S

Af

Xa

b

c

a

ARRAA

KNEMELL

1

rsebtasaaret2

lss2

0d

Ecological Engineering 37 (2011) 1567– 1571

Contents lists available at ScienceDirect

Ecological Engineering

j ourna l ho me page: www.elsev ier .com/ locate /eco leng

hort communication

new model of net ecosystem carbon exchange for the deciduous-dominatedorest by integrating MODIS and flux data

uguang Tanga,b, Dianwei Liua, Kaishan Songa, J. William Mungerc, Bai Zhanga, Zongming Wanga,∗

Northeast Institute of Geography and Agroecology, Chinese Academy of Sciences, Changchun 130012, ChinaGraduate University of Chinese Academy of Sciences, Beijing 100049, ChinaDivision of Applied Sciences, Department of Earth and Planetary Sciences, Harvard University, Cambridge, MA, USA

r t i c l e i n f o

rticle history:eceived 29 December 2010eceived in revised form 23 March 2011ccepted 27 March 2011vailable online 22 April 2011

a b s t r a c t

The eddy covariance technique provides continuous measurements of plot-level net ecosystem carbonexchange (NEE) across a wide range of vegetation types. However, these NEE estimates only representfluxes at the tower footprint scale. To quantify the NEE over regions or continents, flux tower measure-ments need to be up-scaled to large areas. In the present study, we propose a new NEE model solelybased on Moderate Resolution Imaging Spectroradiometer data, including enhanced vegetation index

′

eywords:EEddy covarianceODIS

VIST

(EVI), land surface water index (LWSI), land surface temperature (LST), and Terra nighttime LST . Site-specific data from the deciduous-dominated Harvard Forest flux site were used. Analysis covered sixyears (2001–2006) of CO2 flux data. The data of the first four years were used for model building and therest as validation set. Compared with the model based solely on EVI, we also introduced LST and LSWI intothe new model. The results showed that this method could further improve the precision (R2 and RMSEreached 0.857 and 1.273, respectively) and generally capture the expected seasonal patterns of NEE.

w(amiaaFcette(obnp

SWI

. Introduction

Net carbon uptake by terrestrial ecosystems plays a significantole in the global carbon cycle and helps to mitigate atmo-pheric increases due to emissions from burning fossil fuels (Alleyt al., 2007). Net ecosystem carbon exchange (NEE), the differenceetween photosynthetic uptake and release of CO2 by respira-ion from autotrophs (vegetation) and heterotrophs (free-livingnd fauna in the soil and symbiotic microorganisms), repre-ents the net exchange of CO2 between terrestrial ecosystemsnd the atmosphere. Accurately estimating the spatial patternsnd temporal dynamics of NEE of terrestrial ecosystems at theegional and global scale is of great interest to human soci-ty and is necessary for understanding the carbon cycle of theerrestrial biosphere (Jiménez et al., 2008; Xiao et al., 2008,010).

Despite the consensus that temperate forests in the middle-atitude districts of the Northern Hemisphere provide an important

ink for storing atmospheric CO2, the size and distribution of theink still remain uncertain (Houghton et al., 1999; Law et al.,002). Traditionally, inventory studies of biomass and soil carbon

∗ Corresponding author. Tel.: +86 43185542364; fax: +86 43185542299.E-mail address: wangzm.cas@gmail.com (Z. Wang).

bRtfpee

925-8574/$ – see front matter © 2011 Elsevier B.V. All rights reserved.oi:10.1016/j.ecoleng.2011.03.030

© 2011 Elsevier B.V. All rights reserved.

ere used to quantify an ecosystem NEE over a specific periodClark et al., 2001). In recent years, evaluating carbon balancend its seasonal and annual variations in terrestrial ecosystemsore precisely has become possible (Baldocchi et al., 2001). This

mprovement is due to the development of the eddy covari-nce technique, which offers an alternative way to continuouslynd steadily assess the long-term ecosystem carbon exchange.urthermore, carbon budgets and the effects of environmentalontrols have been quantified with this technique for many for-st types across the continent (Powell et al., 2006). However,hese NEE estimates only represent fluxes at the scale of theower footprint ranging between a hundred meters and sev-ral kilometers relying on homogeneous vegetation and fetchGöckede et al., 2008). To quantify the net exchange of CO2ver regions or continents, flux tower measurements need toe up-scaled to large areas (Xiao et al., 2008, 2010). Therefore,ew approaches are critically needed to extend the role of fieldlots to capture regional variation and to bridge a major gapetween field and satellite observations (Gregory et al., 2010).emote sensing provides observations of ecosystems with spa-ially and temporally consistent coverage, and is a valuable toolor up-scaling carbon fluxes. Earlier studies have shown good

erformance in integrating Gross Primary Production (GPP) orcosystem respiration (Re) with MODIS (Wylie et al., 2007; Simst al., 2008; Xiao et al., 2010). However, to our knowledge, few

1568 X. Tang et al. / Ecological Engineering 37 (2011) 1567– 1571

F t site

f

sm

ebliomdsfl

2

2

(sm

Ff

ig. 1. Comparison between NEE, EVI, and LSWI in 2001–2006 at the Harvard Foresrom 1 to 46.

tudies have scaled eddy flux NEE measurements to this simplerodel.So, in this study, we investigated the deciduous broadleaf for-

st of the Harvard Forest site. The objectives were: (1) to examineiophysical performance of vegetation indices (EVI and LSWI) and

and surface temperature (LST) in relation to seasonal dynam-cs of NEE fluxes, and (2) in contrast to the earlier study basednly on EVI and LST, to estimate GPP or Re, to develop a new

odel on NEE estimation of deciduous broadleaf forest by intro-

ucing LSWI. This study will explore the potential of remoteensing for studying and monitoring vegetation and the carbonuxes.

tFi(

ig. 2. Comparison between NEE, LST, and LST′ in 2001–2006 at the Harvard Forest site from 1 to 46.

for each eight-day period. For x-axis, the data refer to week of year (WOY) ranging

. Methods

.1. Study sites

The analysis was based on the data from the Harvard Forest site42.538◦N and 72.171◦W; elev. 340 m). NEE, fluxes of latent andensible heat, and associated meteorological variables have beeneasured since 1989. The Harvard tower is located in a mixed

emperate forest, approximately 110 km west of Boston, MA, USA.orest composition is dominated by deciduous species, includ-ng red oak (Quercus rubra L.; 36% of basal area) and red maple22%), with lesser quantities of other hardwoods, including black

or each eight-day period. For x-axis, the data refer to week of year (WOY) ranging

X. Tang et al. / Ecological Engineering 37 (2011) 1567– 1571 1569

F the pv

olC(Mi(

2

Etsunafarc(Sdf2

te

TvMtRoOwv7(biE

2

e

FN

ig. 3. A linear comparison between the predicted and measured NEE values duringalues are from simulations of TG model using eight-day MODIS composites.

ak (Quercus velutina Lam.), white oak (Quercus alba L.), and yel-ow birch (Betula alleghaniensis Britton) covering 14% of basal area.onifers include eastern hemlock (13% of basal area), red pinePinus resinosa Aiton; 8%), and white pine (Pinus strobus L.; 6%).

ean annual temperature is 6.5 ◦C and mean annual precipitations 1000 mm. Plant-growing season usually starts around mid-May∼day of year 130) and lasts about 160 days (Urbanski et al., 2007).

.2. Modis products

Optical remote sensing systems measure the reflectance of thearth’s surface. The reflectance depends on biophysical proper-ies (e.g., biomass, leaf area, and stand age), soil moisture, andun-sensor geometry. Therefore, reflectance of MODIS can provideseful information for estimating NEE. This temperature and green-ess (TG) model includes two vegetation indices as input data: EVInd LSWI. EVI directly adjusts the reflectance in the red band as aunction of the reflectance in the blue band, accounting for residualtmospheric contamination, variable soil, and canopy backgroundeflectance (Huete et al., 2002). LSWI was shown to be stronglyorrelated with leaf water content (equivalent water thickness)Jackson et al., 2004) and soil moisture overtime (Fensholt and

andholt, 2003). The other explanatory variable is LST. The LSTerived from MODIS is a measure of soil temperature at the sur-ace, which agreed with in situ measured LST within 1 K in the63–322 K (Wan et al., 2002). A few studies have demonstrated that

luva

ig. 4. A comparison of the seasonal dynamics of weekly NEE at the Harvard Forest site

EE is from the CO2 flux tower site. For x-axis, the data refer to week of year (WOY) rang

eriods 2001–2004 (a) and 2005–2006 (b) at the Harvard Forest site. Predicted NEE

he satellite-derived LST was strongly correlated with Re (Rahmant al., 2005; Schubert et al., 2010).

Therefore, we selected EVI, LSWI, and LST (here, LST includeserra daytime LST and Terra nighttime LST′) as the explanatoryariables to estimate NEE. All these variables were derived fromODIS data, which also avoided the complications and difficul-

ies in merging disparate data sources. The eight-day Land Surfaceeflectance (MOD09A1) data and LST data (MOD11A2) werebtained from the 7 × 7 km subsets of MODIS products available atak Ridge National Laboratory’s Distributed Active Archive Centereb site (http://www.modis.ornl.gov/modis/index.cfm). Average

alues for the central 3 × 3 km area were extracted within the × 7 km cutouts to better represent the flux tower footprintRahman et al., 2005). Reflectance values of these four spectralands – blue, red, near infrared (841–875 nm), and shortwave

nfrared (1628–1652 nm) – in 2001–2006 were used to calculateVI and LSWI.

.3. Calculation of tower-based C fluxes

The direct, long-term measurement of carbon fluxes by theddy covariance technique offers the possibility of assessing the

ocal carbon sequestration rates of forests and different land-ses. This technique also provides a better understanding of theulnerability of the carbon balance of ecosystems to climate vari-bility. In addition, it can be used to validate ecosystem models

during the period 2005–2006. Predicted NEE is from the TG model, and measureding from 1 to 46.

1 ginee

ammFda2eaippvf

2

Mtssuctms

3

3

tdfottoiticeagtdeooGiw2

efpfiiS3

vuMeep

w0ctts

3

nsob

N

rm1

N

swLpv

mvt(dawtpe1NiwWEywwtd

3

570 X. Tang et al. / Ecological En

nd provide data for land surface exchange schemes in globalodels (Valentini et al., 2000). We used six-year eddy covarianceeasurements (2001–2006) of NEE of CO2 made at the Harvard

orest site (http://public.ornl.gov/ameriflux/). Site-specific proce-ures, including quality control, flux corrections, and data editing,re described elsewhere (Barford et al., 2001; Urbanski et al.,007). The Level 4 product consists of NEE data with four differ-nt time steps, including half-hourly, daily, weekly (eight-day),nd monthly. We used weekly NEE data to match the composit-ng intervals of MODIS. Moreover, the average NEE over such aeriod was shown to largely eliminate micrometeorological sam-ling errors, with the remaining spatial variability representingariation in ecosystem attributes (Oren et al., 2006), here accountedor by data from MODIS.

.4. Model development

We developed a predictive NEE model based on site-specificODIS and flux data. The explanatory variables included EVI, day-

ime and nighttime LST, and LSWI. The target variable was NEE. Weplit the site-level data set into a training set (2001–2004) and a testet (2005–2006). Altogether, a total of 155 and 83 data points weresed for the training and test sets, respectively. After analyzing theorrelation between these parameters and NEE observed, based onhe multiple linear regression method, the model with the maxi-

um coefficient of determination (R2) and the minimal root meanquare error (RMSE) was chosen as the best model.

. Results and discussion

.1. Seasonal dynamics of NEE with EVI, LSWI, LST and LST′

Figs. 1 and 2 showed that the NEE time series in 2001–2006 athe Harvard Forest site had distinct seasonal cycles. The seasonalynamics of NEE can be explained in part by the change of land sur-ace temperature. During the winter season of 2001–2006 (Weekf year – WOY ranging from 1 to 16 and from 39 to 46), because lowemperature and frozen soils inhibited the photosynthetic activi-ies of the deciduous broadleaf forest, NEE values were positive,wing to GPP values were near zero and NEE were mostly dom-nated by ecosystem respiration. From WOY 17 of each year, asemperature rose over the minimum limit of photosynthetic activ-ties, vegetation began to grow and the ecosystem photosynthesisapability gradually increased, the carbon uptake of photosyntheticxceeded the release of CO2 through respiration. NEE also increasednd reached its peak value during WOY 25–30. Subsequently, NEEradually declined with temperature and forest started to wither. Inhe whole process, EVI and LSWI had consistent trend because theyirectly reflected the growth status of the deciduous broadleaf for-st. In addition, spring phenology is thought to exert a major effectn carbon balance. For all phenological indicators, earlier springnset consistently, but not always significantly, resulted in higherPP and Re for both seasonal and annual flux integrals. The increase

n Re was less than that in GPP; in response to earlier spring onsetould increase springtime NEE and this effect was pronounced as

004 by 2–4 g C m−2 day−1 more than the other years.Fig. 1 showed that EVI and LSWI of the deciduous broadleaf for-

st had strong seasonal dynamics. The vegetation indices derivedrom MODIS data captured well the beginning and ending of thelant growing season in 2001–2006. The Pearson correlation coef-

cients were 0.887 and 0.463, respectively. EVI began an abrupt

ncrease on WOY 17, and reached its peak during WOY 25–28.ubsequently, EVI gradually declined and remained low after WOY9. In the winter months (December, January, and February), LSWI

ide

ring 37 (2011) 1567– 1571

alues were high, which was attributed to the snow cover on andnder the forest canopy. As snow melted in the spring (aroundarch), LSWI gradually declined. The LSWI time series data can be

asily recognized with a spring trough and a fall trough. It is inter-sting to note that LSWI increases again as the plant growing seasonrogresses, and declines gradually in the late growing season.

Both daytime LST and nighttime LST′ had strong correlationsith NEE. The Pearson correlation coefficients were 0.691 and

.763, respectively. It implies that nighttime LST′ is more closelyorrelated with NEE than daytime LST. This may be attributed tohe relationship between nighttime LST′ and ecosystem respira-ion. However, we were not aware of the influencing degree. Theeasonal dynamics of NEE also agreed well with that of LST (Fig. 2).

.2. Predicted NEE from eight-day composites of MODIS data

After analyzing the relationships between EVI, daytime LST,ighttime LST′, and LSWI with NEE measured at the Harvard Forestite from 2001 to 2004, we established the regression model basedn each single variable. And the model based on EVI exhibited theest performance. NEE is calculated using the following equation:

EE = 7.527 − 20.339 × EVI (R2 is 0.830 and RMSE is 1.438) (1)

Then we developed a new model using the multiple linearegression method. Compared with the others, we got the bestodel with the maximum R2 of 0.857 and the minimal RMSE of

.273. The new empirical TG model for NEE was defined as:

EE = 26.523 − 15.828 × EVI − 0.076 × LST′ − 3.604 × LSWI (2)

The TG model was run at eight-day time scale using the site-pecific EVI, nighttime LST′, and LSWI to predict NEE. Daytime LSTas not used because of its high correlation of 0.96 with nighttime

ST′. Model performance was then evaluated using scatter plots ofredicted (NEEpre) versus measured NEE (NEEmea) and the seasonalariations between them in 2005–2006.

Generally, the seasonal dynamics of NEE simulated by the TGodel solely based on MODIS data agreed reasonably well with the

ariation of NEEmea from the CO2 flux tower site in 2005–2006 andhe simple linear regression model also showed good agreementFigs. 3 and 4). Fig. 3 indicated that the residuals were not randomlyistributed. In absolute magnitude, low NEE values were generallyssociated with lower prediction errors, whereas high NEE valuesere associated with higher prediction errors. This suggests that

he uncertainties of carbon flux measurements are directly pro-ortional to the magnitude of the fluxes. A few discrepancies alsoxisted between NEEmea and NEEpre. For instance, during WOY9–20 in 2005 (Fig. 4a), the NEEpre values were higher than theEEmea values. It may be explained in part by an abrupt increase

n EVI in the corresponding period. Occasionally, the NEEpre valuesere lower than NEEmea, as in the peak of the 2006 growing season.hen MODIS LST data missed, we used Eq. (1) exclusively based on

VI to evaluate the missing NEEpre values. The NEEpre of the wholeear in 2006 at the deciduous-dominated Harvard Forest (Fig. 4b)as −477.47 g C/m2, whereas the sum of 46 weekly NEEmea valuesas −462.26 g C/m2 – only a small difference of 3.3%. Therefore,

his simpler empirical model is a good indicator of both seasonalynamics and annual NEE.

.3. Analysis of uncertainties

Despite the encouraging performance of our predictive modeln estimating NEE and capturing the dynamics patterns of theeciduous-dominated forest ecosystem carbon exchange, the NEEstimates still contain significant uncertainties.

ginee

taNrlae2v

decemCsat(tvp

4

tdpbaTssseoN

A

ttSwTsm

R

A

B

B

B

B

C

F

G

G

H

H

J

J

J

L

O

P

P

R

R

S

S

U

V

W

W

Xmation of net ecosystem carbon exchange for the conterminous United States bycombining MODIS and AmeriFlux data. Agric. Forest Meteorol. 148, 1827–1847.

X. Tang et al. / Ecological En

In the present study, weekly NEE data were used to matchhe compositing intervals of MODIS. The average NEE over such

period well reflected the seasonal dynamics of CO2 exchange.evertheless, the magnitudes of sinks and sources fluctuate

emarkably on annual or longer timescales due to variable climate,and use change, disturbance by fire and pests, and changes in thege distribution and species composition of the ecosystem (Battlet al., 2000; Johnson et al., 2007; Pietrzykowski and Krzaklewski,007). Thus, the reliability of this TG model needs to be furtheralidated under different time scales.

We selected only EVI, LSWI, and LST (here, LST includes Terraaytime LST and nighttime LST′) as the explanatory variables instimating NEE. However, the factors controlling CO2 exchange areomplicated. NEE is the sum of canopy photosynthesis (GEE) andcosystem respiration (Re). GEE is controlled by canopy develop-ent and nutrient status, light, temperature, ambient humidity,

O2 concentration, and soil moisture (Ruimy et al., 1996). Fieldtudies of Re have identified temperature, soil moisture, nutrientvailability, stocks of living and dead biomass, ecosystem produc-ivity, and seasonal carbon allocation as the controlling factorsBoone et al., 1998). Thus, in future research, additional explana-ory variables should be selected to better account for live and deadegetation carbon pools, and other factors that influence decom-osition of woody detritus and soil respiration (Xiao et al., 2008).

. Conclusion

Time series analysis of NEE with EVI, LSWI, LST, and LST′ showedhat all these variables of the deciduous-dominated forest hadistinct seasonal dynamics. NEE can be explained in part by tem-erature. In the whole process, EVI and LSWI had consistent trendecause they directly reflected the growing state. Moreover, therelso existed a strong correlation between NEE and these factors.hus, we integrated MODIS and flux data at the Harvard Forestite to develop a predictive NEE model. Compared with the modelolely based on EVI, we also introduced LST and LSWI. The resultshowed that this method further improved the precision and gen-rally captured the expected seasonal patterns of NEE. Therefore,ur empirical approach has great potential for scaling up plot-levelEE of CO2 to large areas.

cknowledgements

This study was supported by the Strategic Frontier Program ofhe Chinese Academy of Sciences (Grant no. XDA05050101) andhe Knowledge Innovation Program of the Chinese Academy ofciences (Grant no. KZCX2-YW-QN305). Partial data of this studyas presented at the 2010 International Conference on Multimedia

echnology. We thank two anonymous reviewers for their con-tructive comments and the many participants involved in theaintenance and operation of the Harvard Forest flux site.

eferences

lley, R., Berntsen, T., Bindoff, N.L., Chen, Z., Chidthaisong, A., Fridlingstein, P., 2007.

Climate change 2007: the physical science basis, summary for policymakers.Intergovern. Panel Clim. Change, 2–21.

aldocchi, D., Falge, E., Gu, L., 2001. FLUXNET: a new tool to study the temporal andspatial variability of ecosystem scale carbon dioxide, water vapor, and energyflux densities. Bull. Am. Meteorol. Soc. 82, 2415–2434.

X

ring 37 (2011) 1567– 1571 1571

arford, C.C., Wofsy, S.C., Goulden, M.L., Munger, J.W., 2001. Factors controllinglong and short term sequestration of atmospheric CO2 in a mid-latitude forest.Science 294, 1688–1691.

attle, M., Bender, M.L., Tans, P.P., 2000. Global carbon sinks and their variabilityinferred from atmospheric O2 and delta 13C. Science 287, 2467–2470.

oone, R.D., Nadelhoffer, K.J., Canary, J.D., Kaye, J.P., 1998. Roots exert a strong influ-ence on the temperature sensitivity of soil respiration. Nature 396, 570–572.

lark, D.A., Brown, S., Kicklighter, D.W., Chambers, J.Q., Thomlinson, J.R., Ni, J., 2001.Measuring net primary production in forests: concepts and field methods. Ecol.Appl. 11, 356–370.

ensholt, R., Sandholt, I., 2003. Derivation of a shortwave infrared water stressindex from MODIS near and shortwave infrared data in a semiarid environment.Remote Sens. Environ. 87, 111–121.

öckede, M., Foken, T., Aubinet, M., Aurela, M., Banaz, J., Bernhofer, C., 2008. Qualitycontrol of CarboEurope flux data, part 1: coupling footprint analyses with fluxdata quality assessment to evaluate sites in forest ecosystems. Biogeosciences5, 433–450.

regory, P.A., George, V.N.P., Joseph, M., David, E.K., John, K.C., James, J., 2010.High resolution forest carbon stocks and emissions in the Amazon. PNAS 107,16738–16742.

oughton, R.A., Hackler, J.L., Lawrence, K.T., 1999. The U.S. carbon budget: contri-butions from land-use change. Science 285, 574–578.

uete, A., Didan, K., Miura, T., Rodriguez, E.P., Gao, X., Ferreira, L.G., 2002. Overview ofthe radiometric and biophysical performance of the MODIS vegetation indices.Remote Sens. Environ. 83, 195–213.

ackson, 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 derivednormalized difference water index for corn and soybeans. Remote Sens. Environ.92, 475–482.

iménez, J.J., Lal, R., Russo, R.O., Leblanc, H.A., 2008. The soil organic carbon inparticle-size separates under different re-growth forest stands of north easternCosta Rica. Ecol. Eng. 34, 300–310.

ohnson, D., Murphy, J.D., Walker, R.F., Glass, D.W., Miller, W.W., 2007. Wildfireeffects on forest carbon and nutrient budgets. Ecol. Eng. 31, 183–192.

aw, B.E., Falge, E., Gu, L., Baldocchi, D.D., Bakwin, P., Berbigier, P., 2002. Environ-mental controls over carbon dioxide and water vapor exchange of terrestrialvegetation. Agric. Forest Meteorol. 113, 97–120.

ren, R., Hsieh, C.L., Stoy, P., Albertson, J., Rmccarthy, H., Harrell, P., Katul, G., 2006.Estimating the uncertainty in annual net ecosystem carbon exchange: spatialvariation in turbulent fluxes and sampling errors in eddy-covariance measure-ments. Global Change Biol. 12, 883–896.

ietrzykowski, M., Krzaklewski, W., 2007. An assessment of energy efficiency inreclamation to forest. Ecol. Eng. 30, 341–348.

owell, T.L., Bracho, R., Li, J.H., Dore, S., Hinkle, C.R., Drake, B.G., 2006. Environmentalcontrols over net ecosystem carbon exchange of scrub oak in central Florida.Agric. Forest Meteorol. 141, 19–34.

ahman, A.F., Sims, D.A., Cordova, V.D., El-masri, B.Z., 2005. Potential of MODIS EVIand surface temperature for directly estimating per-pixel ecosystem C fluxes.Geophys. Res. Lett. 32, L19404.1–L19404.6.

uimy, A., Jarvis, P.G., Baldocchi, D.D., Saugier, B., 1996. CO2 fluxes over plantcanopies and solar radiation: a review. Adv. Ecol. Res. 26, 1–68.

chubert, P., Eklundeh, L., Lund, M., Nilsson, M., 2010. Estimating northern peat-land CO2 exchange from MODIS time series data. Remote Sens. Environ. 114,1178–1189.

ims, D.A., Rahman, A.F., Vicente, D.C., 2008. A new model of gross primary produc-tivity for North American ecosystems based solely on the enhanced vegetationindex and land surface temperature from MODIS. Remote Sens. Environ. 112,1633–1646.

rbanski, S., Barford, C., Wofsy, S., Kucharik, C., Pyle, E., Budney, J., Mckain, K., 2007.Factors controlling CO2 exchange on time scales from hourly to decadal at Har-vard Forest. J. Geophys. Res. 112, 1–25.

alentini, R., Matteucci, G., Dolman, A.J., 2000. Respiration as the main determinantof carbon balance in European forests. Nature 404, 861–865.

an, Z.M., Zhang, Y.L., Zhang, Q.C., Li, Z.L., 2002. Validation of the land surfacetemperature products retrieved from Terra Moderate Resolution Imaging Spec-troradiometer data. Remote Sens. Environ. 83, 163–180.

ylie, B.K., Fosnight, E.A., Gilmanov, T.G., Frank, A.B., Morgan, J.A., Haferkamp, M.R.,Meyers, T.P., 2007. Adaptive data-driven models for estimating carbon fluxes inthe Northern Great Plains. Remote Sens. Environ. 106, 399–413.

iao, J., Zhuang, Q., Baldocchi, D.D., Law, B.E., Richardson, A.D., Chen, J.Q., 2008. Esti-

iao, J., Zhuang, Q., Law, B.E., Chen, J., Baldocchi, D.D., Cook, D.R., 2010. A continuousmeasure of gross primary production for the conterminous U.S. derived fromMODIS and AmeriFlux data. Remote Sens. Environ. 114, 576–591.