza-Payan, J., Rodriguez, J.C.,
Prueger, J., Kustas, W., Douglas, M., 2007. Changes in
vegetation condition and surface fluxes during NAME 2004.
J. Climate 20, 1810?1820.
Wofsy, S.C., Goulden, M.L., Munger, J.W., Fan, S.-M., Bakwin, P.S.,
Daube, B.C., Bassow, S.L., Bazzaz, F.A., 1993. Net exchange of
CO2 in a mid-latitude forest. Science 260, 1314?1317.
Wylie, B.K., Fosnight, E.A., Gilmanov, T.G., Frank, A.B., Morgan,
J.A., et al., 2007. Adaptive data-driven models for estimating
carbon fluxes in the Northern Great Plains. Remote Sens.
Environ. 106, 399?413.
Xiao, J., Moody, A., 2004. Photosynthetic activity of US biomes:
responses to the spatial variability and seasonality of
precipitation and temperature. Global Change Biol. 10,
437?451.
Xiao, J., Moody, A., 2005. A comparison of methods for
estimating fractional green vegetation cover within a dsert-
to-upland transition zone in central New Mexico, USA.
Remote Sens. Environ. 98, 237?250.
Xiao, X., Zhang, Q., Saleska, S., Hutyra, L., Camargo, P.D., et al.,
2005. Satellite-based modeling of gross primary production
in a seasonally moist tropical evergreen forest. Remote
Sens. Environ. 94, 105?122.
Xu, L., Baldocchi, D.D., 2004. Seasonal variation in carbon
dioxide exchange over a Mediterranean annual grassland in
California. Agric. For. Meteorol. 123, 79?96.
Yamaji, T., Sakai, T., Endo, T., Baruah, P.J., Akiyama, T., Saigusa,
N., Nakai, Y., Kitamura, K., Ishizuka, M., Yasuoka, Y., 2008.
Scaling-up technique for net ecosystem productivity of
deciduous broadleaved forests in Japan using MODIS data.
Ecol. Res. 23, 765?775.
Yang, L., Huang, C., Homer, C., Wylie, B.K., Coan, M.J., 2003. An
approach for mapping large-area impervious surfaces:
synergistic use of Landsat-7 ETM+ and high spatial
resolution imagery. Can. J. Remote Sens. 29, 230?240.
Yang, F., Ichii, K., White, M.A., Hashimoto, H., Michaelis, A.R.,
Votava, P., Zhu, A.-X., Huete, A., Running, S.W., Nemani,
R.R., 2007. Developing a continental-scale measure of gross
primary production by combining MODIS and AmeriFlux
data through support machine approach. Remote Sens.
Environ. 110, 109?122.
Zhao, M., Heinsch, F.A., Nemani, R.R., Running, S.W., 2005.
Improvements of the MODIS terrestrial gross and net
primary production global data set. Remote Sens. Environ.
95, 164?175.
Zhuang, Q., McGuire, A.D., Melillo, J.M., Clein, J.S., Dargaville, R.J.,
Kicklighter, D.W., Myneni, R.B., Dong, J., Romanovsky, V.E.,
Harden, J., Hobbie, J.E., 2003. Carbon cycling in extratropical
terrestrial ecosystems of the Northern Hemisphere during
the 20th Century: a modeling analysis of the influences of soil
thermal dynamics. Tellus 55B, 751?776.