publications · 2020. 2. 18. · improving ecosystem productivity modeling through spatially...

Improving ecosystem productivity modelingthrough spatially explicit estimationof optimal light use efficiencyNima Madani1,2, John S. Kimball1,2, David L. R. Affleck3, Jens Kattge4, Jon Graham5,Peter M. van Bodegom6, Peter B. Reich7,8, and Steven W. Running2

1Flathead Lake Biological Station, University of Montana, Polson, Montana, USA, 2Numerical Terradynamic Simulation Group,University of Montana, Missoula, Montana, USA, 3College of Forestry and Conservation, University of Montana, Missoula,Montana, USA, 4Max Planck Institute for Biogeochemistry, Jena, Germany, 5Department of Mathematical Sciences, Universityof Montana, Missoula, Montana, USA, 6Department of Systems Ecology, VU University Amsterdam, Amsterdam, Netherlands,7Department of Forest Resources, University of Minnesota, St. Paul, Minnesota, USA, 8Hawkesbury Institute for theEnvironment, University of Western Sydney, Sydney, New South Wales, Australia

Abstract A common assumption of remote sensing-based light use efficiency (LUE) models for estimatingvegetation gross primary productivity (GPP) is that plants in a biome matrix operate at their photosyntheticcapacity under optimal climatic conditions. A prescribed constant biomemaximum light use efficiency parameter(LUEmax) defines the maximum photosynthetic carbon conversion rate under these conditions and is a largesource of model uncertainty. Here we used tower eddy covariance measurement-based carbon (CO2) fluxes forspatial estimation of optimal LUE (LUEopt) across North America. LUEopt was estimated at 62 Flux Network sitesusing tower daily carbon fluxes and meteorology, and satellite observed fractional photosynthetically activeradiation from theModerate Resolution Imaging Spectroradiometer. A geostatistical model was fitted to 45 fluxtower-derived LUEopt data points using independent geospatial environmental variables, including global planttraits, soil moisture, terrain aspect, land cover type, and percent tree cover, and validated at 17 independenttower sites. Estimated LUEopt shows large spatial variability within and among different land cover classesindicated from the sparse tower network. Leaf nitrogen content and soil moisture regime are major factorsexplaining LUEopt patterns. GPP derived from estimated LUEopt shows significant correlation improvementagainst tower GPP records (R2 = 76.9%; mean root-mean-square error (RMSE) = 257gCm�2 yr�1), relative toalternative GPP estimates derived using biome-specific LUEmax constants (R

2 =34.0%; RMSE=439gCm�2 yr�1).GPP determined from the LUEopt map also explains a 49.4% greater proportion of tower GPP variability atthe independent validation sites and shows promise for improving understanding of LUE patterns andenvironmental controls and enhancing regional GPP monitoring from satellite remote sensing.

1. Introduction

Satellite remote sensing, as the only means of monitoring vegetation changes at global scales, has beenwidely applied to determine plant productivity and ecosystem dynamics [e.g., Running et al., 2004; Xiao et al.,2005; Kimball et al., 2006; Zhao and Running, 2010]. However, estimation of plant carbon uptake at large scalesusing remote sensing products is bound with uncertainties [Heinsch et al., 2006; Hilker et al., 2008]. Whileupscaled site level vegetation gross primary production (GPP) estimates show a mean global terrestrialcarbon uptake of 123±8PgCyr�1 [Beer et al., 2010]; remote sensing productivity estimates show significantlylower levels (~109.3 PgC yr�1) [Zhao et al., 2005].

Optical remote sensing data-driven methods for estimating GPP generally rely on spectral vegetation indicesof photosynthetic canopy cover derived from visible and near-infrared reflectances and other ancillarybiophysical inputs including general land cover and plant functional type characteristics, incident solarradiation, and surface meteorology [Kimball et al., 2009]. However, limitations with respect to available groundtruth data for model development, calibration and validation, and poor model assumptions are among themain sources of uncertainties in remote sensing-based ecosystem productivity models [Ahl et al., 2004; Yuanet al., 2007]. Additionally, having a good estimation of ecosystem productivity requires detailed knowledge ofvegetation phenology [Jin et al., 2013] and canopy photosynthetic response to variations in environmentalconditions [Li et al., 2008] within and between plant functional types.

MADANI ET AL. ©2014. American Geophysical Union. All Rights Reserved. 1755

PUBLICATIONSJournal of Geophysical Research: Biogeosciences

RESEARCH ARTICLE10.1002/2014JG002709

Key Points:• Quantifying ecosystem optimal lightuse efficiency

• Optimum light use efficiency showsspatial variability within and amongbiome types

• Spatially explicit optimum light useefficiency dramatically improvesremote sensing ecosystemproductivity modeling

Supporting Information:• Readme• Appendix S1

Correspondence to:N. Madani,[email protected]

Citation:Madani, N., J. S. Kimball, D. L. R. Affleck,J. Kattge, J. Graham, P. M. van Bodegom,P. B. Reich, and S. W. Running (2014),Improving ecosystem productivitymodeling through spatially explicitestimation of optimal light useefficiency, J. Geophys. Res. Biogeosci.,119, 1755–1769, doi:10.1002/2014JG002709.

Received 12 MAY 2014Accepted 30 JUL 2014Accepted article online 6 AUG 2014Published online 2 SEP 2014

http://publications.agu.org/journals/http://onlinelibrary.wiley.com/journal/10.1002/(ISSN)2169-8961http://dx.doi.org/10.1002/2014JG002709http://dx.doi.org/10.1002/2014JG002709

Optical remote sensing-based productivity models such as the Carnegie–Ames–Stanford Approach [Potteret al., 1993], Terrestrial Uptake and Release of Carbon [Ruimy et al., 1996], C-Fix [Veroustraete et al., 2002],and MOD17 [Running et al., 2004] are based on the light use efficiency (LUE) concept [Monteith, 1972; Kumarand Monteith, 1981]. In this logic, plant production is linearly related to photosynthetically active radiation(PAR) absorbed by the vegetation canopy (absorbed photosynthetically active radiation (APAR)) and theefficiency with which this solar radiant energy is transformed into vegetation biomass through netphotosynthesis (light use efficiency).

Maximum light use efficiency (LUEmax) defines the canopy photosynthetic capacity or maximum rate ofconversion of APAR to vegetation biomass (gCMJ�1) under optimal (nonlimiting) environmental conditions[Monteith, 1972]. The LUEmax parameter is reduced under suboptimal temperature and water deficitconditions and varies according to vegetation type and environment [Bartlett et al., 1989; Trapani et al., 1992].This has been the basis of the NASA MODIS (Moderate Resolution Imaging Spectroradiometer) GPP product(MOD17) [Running et al., 2004; Zhao et al., 2005]. The MOD17 GPP product is currently the only remote sensingglobal operational ecosystem productivity data record and has been in production from MODIS on boardthe NASA EOS Terra and Aqua satellites since 2000 and 2002, respectively. The MOD17 GPP product usesMODIS-derived fractional photosynthetically active radiation (FPAR) [Myneni et al., 1999] to estimate terrestrialecosystem GPP globally at 8 day intervals with 1 km spatial resolution. The MOD17 algorithm is based on theassumption that LUEmax variability is conservative within individual biomes [Monteith and Moss, 1977] anddefines general biophysical response characteristics to estimate GPP using a Biome Property Look-Up Table(BPLUT) [Zhao et al., 2005] and global land cover classification [Friedl et al., 2010] that defines 11 general plantfunctional types. However, LUEmax can show large variability even within the same plant functional type [Goetzand Prince, 1996; Gower et al., 1999; Turner et al., 2002], and the fixed parameter for LUEmax is a major sourceof uncertainty for ecosystemproductivitymodeling [Ruimy et al., 1994;Way et al., 2005; Pan et al., 2006;Wang et al.,2010]. The biome background matrix defined using fixed LUEmax values increases model GPP uncertainty,because spatial heterogeneity in vegetation light use efficiency is underrepresented [Turner et al., 2002].

Within a given biome type, and independent of direct environmental forcings, LUE and hence ecosystemproductivity are affected by stand age and soil nutrition [Huston and Wolverton, 2009; Malhi, 2012], leafnitrogen concentrations [Kergoat et al., 2008; Ollinger et al., 2008; Reich, 2012], and canopy structure and leaftraits [Wright et al., 2004; Jones et al., 2012; Rogers, 2013]. These factors are not directly represented by therelatively simple LUE model logic due to limitations of available biophysical data required for modeldevelopment and regional simulations.

Factors constraining plant productivity can be divided into three general groups: constraints governingpotential carbon uptake (stressor factors, e.g., temperature and vapor pressure deficit (VPD)), inherent plantphysiological characteristics (plant functional types and traits, e.g., leaf nitrogen content), and landscapefeatures (e.g., terrain and microclimate regime). Environmental stressor factors such as minimum temperatureand VPD directly affect canopy stomatal conductance and photosynthetic carbon uptake. However, the othertwo factors are ecosystem properties that vary spatially and can influence LUE and ecosystem productivitywithin individual biomes. While stressor factors can affect photosynthesis at daily and finer time scales,the other two factors are assumed to be temporally conservative at coarse spatial scales and over limitedoperational satellite records.

When environmental stressor factors (such as temperature and water deficit) are not constraining tophotosynthetic carbon gain, then ecosystem optimal light use efficiency (LUEopt) can be estimated fromtower eddy covariance measurements of land-atmosphere carbon (CO2) exchange [Kergoat et al., 2008].Tower eddy covariance measurement networks, including Flux Network (FLUXNET) [Baldocchi et al., 2001],record CO2 fluxes and site-specific climate data, including incoming short wave radiation, and provide usefulinformation for validating ecosystem models and understanding terrestrial carbon budgets and underlyingenvironmental controls for different ecosystems [e.g., Running et al., 1999; Yi et al., 2013]. LUEopt is expectedto be spatially heterogeneous and lower than the theoretical maximum rate (LUEmax) due to other limitingecosystem morphological and landscape constraints.

The objective of this study is to improve the accuracy of satellite-based LUE model GPP predictions byimplementing a spatially explicit estimation of LUEopt. We apply a regression modeling approach usingspatially contiguous landscape attributes and in situ tower eddy covariance-based GPP values and supporting

Journal of Geophysical Research: Biogeosciences 10.1002/2014JG002709


biophysical measurements to estimate LUEopt over a North American domain. A selection of tower sitesrepresenting the major North American biomes is used for estimating LUEopt from multiyear observations ateach tower site. A geostatistical regression model is then developed to explain the across-site variability inLUEopt using a set of spatially contiguous predictor variables, including general plant traits and landscapefeatures. The LUEopt predictions are evaluated against other observation-based LUEopt values determinedfrom a set of independent tower validation sites. The LUEopt predictions are also used as primary inputsfor satellite (MODIS) LUE model-based GPP predictions over the domain. The model GPP results are thenevaluated against independent tower GPP data and baseline LUE model simulations derived using biome-specific constant LUEmax inputs.

2. Data and Methods2.1. In Situ LUEopt Estimation

Sixty-two tower sites from the FLUXNET La Thuile database [Baldocchi, 2008] were selected for this studyrepresenting major biomes of North America (see Table S1 in the supporting information). Most of the towersites selected had multiyear (≥2) daily values for GPP and surface meteorology, while only two sites had asingle year of measurements. The tower eddy covariance method intrinsically measures net ecosystem CO2exchange (NEE), while GPP is estimated by applying a model to the eddy covariance measurements topartition NEE into GPP and respiration components [Stoy et al., 2006; Lasslop et al., 2010]. Only the best qualitydaily GPP data determined from tower eddy covariance measured NEE were selected. These data denotedas having the best quality control flag (quality control = 1) are either original data or filled with highconfidence [Agarwal et al., 2010]. Daily global shortwave solar radiationmeasured at the tower sites was used toestimate PAR, which represents approximately half of the total incoming solar radiation [Zhao et al., 2005].

The tower LUEopt estimation was based on the assumption that GPP attains a maximum daily rate (definedas ≥98% of the long-term record) at some point over the multiyear measurement record, where LUE andcanopy photosynthesis are not limited by one ormore environmental stress factors, including light, temperature,or moisture limitations [Kergoat et al., 2008]. In order to avoid the effect of outliers, the higher 0.5% bin ofmeasurements was ignored. Thus, the upper 98–99.5% bin of daily GPP values throughout the measurementyears were sampled, representing the maximum estimated daily GPP level (GPPmax) from each site record.For all days with such criteria, LUE was defined as

LUE ¼ GPPmax= PAR � FPARð Þ (1)using PAR derived from tower incoming short wave solar radiation measurements and colocated MODISMOD15 FPAR (C5) retrievals [Myneni et al., 1999]. The MODIS 1 km resolution, 8 day FPAR data were obtainedfrom the Oak Ridge National Laboratory Distributed Active Archive Center (DAAC), where MODIS landproducts are available in a 7 × 7 km grid centered over individual tower sites. The FPAR data were sampledwithin 3×3 km pixels overlying the individual tower sites, and only the highest-quality values (main algorithmused with no saturation occurrence) were used. The FPAR data were spatially resampled using the MODIS1 km resolution global land cover product [Friedl et al., 2010] to ensure that all 3 × 3 km windows representedthe same land cover type as the local tower footprint. In order to capture the tower footprint, the 3×3 kmFPAR data were spatially averaged for each 8 day time step [Rahman, 2005], and temporal data gaps werefilled using the long-term MODIS FPAR 8 day climatology [Kandasamy et al., 2013]. In order to produce dailyFPAR data consistent with daily flux tower GPP values, the continuous 8 day FPAR record was interpolated to adaily time step using smoothing splines [Wahba, 1975]. The FPAR data for each of the 62 tower sites weretemporally matched with the local tower GPP and PAR observations.

The resultingmean daily estimated LUE in the definedmaximum threshold range fromequation (1) was used asthe LUEopt value for each tower site and was analyzed within and among the major representative NorthAmerican biome types represented by the MODIS land cover classification and selected FLUXNET tower sites.

2.2. Modeling LUEopt Patterns2.2.1. Explanatory Variables Influencing LUEopt PatternsIn order to explain the spatial variability of LUEopt, a set of geospatial environmental data characterizing generalcanopy traits, climate, and landscape terrain characteristics assumed to be important for ecosystem productivitywere considered (see Table 1). Plant traits were characterized by leaf nitrogen content, specific leaf area (SLA; leaf



area divided by foliar dry mass, m2 kg�1), and tree height. A leaf nitrogen map (Figure S1 in the supportinginformation) was developed using 27,116 global in situ observations of leaf nitrogen content per dry massavailable from a global plant trait database [Kattge et al., 2011]. Multiple observations at a given samplinglocation were averaged to obtain a single value for each location, resulting in 3700 global points from which287 points were within North America. These data were later interpolated over the spatial domain using universalkriging [Pebesma, 2004]. The SLA map (Figure S2 in the supporting information) was derived from the samedatabase and similar methodology but was determined from 154 sampling locations in North America out of1759 global points. Tree height data were obtained from a global 1 km resolution map, originally created fromsatellite light detection and ranging data from the Ice, Cloud and land Elevation Satellite [Simard et al., 2011].

Landscape features and terrain were characterized by elevation, aspect, and canopy cover geospatial data.Elevation was obtained from a global Shuttle Radar Topography Mission-based digital elevation map (DEM)with 1 km spatial resolution [Farr et al., 2007]. Terrain aspect was derived from the DEM and converted todimensionless east and north facing units ranging from �1 to 1, where flat terrain is 0, and 1 denotesmaximum eastward and northward aspects [Zar, 1999].

Selected MODIS land products spanning the satellite record from 2000 to 2006 including the MOD12Q1 IGBPstatic (2004) land cover classification [Friedl et al., 2010], MOD44 percent tree cover [Townshend et al., 2011], andMOD13A3 16day Enhanced Vegetation Index (EVI) products [Huete et al., 2002] were mosaicked to make asingle map for each product over the North American domain. The maximum EVI was produced using themaximum recorded values for a pixel over the period of record. Forest stand age was obtained from Pan et al.[2011], which provides forest age map products at 1 km resolution for Canada and the United States.

The long-term average precipitation and temperature, mean annual frozen season, and surface soil moisturewere used to characterize potential climate characteristics influencing LUEopt. Global temperature andprecipitation averages were obtained at 1 km spatial resolution from the WorldClim database; these data areinterpolated from 47,554 and 24,542 global weather stations for precipitation and temperature, respectively[Hijmans et al., 2005]. A global ecoregionmapwas obtained from theworldwildlife fund covering 867 land unitsof distinct biotas [Olson et al., 2001] and was used to define climate zones by aggregating temperature andprecipitation within each ecoregion.

Table 1. List of All the Data Sets Used for Predictive Modeling

Variables Abbreviation Geophysical Data Source

Landscape characteristics MODIS-MOD12 Land cover typesa,b Friedl et al. [2010]MODIS-MOD13 Maximum EVI Huete et al. [2002]MODIS-MOD44 Percent tree covera Townshend et al. [2011]Terrain DEM Elevation (m)

Aspect eastnessa

Aspect northness

Farr et al. [2007]

Climate Bio1 Annual temperature (°C) Hijmans et al. [2005]Bio12 Annual precipitation (mm)

NOAH GLDAS Soil moisturea(kgm�2) Rodell et al. [2004]Frozen days Average number of frozen days Kim and Kimball [2011]

and Kim et al. [2012]Plant traits SLAc Specific leaf area Kattge et al. [2011]

Leaf nitrogen contentc Leaf nitrogen per dry massa(%)Height Global tree height (m) Simard et al. [2011]

Others Ecoregions Global Ecoregions Olson et al. [2001]Stand Age North America stand age map Pan et al. [2011]

aDenotes the variables used in the final model.bLand cover types based on University of Maryland scheme includes: Water, Evergreen Needleleaf Forest, Evergreen

Broadleaf Forest, Deciduous Needleleaf Forest, Deciduous Broadleaf Forest, Mixed Forests, Closed Shrublands, OpenShrublands, Woody Savannas, Grasslands*, Croplands*, Urban and Built-Up, Barren and Sparsely Vegetated, andUnclassified.

c[Shipley, 1995, 2002; Cornelissen, 1996; Cornelissen et al., 1996, 2003, 2004; Atkin et al., 1997, 1999; Hickler, 1999;Medlynet al., 1999; Meziane and Shipley, 1999; Pyankov et al., 1999; Fonseca et al., 2000; Shipley and Lechowicz, 2000; Niinemets,2001; Shipley and Vu, 2002; Loveys et al., 2003;Ogaya and Peñuelas, 2003;Quested et al., 2003; Xu and Baldocchi, 2003; Díazet al., 2004;Wright et al., 2004, 2007; Craine et al., 2005, 2009;Han et al., 2005; Bakker et al., 2005, 2006; Kazakou et al., 2006;Preston et al., 2006; Cavender-Bares et al., 2006; Garnier et al., 2007; Campbell et al., 2007; Kleyer et al., 2008; Reich et al.,2008, 2009; van Bodegom et al., 2008; Fyllas et al., 2009; Kattge et al., 2009; Penuelas et al., 2009; Poorter, 2009; Freschetet al., 2010; Laughlin et al., 2010; Messier et al., 2010; Ordoñez et al., 2010; Onoda et al., 2011].



The mean annual frozen season (days) from 2000 to 2006 was derived from a consistent global classification ofdaily landscape freeze-thaw status derived from satellite passive microwave remote sensing [Kim and Kimball,2011; Kim et al., 2012] and distributed through the National Snow and Ice Data Center DAAC. Global monthlyaverage layer one soil moisture data were obtained from the Global Land Data Assimilation System(GLDAS), representing the outputs of four land surface models and extending from 1948 to 2010 [Rodellet al., 2004]. The aggregated monthly soil moisture data from GLDAS at 0.25° (~27 km) resolution wereaveraged and resampled to the baseline 1 km spatial resolution geographic projection of this investigationusing bilinear interpolation.2.2.2. Spatial Modeling of LUEoptA generalized additive model (GAM) [Hastie et al., 1986;Wood, 2006] was used to estimate the spatial variabilityof LUEopt across 45 FLUXNET tower sites having multiyear daily carbon flux measurements. The GAM describesthe relationship between two or more explanatory variables and the response variable (LUEopt) by fittingadditive and smoothed functions of those explanatory variables [Wood, 2006]. The GAM generally addsnonparametric smoothers to the parametric part of a generalized linear model (GLM). This model can provideimprovement over a GLM through the addition of appropriate smoothing functions [Guisan et al., 2002].The general model structure followed by Wood [2006] is

g μið Þ ¼ X�i θ þ f 1 X1ið Þ þ f 2 X2ið Þ þ f 3 X3ið Þ þ… (2)where μi ≡ E (Yi) and Yi follows an exponential family distribution. Yi is the response variable, Xi* is the ith rowof the model matrix, θ is the corresponding parameter vector so that Xi

*θ is a linear function of θ, and fi is theith smoothed function of X covariates.

The GAM was optimized using stepwise variable selection by means of Akaike information criterion [Burnhamand Anderson, 2002]. The spatially correlated residuals of the best model ε (s) (the stochastic part of the model)were explained using a semivariogram model [Pebesma, 2004; Hengl et al., 2009], based on the assumption thatflux towers with the same land cover type that are closer to each other are more likely to have similar LUEoptvalues. The spatial continuity between pairs of flux towers was first examined using the empirical semivariogram:

γ̂ hð Þ ¼ 12n hð Þ

Xi; jð Þjhij¼h yi � yj

� �2(3)

where n(h) is the number of pairs of points separated by vector h, hij= (h1ij , h2ij , …) is the vector oforientation between sites i and j, and yi is the observed response at the ith location. After calculation of thesemivariogram, an isotropic exponential variogram model was fitted

γ hð Þ ¼ 0 h ¼ 0aþ σ2 � að Þ 1� e�3j hj jj=r� � h ≠ 0

((4)

where a ≥ 0, σ2≥ a, and r ≥ 0; where h is the distance, a is the nugget effect, σ2 is the sill, and r is the range.

The full model (regression kriging) representing the stochastic and semivariogram deterministic terms wasthen used for spatial prediction of LUEopt at 1 km resolution over the North American domain using theselected spatially contiguous explanatory variables (represented in Table 1) and was cross validated usingtower-based LUEopt values from a set of 17 independent tower sites representingmajor North American biomes(For the location of test sites, refer to Figure S4 in the supporting information).

2.3. Gross Primary Production Modeling

In order to assess potential gain in GPP accuracy by using the spatially explicit LUEopt estimates, both thepredicted LUEopt and theMOD17 LUEmax values were used to estimate GPP and compared against all tower fluxmeasurement-based GPP records. The MOD17 algorithm [Running et al., 2004] was used for predicting GPP at adaily time step over a single year per tower site with the lowest fraction of missing values. Two sets of GPPsimulations were conducted as follows:

GPPLUEopt ¼ LUEopt � f VPD � f T� � � FPAR � PAR (5)

GPPLUEmax ¼ LUEmax � f VPD � f Tð Þ � FPAR � PAR (6)where f VPD and fTare vapor pressure deficit and temperature scalars that reduce LUEmax and LUEopt undersuboptimal conditions that vary by land cover type [Zhao et al., 2005]. The daily meteorological inputs



(VPD, T, and PAR) were derived from tower site level measurements, while daily FPAR was obtained fromthe MODIS MOD15 FPAR (C5) product. The MOD17 BPLUT for the VPD and T scalars is summarized inTable S3 in the supporting information.

3. Results3.1. LUEopt Estimation for Selected Tower Sites

LUEopt derived by applying the LUE model in equation (1) range from 0.28 to 2.82 (g CMJ�1) for tower sites

representing the different land cover types. Shrubland sites have the lowest LUEopt (0.508 ± 0.01), whilecropland sites have the highest LUEopt rates of the nine major North American land cover and plantfunctional types represented. The estimated tower LUEopt results show high spatial variability both withinand among biome types (Figures 1 and 2), while croplands have the highest range of LUEopt spatial variability.The LUEopt results are summarized for individual tower sites in Table S1 in the supporting information.

Comparison of average LUEopt values for each land cover type derived from this study with the biome-specificconstant LUEmax values used in the MODIS MOD17 LUE algorithm [Running et al., 2004] shows relatively largedifferences in these parameters for similar land cover types (Table 2). The MOD17 algorithm prescribes muchlower optimum LUE rates for croplands, deciduous broadleaf forest, grasslands, and mixed forest plantfunctional types than the estimated mean LUEopt rates inferred from the tower GPP values, implying thatMOD17 underestimates GPP for these land cover types (at least under near optimal conditions); the largestLUE differences occur for croplands, where the LUEopt results are approximately 2.5 times larger than MOD17.The prescribed MOD17 LUEmax rates are also lower than the lower 95th percentile of tower-estimated LUEoptrates for cropland (1.04 versus 1.77 g CMJ�1), grassland (0.86 versus 0.91 g CMJ�1), and deciduous broadleafforest (1.16 versus 1.29 g CMJ�1), which are three important biomes that cover ≈20% of the total area(and ≈1/3 of the vegetated area) of the study domain. In contrast, MOD17 LUEmax is approximately 2.5 timeshigher than the upper 95th percentile of LUEopt for closed shrublands (1.28 versus 0.53 g CMJ

�1).

Figure 1. Location of all FLUXNET tower sites used in this investigation and the relative magnitudes of estimated LUEopt(g CMJ�1) for each site, overlain on the MODIS-MOD17 maximum light use efficiency (LUEmax) values based on MODIS(MOD12Q1) IGBP land cover classes, including OSH (Open Shrubland), GRA (Grassland), CRO (Croplands), MF (Mixed Forest),DNF (Deciduous Needleleaf Forest), DBF (Deciduous Broadleaf Forest), SA (Savanna), WSA (Woody Savanna), EBF(Evergreen Broadleaf Forest), and CSH (Closed Shrubland).



3.2. LUEopt Prediction Overthe North American Domain

The best GAM used to derive LUEopt,based on its predictive power (root-mean-square error (RMSE)) inestimating LUEopt from the tower testsites, used land cover type (grasslandsand croplands), leaf nitrogen content,soil moisture, percent tree cover, andterrain aspect eastness coefficients.The model showed an adjusted R2 of68.2% with an associated RMSE of0.22 g CMJ�1 for the tower trainingsite-derived LUEopt values. The GAMused a linear relationship betweenall covariates except soil moisture,which showed a nonlinear relationship(Figure S6 in the supportinginformation) with tower observation-derived LUEopt. The full model(combination of GAM and krigingof residuals) increased the GAMaccuracy in explaining LUEopt spatialvariability among the tower trainingsites (R2 =88.6%, RMSE=0.15gCMJ�1)and validation sites (R2 = 91.1%, RMSEof 0.22 g CMJ�1), so that 88.6% ofthe variance across all sites (trainingand validation) was explained(RMSE= 0.17 g CMJ�1) (Figure 3). As

indicated by Kergoat et al. [2008], leaf nitrogen content was the most important factor explaining spatialvariability in LUEopt, accounting for 43% of the observed spatial variability across all tower sites (p< 0.0001)(Figure S5 in the supporting information). Surface soil moisture was the next most important factor,accounting for 19% (p< 0.0001) of LUEopt variability. Among land cover types, only grassland andcropland identifiers were significant in explaining LUEopt variability between sites (p< 0.01 andp< 0.0001, respectively) and were used in the fitted GAM (GAM coefficients of the covariates aresummarized in Table S2 in the supporting information). The resulting model predictions showed thatLUEopt for the three training cropland sites were underestimated by 11% compared to the tower-estimatedLUEopt for these sites.

Figure 2. Box plot of estimated tower LUEopt (gCMJ�1) representing different

land cover (MODIS MOD12Q1) types for 62 FLUXNET sites having multiyeardata records that were used in this study. Each symbol (black circles) representsan individual tower site grouped by land cover class; the numbers above eachbox plot denote the number of tower sites representing the specified landcover class, including OSH (Open Shrubland), ENF (Evergreen NeedleleafForest), CRO (Croplands), GRA (Grassland), MF (Mixed Forest), WSA (WoodySavanna), DBF (Deciduous Broadleaf Forest), EBF (Evergreen Broadleaf Forest),and CSH (Closed Shrubland); and the asterisk denotes MOD17 maximum lightuse efficiency (LUEmax) values (gCMJ

�1) for each land cover type.

Table 2. Comparison of Biome Average and Spatial Standard Deviation of Estimated LUEopt (g CMJ�1) From This Study

and the Prescribed Biome-Specific LUEmax Values Used in the MODIS MOD17 Operational Algorithm and GPP Product forNorth American Land Cover Types With the Greatest Proportional Coverage Over the Study Domain

Land Cover Area (%) LUEopt (This Study) LUEmax (MOD17)

Open Shrubland (OSH) 27.86 0.631 ± 0.37 0.841Evergreen Needleleaf Forest (ENF) 12.16 0.835 ± 0.24 0.962Cropland (CRO) 10.86 2.201 ± 0.66 1.044Grassland (GRA) 7.81 1.294 ± 0.32 0.86Mixed Forest (MF) 6.94 1.171 ± 0.23 1.051Woody Savanna (WSA) 5.53 0.983 ± 0.20 1.239Deciduous Broadleaf Forest (DBF) 1.98 1.453 ± 0.14 1.165Evergreen Broadleaf Forest (EBF) 1.50 0.980 ± 0.21 1.268Closed Shrubland (CSH) 0.50 0.508 ± 0.01 1.281



The predicted LUEopt map for theNorth American domain is defined at1 km spatial resolution consistent withthe MODIS land cover classificationinputs and shows that, as expected,areas of intense agriculture have thehighest regional LUEopt (Figure 4a).The model predictions also show thatthe largest relative prediction error isassociated with regions where therewere only a limited number of towersites available for model development(Figure 4b). These areas are generallycropland regions (Figure 1).

3.3. GPP Modeling Improvements

Predicted LUEopt at the tower sites wasused as an ancillary input to theMOD17 LUE algorithm in place of aprescribed LUEmax for each biome typeto estimate daily GPP at each site; theGPP simulations were conducted ateach site over one specific year havingthe longest available tower GPP record.The resulting daily GPP estimates

derived using LUEopt (GPPLUEopt) estimationwere evaluated against independent tower GPP values and daily GPPestimates determined using prescribed LUEmax constants (GPPLUEmax) from the MOD17 algorithm. Tower sitecomparisons between LUEopt and MOD17 LUEmax and the resulting annual GPP simulations are summarized inFigures 5a and 5b. These results indicate that the use of a spatially explicit LUEopt input dramatically improvesGPP estimation accuracy relative to the tower data and baseline MOD17 calculations derived from biome-prescribed LUEmax constants. A list of sites used for validation and associated summary of annual GPP results for

Figure 3. The relationship between model-predicted and tower-estimatedLUEopt (g CMJ

�1) for all FLUXNET sites represented in this study. The pre-dicted LUEopt values account for 88.6% (R

2) of spatial variability in towerGPP, with associated RMSE differences of 0.17 g CMJ�1.

Figure 4. (a) Predicted map of LUEopt (g CMJ�1) for the North American domain, encompassing all vegetated land areas. Croplands have the highest estimated

LUEopt, with generally lower LUEopt levels at higher latitudes and elevations. Gray shading denotes mean LUEopt variability by latitude and longitude; latitudinalmeans range from 0.2 to 1.6 (g CMJ�1), and longitudinal means range from 0.2 to 1.2 (g CMJ�1). (b) Map of the model standard error as a proportion (percentage)of the predicted LUEopt; greater relative model uncertainty occurs over regions with sparse tower data.



each tower site and model are presented in Table S4 in the supporting information. The RMSE differencebetween GPPLUEopt and the tower data was 257gCm

�2 yr�1 for all 62 tower sites, which represented a 41%improvement in model performance relative to the baseline MOD17 GPPLUEmax calculations (RMSE:439gCm�2 yr�1). The model results for the independent tower validation sites also indicate that the LUEoptinputs provide more accurate simulations of observed annual GPP spatial variability compared to the baselineMOD17 simulations, accounting for more than 4 times asmuch observed GPP variability (R2 of 64.2% versus R2 of14.8%). For both sets of simulations, we used local tower meteorological data, including incoming shortwavesolar radiation, air temperature, and MODIS FPAR inputs, and the only differences were the baseline LUEmax andmodel-predicted LUEopt inputs to the LUE model GPP calculations.

The model GPP results show that for all land cover types represented, the predicted LUEopt inputs produce moreaccurate predictions of ecosystemGPP (Table 3) relative to the use of prescribed biome-specific LUEmax constantsin the baseline MOD17 simulations. The seasonal progression of estimated daily GPP derived from alternative

Figure 5. Comparison of annual GPP (g Cm�2 yr�1) estimated from daily LUE model simulations using (a) LUEopt and (b) baseline-prescribed LUEmax inputs relativeto tower GPP from all the 62 tower sites and 17 independent tower test sites (inset); the symbols denote the dominant land cover type of each tower site, whilesymbol colors denote the magnitude of LUEopt and LUEmax used to estimate GPP.

Table 3. Comparison Between Modeled Gross Primary Production and Tower Eddy Covariance Measurement-Based GPPfor the North American Land Cover Types Represented in This Studya

GPP-LUEopt GPP-LUEmax

Land cover type R2d RMSEc mean residual error (MRE)b R2d RMSEc MREb

Open Shrubland (OSH) 0.34 177 �139 0.09 299 142Evergreen Needle Leaf Forest (ENF) 0.60 308 �19 0.39 412 144Cropland (CRO) 0.75 233 202 0.80 733 �555Grassland (GRA) 0.93 158 �99 0.94 270 �307Mixed Forest (MF) 0.72 238 24 0.08 399 �124Woody Savanna (WSA) - 20 �17 - 325 270Deciduous Broadleaf Forest (DBF) 0.66 223 �61 0.52 406 �380Evergreen Broadleaf Forest (EBF) - 443 �304 - 567 145Closed Shrubland (CSH) - 245 �243 - 600 331

aFor all land cover types, GPP is modeled using LUEopt inputs and shows improvements over the baseline MOD17simulations derived using prescribed LUEmax inputs.bMRE (mean residual error).

cRMSE (root-mean-square error).dThe hyphen (-) denotes only two tower sites representing the land cover type.



LUEopt and LUEmax inputs is presented in Figure 6 for four selected tower sites representing the major NorthAmerican biome types, based on having the largest aerial coverage within the domain and representingshrubland, evergreen needleleaf forest, cropland, and grassland biome types. The prescribed LUEmax inputs fromthe MOD17 operational algorithm are generally higher than the average estimated LUEopt values for closedshrubland and evergreen needleleaf forest and lower for cropland and grasslands, while LUEopt levels from thisinvestigation are generally independent of land cover type and exhibit large spatial variability. Overall, the use ofalternative LUEopt inputs leads to 58.5% lower RMSE differences and 42.9% higher daily R

2 correspondenceagainst the tower data and relative to the LUEmax-based GPP simulations at the tower sites.

4. Discussion

The tower analysis showed that LUEopt is spatially heterogeneous within individual land cover types andacross the landscape. The largest LUEopt variability was within croplands, which also showed the highestLUEopt compared to other land cover types. Previous studies have shown that LUEmax is underestimatedfor Maize and Soybean in the MODIS MOD17 algorithm [Xin et al., 2013], and a cropland field study showedthat LUEmax of Maize can be as high as 3.84 g CMJ

�1 [Lindquist et al., 2005].

Spatial modeling of tower-estimated LUEopt facilitated GAM-based estimation of LUEopt regional patterns acrossNorth America. Among the factors influencing LUEopt patterns, percent tree cover as a surrogate for vegetatedarea was positively associated with LUEopt. Terrain aspect as a surrogate for solar illumination andmicroclimate(air temperature and precipitation) has the lowest impacts as predictor variables in our model. Our resultsalso highlight the importance of soil moisture and leaf nitrogen content in determining LUEopt patterns. Leaf

Figure 6. Seasonal comparison of tower daily GPP and corresponding model-simulated GPP using baseline-prescribedLUEmax inputs and alternative LUEopt inputs to the MOD17 LUE algorithm for selected FLUXNET validation sites representingCSH (Closed Shrubland), ENF (Evergreen Needleleaf Forest), CRO (Croplands), and GRA (Grassland) land cover types. Fornorthern latitudes and shrublands, MOD17 LUEmax is higher than LUEopt, resulting in GPPLUEmax overestimation of GPPfor CSH and ENF. For CRO and GRA, GPPLUEmax is lower than GPPLUEopt and tower GPP. GPPLUEopt shows overall betteraccuracy than GPPLUEmax against the tower GPP records.



nitrogen content, because of its relationship with rubisco and photosynthetic capacity, is an important factordetermining plant productivity. The availability of large databases for plant traits [Kattge et al., 2011] provided aunique opportunity to include leaf nitrogen content data in the LUEopt prediction model. Leaf traits such asleaf nitrogen content and SLA can be temporally dynamic in response to seasonal canopy changes, standage, and disturbance recovery [Parolin et al., 2002; Nouvellon et al., 2010], affecting canopy photosyntheticcapacity, light use efficiency, and GPP, and others have shown that canopy N can explain a substantialamount of variation in productivity from minute to year scales [Kergoat et al., 2008; Ollinger et al., 2008; Reich,2012]. Our investigation focus was on estimating spatial patterns in LUEopt using limited ground observationsof general leaf traits, while the effects of temporal leaf trait variations on LUEopt and associated GPPcalculations were not explicitly represented and require further investigation. Some attempts have beenmade to infer LUE by estimating leaf nitrogen or chlorophyll content using remote sensing [e.g., Grace et al.,2007; Goerner et al., 2009; Frankenberg et al., 2011; Schlemmer et al., 2013], which is expected to improveglobal ecosystem productivity estimation. However, the general methods and leaf trait data used in this study,including leaf nitrogen, show promising results for spatially continuous estimation of LUEopt, and improvedestimation of GPP spatial and temporal dynamics.

The GAM results showed that a static land cover classification was only useful when grassland and croplandclasses were considered as independent predictors of LUEopt spatial variability. The covariates used in this studywere also generally insufficient for explaining LUEopt spatial variability within cropland and grassland areas duein part to sparse tower representation of grassland and cropland heterogeneity. These results also imply theneed for testing other covariates, including temporally dynamic land cover (e.g., maize and soybean rotation)and irrigation regime inputs, as potential explanatory variables in these areas. However, despite the fact that ourmodel underestimated LUEopt for these sites (Figure 3), GPP estimation accuracy was significantly improvedusing estimated LUEopt compared to baseline LUEmax inputs to the MOD17 algorithm.

In this study, we used a limited number of predictor variables reported to have an important role in the carbonuptake capacity of ecosystems to predict LUEopt at regional scales. Other factors such as groundwater storageand soil chemistry also influence plant photosynthesis andmay provide additional landscape characteristics forestimating LUEopt and productivity. Additionally, uncertainty associated with the covariates used in the spatialregression model likely has a negative impact on LUEopt and GPP accuracy. The model predictions andvalidation activities from this studywere also derived from a relatively sparse North American tower observationnetwork that may not fully capture the range of variability in regional vegetation and climate patterns. Despitethese limitations and uncertainties, our model results produce regional patterns of LUEopt with favorableaccuracy that enhanced the accuracy of higher-order GPP simulations from a satellite data-driven LUEalgorithm. These findings and the global availability of similar plant traits information and geospatial datarequired for model extrapolation imply the potential for similar global mapping of LUEopt utilizing moreextensive plant trait and tower (FLUXNET) measurement records available from global networks andspanning a broader range of global biomes. New global biophysical data from next generation satellitesensors may also lead to better LUEopt and GPP predictions; these new observations include canopyfluorescence, landscape freeze-thaw, and soil moisture dynamics from the NASA Orbiting CarbonObservatory 2 and Soil Moisture Active Passive missions that may provide near direct measures of LUE andunderlying environmental controls. LUEopt also likely varies temporally with changes in vegetation andenvironmental conditions, whereas this study only provides static map of LUEopt from limited towerobservations and spatially coarse geophysical data. Future research and new satellite observations mayenable temporal modeling of LUEopt while also considering land cover change and disturbance recoveryimpacts. This is especially important for cropland regions with annual rotation of C4 and C3 crops.

Here our primary focus was on predicting optimal light use efficiency, while LUEopt predictions was used forenhancing GPP estimation accuracy relative to using prescribed biome LUEmax constants in the MOD17 LUEmodel. The MOD17 algorithm for GPP modeling only accounts for VPD and air temperature as direct dailystressor factors. Even though GPPLUEopt showed improved accuracy over the GPPLUEmax estimates andrelative to the tower GPP data, the remaining unexplained variance between the model predictions andtower GPP values imply that additional environmental factors are needed to further improve model accuracy(Figure 6); the application of additional dynamic stressor factors such as freeze-thaw and soil moisture statusmay improve the seasonality of LUE-modeled GPP [Kimball et al., 2009].



In summary, our results showed that the spatial variability in LUEopt between sites and within individualbiome types should not be ignored. The LUEopt predictions from this study led to a 42.9% improvement inLUE model and tower GPP correspondence. The LUEopt retrievals show large spatial variability that is largelyindependent from the static land cover classification data, and even though the current study domain waslimited to North America, the potential exists for extrapolating these methods to a global domain using alarger global tower network [Baldocchi, 2008] and plant traits database [Kattge et al., 2011]. Spatially explicitLUEopt data derived from general landscape characteristics and plant traits information, and associatedimprovements in GPP estimation accuracy, should promote better understanding of terrestrial carbon sinksand sources and biospheric capacity for mitigation of the human carbon footprint.

ReferencesAgarwal, D. A., M. Humphrey, N. F. Beekwilder, K. R. Jackson, M. M. Goode, and C. Van Ingen (2010), A data-centered collaboration portal to

support global carbon-flux analysis, Concurrency Computat.: Pract. Exper., 22(17), 2323–2334, doi:10.1002/cpe.1600.Ahl, D. E., S. T. Gower, D. S. Mackay, S. N. Burrows, J. M. Norman, and G. R. Diak (2004), Heterogeneity of light use efficiency in a northern

Wisconsin forest: Implications for modeling net primary production with remote sensing, Remote Sens. Environ., 93(1–2), 168–178,doi:10.1016/j.rse.2004.07.003.

Atkin, O. K., M. Schortemeyer, N. McFarlane, and J. R. Evans (1999), The response of fast- and slow-growing Acacia species to elevated atmosphericCO2: An analysis of the underlying components of relative growth rate, Oecologia, 120(4), 544–554, doi:10.1007/s004420050889.

Atkin, O., M. Westbeek, M. L. Cambridge, H. Lambers, and T. L. Pons (1997), Leaf respiration in light and darkness (a comparison of slow-andfast-growing Poa species), Plant Physiol., 113, 961–965.

Bakker, C., J. Rodenburg, and P. M. van Bodegom (2005), Effects of Ca- and Fe-rich Seepage on P Availability and Plant Performance inCalcareous Dune Soils, Plant Soil, 275(1–2), 111–122, doi:10.1007/s11104-005-0438-1.

Bakker, C., P. M. Van Bodegom, H. J. M. Nelissen, W. H. O. Ernst, and R. Aerts (2006), Plant responses to rising water tables and nutrientmanagement in calcareous dune slacks, Plant Ecol., 185(1), 19–28, doi:10.1007/s11258-005-9080-5.

Baldocchi, D. (2008), TURNER REVIEWNo. 15. ’Breathing’ of the terrestrial biosphere: Lessons learned from a global network of carbon dioxideflux measurement systems, Aust. J. Bot., 56(1), 1–26.

Baldocchi, D., et al. (2001), FLUXNET: A new tool to study the temporal and spatial variability of ecosystem-scale carbon dioxide, water vapor,and energy flux densities, Bull. Am. Meteorol. Soc., 82, 2415–2434, doi:10.1175/1520-0477(2001).

Bartlett, D. S., G. J. Whiting, and J. M. Hartman (1989), Use of vegetation indices to estimate indices to estimate intercepted solar radiationand net carbon dioxide exchange of a grass canopy, Remote Sens. Environ., 30(2), 115–128.

Beer, C., et al. (2010), Terrestrial gross carbon dioxide uptake: Global distribution and covariation with climate, Science, 329(5993), 834–8,doi:10.1126/science.1184984.

Burnham, K. P., andD. R. Anderson (2002),Model Selection andMultimodel Inference: A Practical Information-Theoretic Approach, Springer, New York.Campbell, C., L. Atkinson, J. Zaragoza-Castells, M. Lundmark, O. Atkin, and V. Hurry (2007), Acclimation of photosynthesis and respiration is

asynchronous in response to changes in temperature regardless of plant functional group, New Phytol., 176(2), 375–389, doi:10.1111/j.1469-8137.2007.02183.x.

Cavender-Bares, J., A. Keen, and B. Miles (2006), Phylogenetic structure of floridian plant communities depends on taxonomic and spatialscale, Ecology, 87(7), S109–S122, doi:10.2307/20069161.

Cornelissen, J. (1996), An experimental comparison of leaf decomposition rates in a wide range of temperate plant species and types, J. Ecol.,84(4), 573–582.

Cornelissen, J. H. C., B. Cerabolini, P. Castro-Díez, P. Villar-Salvador, G. Montserrat-Martí, J. P. Puyravaud, M. Maestro, M. J. A. Werger, andR. Aerts (2003), Functional traits of woody plants: Correspondence of species rankings between field adults and laboratory-grown seedlings?,J. Veg. Sci., 14(3), 311–322, doi:10.1111/j.1654-1103.2003.tb02157.x.

Cornelissen, J. H. C., H. M. Quested, D. Gwynn-jones, R. S. P. Van logtestijn, M. A. H. De beus, A. Kondratchuk, T. V. Callaghan, and R. Aerts (2004),Leaf digestibility and litter decomposability are related in a wide range of subarctic plant species and types, Funct. Ecol., 18(6), 779–786,doi:10.1111/j.0269-8463.2004.00900.x.

Cornelissen, J., P. Diez, and R. Hunt (1996), Seedling growth, allocation and leaf attributes in a wide range of woody plant species and types,J. Ecol., 84(5), 755–765.

Craine, J. M., W. G. Lee, W. J. Bond, R. J. Williams, and L. C. Johnson (2005), Environmental constraints on a global relationship among leaf androot traits of grasses, Ecology, 86(1), 12–19, doi:10.1890/04-1075.

Craine, J. M., et al. (2009), Global patterns of foliar nitrogen isotopes and their relationships with climate, mycorrhizal fungi, foliar nutrientconcentrations, and nitrogen availability, New Phytol., 183(4), 980–992.

Díaz, S., et al. (2004), The plant traits that drive ecosystems: Evidence from three continents, J. Veg. Sci., 15(3), 295–304, doi:10.1658/1100-9233(2004)015[0295:TPTTDE]2.0.CO;2.

Farr, T., P. Rosen, and E. Caro (2007), The shuttle radar topography mission, Rev. Geophys., 45, RG2004, doi:10.1029/2005RG000183.Fonseca, C. R., J. M. Overton, B. Collins, and M. Westoby (2000), Shifts in trait-combinations along rainfall and phosphorus gradients, J. Ecol.,

88(6), 964–977, doi:10.1046/j.1365-2745.2000.00506.x.Frankenberg, C., et al. (2011), New global observations of the terrestrial carbon cycle from GOSAT: Patterns of plant fluorescence with gross

primary productivity, Geophys. Res. Lett., 38, L17706, doi:10.1029/2011GL048738.Freschet, G. T., J. H. C. Cornelissen, R. S. P. van Logtestijn, and R. Aerts (2010), Evidence of the “plant economics spectrum” in a subarctic flora,

J. Ecol., 98(2), 362–373, doi:10.1111/j.1365-2745.2009.01615.x.Friedl, M. A., D. Sulla-Menashe, B. Tan, A. Schneider, N. Ramankutty, A. Sibley, and X. Huang (2010), MODIS Collection 5 global land cover:

Algorithm refinements and characterization of new datasets, Remote Sens. Environ., 114(1), 168–182, doi:10.1016/j.rse.2009.08.016.Fyllas, N. M., et al. (2009), Basin-wide variations in foliar properties of Amazonian forest: Phylogeny, soils and climate, Biogeosciences, 6,

2677–2708, doi:10.5194/bg-6-2677-2009.Garnier, E., et al. (2007), Assessing the effects of land-use change on plant traits, communities and ecosystem functioning in grasslands:

A standardized methodology and lessons from an application to 11 European sites, Ann. Bot., 99(5), 967–85, doi:10.1093/aob/mcl215.

AcknowledgmentsThis study was conducted with fundingprovided by the National Aeronauticsand Space Administration (NASA) EarthScience program (NNX11AD46G,NNX09AP52G). The study has beensupported by the TRY initiative on planttraits (http://www.try-db.org). The TRYinitiative and database is hosted, devel-oped, and maintained by J. Kattge andG. Bönisch (Max Planck Institute forBiogeochemistry, Jena, Germany). TRYis/has been supported by DIVERSITAS,IGBP, the Global Land Project, the UKNatural Environment Research Councilthrough its program QUEST (Quantifyingand Understanding the Earth System),the French Foundation for BiodiversityResearch, and GIS “Climat Environnementet Société,” France. This work used eddycovariance data acquired by the FLUXNETcommunity and in particular by thefollowing networks: AmeriFlux (U.S.Department of Energy, Biological andEnvironmental Research, TerrestrialCarbon Program (DE-FG02-04ER63917and DE-FG02-04ER63911)), Fluxnet-Canada (supported by CFCAS, NSERC,BIOCAP, Environment Canada andNRCan). We acknowledge the financialsupport to the eddy covariance dataharmonization provided by CarboEuropeIP,FAO-GTOS-TCO, iLEAPS, Max PlanckInstitute for Biogeochemistry, NationalScience Foundation, University ofTuscia, Université Laval, EnvironmentCanada and U.S. Department of Energyand the database development andtechnical support from Berkeley WaterCenter, Lawrence Berkeley NationalLaboratory, Microsoft Research eScience,Oak Ridge National Laboratory,University of California–Berkeley andthe University of Virginia.



http://dx.doi.org/10.1002/cpe.1600http://dx.doi.org/10.1016/j.rse.2004.07.003http://dx.doi.org/10.1007/s004420050889http://dx.doi.org/10.1007/s11104-005-0438-1http://dx.doi.org/10.1007/s11258-005-9080-5http://dx.doi.org/10.1175/1520-0477(2001)http://dx.doi.org/10.1126/science.1184984http://dx.doi.org/10.1111/j.1469-8137.2007.02183.xhttp://dx.doi.org/10.1111/j.1469-8137.2007.02183.xhttp://dx.doi.org/10.2307/20069161http://dx.doi.org/10.1111/j.1654-1103.2003.tb02157.xhttp://dx.doi.org/10.1111/j.0269-8463.2004.00900.xhttp://dx.doi.org/10.1890/04-1075http://dx.doi.org/10.1658/1100-9233(2004)015[0295:TPTTDE]2.0.CO;2http://dx.doi.org/10.1658/1100-9233(2004)015[0295:TPTTDE]2.0.CO;2http://dx.doi.org/10.1029/2005RG000183http://dx.doi.org/10.1046/j.1365-2745.2000.00506.xhttp://dx.doi.org/10.1029/2011GL048738http://dx.doi.org/10.1111/j.1365-2745.2009.01615.xhttp://dx.doi.org/10.1016/j.rse.2009.08.016http://dx.doi.org/10.5194/bg-6-2677-2009http://dx.doi.org/10.1093/aob/mcl215http://www.try-db.org

Goerner, A., M. Reichstein, and S. Rambal (2009), Tracking seasonal drought effects on ecosystem light use efficiency with satellite-based PRIin a Mediterranean forest, Remote Sens. Environ., 113(5), 1101–1111, doi:10.1016/j.rse.2009.02.001.

Goetz, S. J., and S. D. Prince (1996), Remote sensing of net primary production in boreal forest stands, Agric. For. Meteorol., 78(3–4), 149–179,doi:10.1016/0168-1923(95)02268-6.

Gower, S. T., C. J. C. Kucharik, and J. J. M. Norman (1999), Direct and Indirect Estimation of Leaf Area Index, f APAR, and Net Primary Productionof Terrestrial Ecosystems, Remote Sens. Environ., 4257(99), 29–51.

Grace, J., C. Nichol, M. Disney, P. Lewis, T. Quaife, P. Bowyer, C. Building, W. M. Road, and G. Street (2007), Can we measure terrestrialphotosynthesis from space directly, using spectral reflectance and fluorescence?, Global Change Biol., 13(7), 1484–1497, doi:10.1111/j.1365-2486.2007.01352.x.

Guisan, A., T. C. Edwards, and T. Hastie (2002), Generalized linear and generalized additive models in studies of species distributions: Settingthe scene, Ecol. Modell., 157(2–3), 89–100.

Han, W., J. Fang, D. Guo, and Y. Zhang (2005), Leaf nitrogen and phosphorus stoichiometry across 753 terrestrial plant species in China,New Phytol., 168(2), 377–385, doi:10.1111/j.1469-8137.2005.01530.x.

Hastie, T., R. Tibshirani, T. Hastie, and R. Tibshirani (1986), Generalized additive models, Stat. Sci., 1(3), 297–318.Heinsch, F. A., et al. (2006), Evaluation of remote sensing based terrestrial productivity from MODIS using regional tower eddy flux network

observations, IEEE Trans. Geosci. Remote Sens., 44(7), 1908–1925, doi:10.1109/TGRS.2005.853936.Hengl, T., H. Sierdsema, A. Radović, A. Dilo, and A. Radovic (2009), Spatial prediction of species’ distributions from occurrence-only records:

Combining point pattern analysis, ENFA and regression-kriging, Ecol. Modell., 220(24), 3499–3511, doi:10.1016/j.ecolmodel.2009.06.038.Hickler, T. (1999), Plant functional types and community characteristics along environmental gradients on Öland’s Great Alvar (Sweden), Lund.Hijmans, R. J., S. E. Cameron, J. L. Parra, P. G. Jones, A. Jarvis, S. E. Hijmans, J. L. Cameron, P. G. J. Parra, and R. J. A. Jarvis (2005), Very high

resolution interpolated climate surfaces for global land areas, Int. J. Climatol., 25(15), 1965–1978, doi:10.1002/joc.1276.Hilker, T., N. C. Coops, M. A. Wulder, T. A. Black, and R. D. Guy (2008), The use of remote sensing in light use efficiency based models

of gross primary production: A review of current status and future requirements, Sci. Total Environ., 404(2–3), 411–23, doi:10.1016/j.scitotenv.2007.11.007.

Huete, A., K. Didan, T. Miura, E. P. Rodriguez, X. Gao, and L. G. Ferreira (2002), Overview of the radiometric and biophysical performance of theMODIS vegetation indices, Remote Sens. Environ., 83(1–2), 195–213, doi:10.1016/s0034-4257(02)00096-2.

Huston, M. A., and S. Wolverton (2009), The global distribution of net primary production: Resolving the paradox, Ecol. Monogr., 79(3), 343–377,doi:10.1890/08-0588.1.

Jin, C., X. Xiao, L. Merbold, A. Arneth, E. Veenendaal, and W. L. Kutsch (2013), Phenology and gross primary production of two dominantsavanna woodland ecosystems in Southern Africa, Remote Sens. Environ., 135, 189–201, doi:10.1016/j.rse.2013.03.033.

Jones, M. O., J. S. Kimball, L. A. Jones, and K. C. McDonald (2012), Satellite passive microwave detection of North America start of season,Remote Sens. Environ., 123, 324–333, doi:10.1016/j.rse.2012.03.025.

Kandasamy, S., F. Baret, A. Verger, P. Neveux, and M. Weiss (2013), A comparison of methods for smoothing and gap filling time series ofremote sensing observations – application to MODIS LAI products, Biogeosciences, 10(6), 4055–4071, doi:10.5194/bg-10-4055-2013.

Kattge, J., W. Knorr, T. Raddatz, and C. Wirth (2009), Quantifying photosynthetic capacity and its relationship to leaf nitrogen content forglobal-scale terrestrial biosphere models, Global Change Biol., 15(4), 976–991, doi:10.1111/j.1365-2486.2008.01744.x.

Kattge, J., et al. (2011), TRY - a global database of plant traits, Global Change Biol., 17(9), 2905–2935, doi:10.1111/j.1365-2486.2011.02451.x.Kazakou, E., D. Vile, B. Shipley, C. Gallet, and E. Garnier (2006), Co-variations in litter decomposition, leaf traits and plant growth in species

from a Mediterranean old-field succession, Funct. Ecol., 20(1), 21–30, doi:10.1111/j.1365-2435.2006.01080.x.Kergoat, L., S. Lafont, A. Arneth, V. Le Dantec, and B. Saugier (2008), Nitrogen controls plant canopy light-use efficiency in temperate and

boreal ecosystems, J. Geophys. Res., 113, G04017, doi:10.1029/2007JG000676.Kim, Y., and J. Kimball (2011), Developing a global data record of daily landscape freeze/thaw status using satellite passive microwave

remote sensing, IEEE Trans. Geosci. Remote Sens., 49(3), 949–960, doi:10.1109/TGRS.2010.2070515.Kim, Y., J. S. Kimball, K. Zhang, and K. C. McDonald (2012), Satellite detection of increasing Northern Hemisphere non-frozen seasons from

1979 to 2008: Implications for regional vegetation growth, Remote Sens. Environ., 121, 472–487, doi:10.1016/j.rse.2012.02.014.Kimball, J. S., M. Zhao, K. C. McDonald, and S. W. Running (2006), Satellite remote sensing of terrestrial net primary production for the pan-Arctic

basin and Alaska, Mitig. Adapt. Strateg. Global Change., 11(4), 783–804, doi:10.1007/s11027-005-9014-5.Kimball, J. S., L. A. Jones, Z. Ke, F. A. Heinsch, K. C. McDonald, and W. C. Oechel (2009), A satellite approach to estimate land – Atmosphere

CO2 exchange for Boreal and Arctic Biomes using MODIS and AMSR-E, IEEE Trans. Geosci. Remote Sens., 47(2), 569–587, doi:10.1109/tgrs.2008.2003248.

Kleyer, M., et al. (2008), The LEDA Traitbase: A database of life-history traits of the Northwest European flora, J. Ecol., 96(6), 1266–1274,doi:10.1111/j.1365-2745.2008.01430.x.

Kumar, M., and J. L. Monteith (1981), Remote sensing of plant growth, in Plants and the Daylight Spectrum, edited by P. A. Huxley, pp. 133–144,Academic Press, London.

Lasslop, G., M. Reichstein, D. Papale, A. D. Richardson, A. Arneth, A. Barr, P. Stoy, and G. Wohlfahrt (2010), Separation of net ecosystemexchange into assimilation and respiration using a light response curve approach: Critical issues and global evaluation, Global ChangeBiol., 16(1), 187–208, doi:10.1111/j.1365-2486.2009.02041.x.

Laughlin, D. C., J. J. Leppert, M. M. Moore, and C. H. Sieg (2010), A multi-trait test of the leaf-height-seed plant strategy scheme with 133 speciesfrom a pine forest flora, Funct. Ecol., 24(3), 493–501, doi:10.1111/j.1365-2435.2009.01672.x.

Li, S.-G., W. Eugster, J. Asanuma, A. Kotani, G. Davaa, D. Oyunbaatar, and M. Sugita (2008), Response of gross ecosystem productivity, lightuse efficiency, and water use efficiency of Mongolian steppe to seasonal variations in soil moisture, J. Geophys. Res., 113, G01019,doi:10.1029/2006JG000349.

Lindquist, J. L., T. J. Arkebauer, D. T. Walters, K. G. Cassman, and A. Dobermann (2005), Maize radiation use efficiency under optimal growthconditions, Agronomy, 97, 72–78.

Loveys, B., L. Atkinson, D. Sherlock, R. Roberts, A. Fitter, and O. Atkin (2003), Thermal acclimation of leaf and root respiration: An investigationcomparing inherently fast-and slow-growing plant species, Global Change Biol., 9(6), 895–910, doi:10.1046/j.1365-2486.2003.00611.x.

Malhi, Y. (2012), The productivity, metabolism and carbon cycle of tropical forest vegetation, J. Ecol., 100(1), 65–75, doi:10.1111/j.1365-2745.2011.01916.x.

Medlyn, B. E., et al. (1999), Effects of elevated [CO2] on photosynthesis in European forest species: A meta-analysis of model parameters,Plant. Cell Environ., 22(12), 1475–1495, doi:10.1046/j.1365-3040.1999.00523.x.

Messier, J., B. J. McGill, and M. J. Lechowicz (2010), How do traits vary across ecological scales? A case for trait-based ecology, Ecol. Lett., 13(7),838–48, doi:10.1111/j.1461-0248.2010.01476.x.



http://dx.doi.org/10.1016/j.rse.2009.02.001http://dx.doi.org/10.1016/0168-1923(95)02268-6http://dx.doi.org/10.1111/j.1365-2486.2007.01352.xhttp://dx.doi.org/10.1111/j.1365-2486.2007.01352.xhttp://dx.doi.org/10.1111/j.1469-8137.2005.01530.xhttp://dx.doi.org/10.1109/TGRS.2005.853936http://dx.doi.org/10.1016/j.ecolmodel.2009.06.038http://dx.doi.org/10.1002/joc.1276http://dx.doi.org/10.1016/j.scitotenv.2007.11.007http://dx.doi.org/10.1016/j.scitotenv.2007.11.007http://dx.doi.org/10.1016/s0034-4257(02)00096-2http://dx.doi.org/10.1890/08-0588.1http://dx.doi.org/10.1016/j.rse.2013.03.033http://dx.doi.org/10.1016/j.rse.2012.03.025http://dx.doi.org/10.5194/bg-10-4055-2013http://dx.doi.org/10.1111/j.1365-2486.2008.01744.xhttp://dx.doi.org/10.1111/j.1365-2486.2011.02451.xhttp://dx.doi.org/10.1111/j.1365-2435.2006.01080.xhttp://dx.doi.org/10.1029/2007JG000676http://dx.doi.org/10.1109/TGRS.2010.2070515http://dx.doi.org/10.1016/j.rse.2012.02.014http://dx.doi.org/10.1007/s11027-005-9014-5http://dx.doi.org/10.1109/tgrs.2008.2003248http://dx.doi.org/10.1109/tgrs.2008.2003248http://dx.doi.org/10.1111/j.1365-2745.2008.01430.xhttp://dx.doi.org/10.1111/j.1365-2486.2009.02041.xhttp://dx.doi.org/10.1111/j.1365-2435.2009.01672.xhttp://dx.doi.org/10.1029/2006JG000349http://dx.doi.org/10.1046/j.1365-2486.2003.00611.xhttp://dx.doi.org/10.1111/j.1365-2745.2011.01916.xhttp://dx.doi.org/10.1111/j.1365-2745.2011.01916.xhttp://dx.doi.org/10.1046/j.1365-3040.1999.00523.xhttp://dx.doi.org/10.1111/j.1461-0248.2010.01476.x

Meziane, D., and B. Shipley (1999), Interacting components of interspecific relative growth rate: Constancy and change under differing conditionsof light and nutrient supply, Funct. Ecol., 13(5), 611–622.

Monteith, J., and C.Moss (1977), Climate and the efficiency of crop production in Britain, Philos. Trans. R. Soc. London B. Biol. Sci., 281(980), 277–294.Monteith, J. L. (1972), Solar radiation and productivity in tropical ecosystems, J. Appl. Ecol., 9(3), 747–766.Myneni, R. B., et al. (1999), MODIS Leaf Area Index (LAI) And Fraction Of Photosynthetically Active Radiation Absorbed By Vegetation (FPAR)

Product (MOD15), edited by Y. Zhang, NASA Goddard Sp. Flight Cent., 121, 126, October.Niinemets, Ü. (2001), Global-scale climatic controls of leaf dry mass per area, density, and thickness in trees and shrubs, Ecology, 82(2), 453–469,

doi:10.1890/0012-9658(2001).Nouvellon, Y., et al. (2010), Within-stand and seasonal variations of specific leaf area in a clonal Eucalyptus plantation in the Republic of Congo,

For. Ecol. Manage., 259(9), 1796–1807, doi:10.1016/j.foreco.2009.05.023.Ogaya, R., and J. Peñuelas (2003), Comparative field study of Quercus ilex and Phillyrea latifolia: Photosynthetic response to experimental

drought conditions, Environ. Exp. Bot., 50(2), 137–148, doi:10.1016/S0098-8472(03)00019-4.Ollinger, S. V., et al. (2008), Canopy nitrogen, carbon assimilation, and albedo in temperate and boreal forests: Functional relations and potential

climate feedbacks, Proc. Natl. Acad. Sci. U.S.A., 105(49), 19,336–41, doi:10.1073/pnas.0810021105.Olson, D. M., et al. (2001), Terrestrial ecoregions of the World: A newmap of life on Earth, BioScience, 51(11), 933–938, doi:10.1641/0006-3568

(2001)051[0933:TEOTWA]2.0.C.Onoda, Y., et al. (2011), Global patterns of leaf mechanical properties, Ecol. Lett., 14(3), 301–12, doi:10.1111/j.1461-0248.2010.01582.x.Ordoñez, A. J. C., P. M. Van Bodegom, J. M. Witte, P. Ruud, J. C. Ordoñez, P. M. van Bodegom, R. P. Bartholomeus, H. F. van Dobben, and

R. Aerts (2010), Leaf habit and woodiness regulate different leaf economy traits at a given nutrient supply, Ecology, 91(11), 3218–3228,doi:10.1890/09-1509.1.

Pan, Y., R. Birdsey, J. Hom, K. Mccullough, and K. Clark (2006), Improved estimates of net primary productivity from Modis Satellite Data atRegional and Local Scales, Ecol. Soc. Am., 16(1), 125–132.

Pan, Y., J. M. Chen, R. Birdsey, K. McCullough, L. He, and F. Deng (2011), Age structure and disturbance legacy of North American forests,Biogeosciences, 8(3), 715–732, doi:10.5194/bg-8-715-2011.

Parolin, P., N. Armbrüster, and W. J. Junk (2002), Seasonal changes of leaf nitrogen content in trees of Amazonian floodplains, Acta Amazônica,32(2), 231–240.

Pebesma, E. J. (2004), Multivariable geostatistics in S: The gstat package, Comput. Geosci., 30(7), 683–691, doi:10.1016/j.cageo.2004.03.012.Penuelas, J., J. Sardans, J. Llusià, S. M. Owen, J. Carnicer, T. W. Giambelluca, E. L. Rezende, M. Waite, and Ü. Niinemets (2009), Faster returns on

“leaf economics” and different biogeochemical niche in invasive compared with native plant species, Global Change Biol., 16(8), 2171–2185,doi:10.1111/j.1365-2486.2009.02054.x.

Poorter, L. (2009), Leaf traits show different relationships with shade tolerance in moist versus dry tropical forests, New Phytol., 181(4), 890–900,doi:10.1111/j.1469-8137.2008.02715.x.

Potter, C., J. Randerson, and C. Field (1993), Terrestrial ecosystem production: A process model based on global satellite and surface data,Global Biogeochem. Cycles, 7(4), 811–841, doi:10.1029/93GB02725.

Preston, K., W. Cornwell, and J. DeNoyer (2006), Wood density and vessel traits as distinct correlates of ecological strategy in 51 Californiacoast range angiosperms, New Phytol., 170(4), 807–18, doi:10.1111/j.1469-8137.2006.01712.x.

Pyankov, V., A. Kondratchuk, and B. Shipley (1999), Leaf structure and specific leaf mass: The alpine desert plants of the Eastern Pamirs Tadjikistan,New Phytol., 143, 131–142.

Quested, H. M., J. H. C. Cornelissen, M. C. Press, T. V. Callaghan, R. Aerts, F. Trosien, P. Riemann, D. Gwynn-Jones, A. Kondratchuk, andS. E. Jonasson (2003), Decomposition of sub-arctic plants with differing nitrogen economies: A functional role for hemiparasites,Ecology, 84(12), 3209–3221, doi:10.1890/02-0426.

Rahman, A. F. (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.

Reich, P. B. (2012), Key canopy traits drive forest productivity, Proc. Biol. Sci., doi:10.1098/rspb.2011.2270, (January).Reich, P. B., M. G. Tjoelker, K. S. Pregitzer, I. J. Wright, J. Oleksyn, and J.-L. Machado (2008), Scaling of respiration to nitrogen in leaves, stems

and roots of higher land plants, Ecol. Lett., 11(8), 793–801, doi:10.1111/j.1461-0248.2008.01185.x.Reich, P. B., J. Oleksyn, and I. J. Wright (2009), Leaf phosphorus influences the photosynthesis-nitrogen relation: A cross-biome analysis of

314 species, Oecologia, 160(2), 207–12, doi:10.1007/s00442-009-1291-3.Rodell, M., et al. (2004), The Global Land Data Assimilation System, Bull. Am. Meteorol. Soc., 85(3), 381–394, doi:10.1175/BAMS-85-3-381.Rogers, A. (2013), The use and misuse of V c,max in Earth System Models, Photosynth. Res., 119(1–2), 15–29, doi:10.1007/s11120-013-9818-1.Ruimy, A., B. Saugier, and G. Dedieu (1994), Methodology for the estimation of terrestrial net primary production from remotely sensed data,

J. Geophys. Res., 99(D3), 5263–5283, doi:10.1029/93JD03221.Ruimy, A., G. Dedieu, and B. Saugier (1996), TURC: A diagnostic model of continental gross primary productivity and net primary productivity,

Global Biogeochem. Cycles, 10(2), 269–285, doi:10.1029/96GB00349.Running, S. W., R. R. Nemani, F. A. Heinsch, M. Zhao, M. Reeves, and H. Hashimoto (2004), A Continuous Satellite-Derived Measure of Global

Terrestrial Primary Production, BioScience, 54(6), 547, doi:10.1641/0006-3568(2004)054[0547:ACSMOG]2.0.CO;2.Running, S., et al. (1999), A global terrestrial monitoring network integrating tower fluxes, flask sampling, ecosystemmodeling and EOS data,

Remote Sens. Environ., 70, 108–127.Schlemmer, M., A. Gitelson, J. Schepers, R. Ferguson, Y. Peng, J. Shanahan, and D. Rundquist (2013), Remote estimation of nitrogen and

chlorophyll contents in maize at leaf and canopy levels, Int. J. Appl. Earth Obs. Geoinf., 25, 47–54, doi:10.1016/j.jag.2013.04.003.Shipley, B. (1995), Structured interspecific determinants of specific leaf area in 34 species of herbaceous angiosperms, Funct. Ecol., 9(2), 312–319.Shipley, B. (2002), Trade-offs between net assimilation rate and specific leaf area in determining relative growth rate: Relationship with daily

irradiance, Funct. Ecol., 16(5), 682–689, doi:10.1046/j.1365-2435.2002.00672.x.Shipley, B., and M. Lechowicz (2000), The functional co-ordination of leaf morphology, nitrogen concentration, and gas exchange in 40 wetland

species, Ecoscience, 7(2), 183–194.Shipley, B., and T.-T. Vu (2002), Dry matter content as a measure of dry matter concentration in plants and their parts, New Phytol., 153(2),

359–364, doi:10.1046/j.0028-646X.2001.00320.x.Simard, M., N. Pinto, J. B. Fisher, and A. Baccini (2011), Mapping forest canopy height globally with spaceborne lidar, J. Geophys. Res., 116,

G04021, doi:10.1029/2011JG001708.Stoy, P. C., G. G. Katul, M. B. S. Siqueira, J.-Y. Juang, K. A. Novick, J. M. Uebelherr, and R. Oren (2006), An evaluation of models for partitioning

eddy covariance-measured net ecosystem exchange into photosynthesis and respiration, Agric. For. Meteorol., 141(1), 2–18,doi:10.1016/j.agrformet.2006.09.001.



http://dx.doi.org/10.1890/0012-9658(2001)http://dx.doi.org/10.1016/j.foreco.2009.05.023http://dx.doi.org/10.1016/S0098-8472(03)00019-4http://dx.doi.org/10.1073/pnas.0810021105http://dx.doi.org/10.1641/0006-3568(2001)051[0933:TEOTWA]2.0.Chttp://dx.doi.org/10.1641/0006-3568(2001)051[0933:TEOTWA]2.0.Chttp://dx.doi.org/10.1111/j.1461-0248.2010.01582.xhttp://dx.doi.org/10.1890/09-1509.1http://dx.doi.org/10.5194/bg-8-715-2011http://dx.doi.org/10.1016/j.cageo.2004.03.012http://dx.doi.org/10.1111/j.1365-2486.2009.02054.xhttp://dx.doi.org/10.1111/j.1469-8137.2008.02715.xhttp://dx.doi.org/10.1029/93GB02725http://dx.doi.org/10.1111/j.1469-8137.2006.01712.xhttp://dx.doi.org/10.1890/02-0426http://dx.doi.org/10.1029/2005GL024127http://dx.doi.org/10.1098/rspb.2011.2270http://dx.doi.org/10.1111/j.1461-0248.2008.01185.xhttp://dx.doi.org/10.1007/s00442-009-1291-3http://dx.doi.org/10.1175/BAMS-85-3-381http://dx.doi.org/10.1007/s11120-013-9818-1http://dx.doi.org/10.1029/93JD03221http://dx.doi.org/10.1029/96GB00349http://dx.doi.org/10.1641/0006-3568(2004)054[0547:ACSMOG]2.0.CO;2http://dx.doi.org/10.1016/j.jag.2013.04.003http://dx.doi.org/10.1046/j.1365-2435.2002.00672.xhttp://dx.doi.org/10.1046/j.0028-646X.2001.00320.xhttp://dx.doi.org/10.1029/2011JG001708http://dx.doi.org/10.1016/j.agrformet.2006.09.001

Townshend, J. R. G., M. Carroll, C. Dimiceli, R. Sohlberg, M. Hansen, and R. DeFries (2011), Vegetation Continuous Fields MOD44B, 2001Percent Tree Cover, Collection 5, Univ. of Maryland.

Trapani, N., A. J. Hall, V. O. Sadras, and F. Vilella (1992), Ontogenetic changes in radiation use efficiency of sunflower (Helianthus annuus L.)crops, F. Crop. Res., 29(4), 301–316, doi:10.1016/0378-4290(92)90032-5.

Turner, D. P., S. T. Gower, W. B. Cohen, M. Gregory, and T. K. Maiersperger (2002), Effects of spatial variability in light use efficiency on satellite-based NPP monitoring, Remote Sens. Environ., 80(3), 397–405, doi:10.1016/S0034-4257(01)00319-4.

Van Bodegom, P. M., B. K. Sorrell, A. Oosthoek, C. Bakker, and R. Aerts (2008), Separating the effects of partial submergence and soil oxygendemand on plant physiology, Ecology, 89(1), 193–204, doi:10.1890/07-0390.1.

Veroustraete, F., H. Sabbe, and H. Eerens (2002), Estimation of carbon mass fluxes over Europe using the C-Fix model and Euroflux data,Remote Sens. Environ., 83(3), 376–399, doi:10.1016/S0034-4257(02)00043-3.

Wahba, G. (1975), Smoothing noisy data with spline functions, Numer. Math., 24, 383–393.Wang, H., G. Jia, C. Fu, J. Feng, T. Zhao, and Z. Ma (2010), Deriving maximal light use efficiency from coordinated flux measurements and

satellite data for regional gross primary production modeling, Remote Sens. Environ., 114(10), 2248–2258, doi:10.1016/j.rse.2010.05.001.Way, J., et al. (2005), Site-level evaluation of satellite-based global terrestrial gross primary production and net primary production monitoring,

Global Change Biol., 11(4), 666–684, doi:10.1111/j.1365-2486.2005.00936.x.Wood, S. N. (2006), Generalized Additive Models: An Introduction With R, Chapman and Hall/CRC, Boca Raton, Fla.Wright, I. J., et al. (2004), The worldwide leaf economics spectrum, Nature, 428(6985), 821–7, doi:10.1038/nature02403.Wright, I. J., et al. (2007), Relationships among ecologically important dimensions of plant trait variation in seven neotropical forests, Ann. Bot.,

99(5), 1003–15, doi:10.1093/aob/mcl066.Xiao, X., Q. Zhang, S. Saleska, L. Hutyra, P. De Camargo, S. Wofsy, S. Frolking, S. Boles, M. Keller, and B. Moore Iii (2005), Satellite-based

modeling of gross primary production in a seasonally moist tropical evergreen forest, Remote Sens. Environ., 94(1), 105–122,doi:10.1016/j.rse.2004.08.015.

Xin, Q., P. Gong, C. Yu, L. Yu, M. Broich, A. Suyker, and R. Myneni (2013), A Production Efficiency Model-Based Method for Satellite Estimates ofCorn and Soybean Yields in the Midwestern US, Remote Sens., 5(11), 5926–5943, doi:10.3390/rs5115926.

Xu, L., and D. D. Baldocchi (2003), Seasonal trends in photosynthetic parameters and stomatal conductance of blue oak (Quercus douglasii)under prolonged summer drought and high temperature, Tree Physiol., 23(13), 865–77.

Yi, Y., J. S. Kimball, L. A. Jones, R. H. Reichle, R. Nemani, and H. A. Margolis (2013), Recent climate and fire disturbance impacts on boreal andarctic ecosystem productivity estimated using a satellite-based terrestrial carbon fluxmodel, J. Geophys. Res. Biogeosciences, 118, 606–622,doi:10.1002/jgrg.20053.

Yuan, W., et al. (2007), Deriving a light use efficiency model from eddy covariance flux data for predicting daily gross primary productionacross biomes, Agric. For. Meteorol., 143(3–4), 189–207, doi:10.1016/j.agrformet.2006.12.001.

Zar, J. H. (1999), Biostatistical Analysis, Prentice Hall, New Jersey.Zhao, M., and S. W. Running (2010), Drought-induced reduction in global terrestrial net primary production from 2000 through 2009, Science,

329(5994), 940–3, doi:10.1126/science.1192666.Zhao, M., F. A. Heinsch, R. R. Nemani, and S. W. Running (2005), Improvements of the MODIS terrestrial gross and net primary production

global data set, Remote Sens. Environ., 95(2), 164–176, doi:10.1016/j.rse.2004.12.011.



http://dx.doi.org/10.1016/0378-4290(92)90032-5http://dx.doi.org/10.1016/S0034-4257(01)00319-4http://dx.doi.org/10.1890/07-0390.1http://dx.doi.org/10.1016/S0034-4257(02)00043-3http://dx.doi.org/10.1016/j.rse.2010.05.001http://dx.doi.org/10.1111/j.1365-2486.2005.00936.xhttp://dx.doi.org/10.1038/nature02403http://dx.doi.org/10.1093/aob/mcl066http://dx.doi.org/10.1016/j.rse.2004.08.015http://dx.doi.org/10.3390/rs5115926http://dx.doi.org/10.1002/jgrg.20053http://dx.doi.org/10.1016/j.agrformet.2006.12.001http://dx.doi.org/10.1126/science.1192666http://dx.doi.org/10.1016/j.rse.2004.12.011

Appendix S1

Table S1. List of all flux data used for modeling ecosystem’s optimum light use efficiency (LUEopt) (g C MJ-1). Site ID with * represents the sites used for testing the model.

Site.ID Country Longitude Latitude IGBP Start.Year Stop.Year Data.Years LUEopt Ref CA-Ca3 Canada -124.9 49.53462 ENF 2001 2005 5 1.007128 [1] CA-Gro Canada -82.1556 48.2167 MF 2003 2005 3 1.072686 [2] CA-Let Canada -112.94 49.7093 GRA 2000 2005 6 1.3845 [1] CA-NS2 Canada -98.5247 55.90583 ENF 2001 2005 5 0.652424 [3] CA-Qcu Canada -74.0365 49.2671 ENF 2001 2006 6 0.544343 [1] CA-SF1 Canada -105.818 54.485 ENF 2003 2005 3 0.876917 [4] CA-SJ3 Canada -104.645 53.8758 ENF 2004 2005 2 0.567476 [1] CA-TP4 Canada -80.3574 42.7098 ENF 2003 2005 3 1.241333 [1] CA-WP1 Canada -112.467 54.9538 MF 2003 2005 3 0.820705 [5] US-Arc USA -98.0406 35.54649 GRA 2005 2006 2 1.633346 [6] US-Aud USA -110.51 31.5907 GRA 2002 2006 5 1.516681 [7] US-Bar USA -71.2881 44.0646 DBF 2004 2005 2 1.294878 [7] US-Bkg USA -96.8362 44.3453 GRA 2004 2006 3 0.92746 [8] US-Blo USA -120.633 38.8952 ENF 2000 2006 7 0.795183 [7] US-Bo1 USA -88.2904 40.0062 CRO 2000 2007 8 1.771081 [9] US-Fmf USA -111.727 35.1426 ENF 2005 2006 2 1.088077 - US-FPe USA -105.102 48.3077 GRA 2000 2006 7 0.94503 [9] US-FR2 USA -97.9962 29.9495 GRA 2004 2006 3 1.156695 [10] US-Goo USA -89.8735 34.2547 CRO 2002 2006 5 1.281304 [7] US-Ha1 USA -72.1715 42.5378 MF 2000 2006 7 1.594431 [11] US-Ho1 USA -68.7402 45.2041 MF 2000 2004 5 1.237908 [11] US-Ho2 USA -68.747 45.2091 MF 2000 2004 5 1.143624 [12] US-IB1 USA -88.2227 41.8593 CRO 2005 2007 3 2.370859 [13] US-Ivo USA -155.75 68.48647 OSH 2003 2006 4 0.282539 [8] US-KS2 USA -80.6715 28.6086 EBF 2000 2006 7 1.132858 [10] US-Los USA -89.9751 46.07917 MF 2001 2005 5 0.868477 [14] US-LPH USA -72.185 42.5419 DBF 2002 2005 4 1.374309 [10] US-Me1 USA -121.505 44.57917 ENF 2004 2005 2 0.803781 [15] US-MMS USA -86.4131 39.3231 DBF 2000 2005 6 1.334926 [16] US-MOz USA -92.2 38.7441 DBF 2004 2006 3 1.522771 [7] US-Ne3 USA -96.4397 41.1797 CRO 2001 2005 5 2.758718 [8] US-NR1 USA -105.546 40.0329 ENF 2000 2003 4 0.670666 [7] US-Oho USA -83.8438 41.5545 DBF 2004 2005 2 1.730531 [17] US-SO3 USA -116.623 33.3772 OSH 2000 2006 7 0.58047 [10] US-SP3 USA -82.1633 29.7548 EBF 2000 2004 5 0.827336 [7] US-SRM USA -110.866 31.8214 OSH 2004 2006 3 1.031525 [18] US-Syv USA -89.3477 46.242 MF 2002 2006 5 1.208443 [7]

Site.ID Country Longitude Latitude IGBP Start.Year Stop.Year Data.Years LUEopt Ref US-Ton USA -120.966 38.4316 WSA 2001 2006 6 0.840432 [10] US-UMB USA -84.7138 45.5598 DBF 2000 2003 4 1.387933 [7] US-Var USA -120.951 38.4133 WSA 2001 2006 6 1.126614 [7] US-WCr USA -90.0799 45.8059 DBF 2000 2006 7 1.545098 [7] US-Wi4 USA -91.1663 46.7393 MF 2002 2005 4 1.282347 [19] US-Wkg USA -109.942 31.7365 GRA 2004 2006 3 1.376162 [20] US-Wrc USA -121.952 45.8205 ENF 2000 2006 7 0.83209 [7] US-Dk3 USA -79.0942 35.9782 MF 2001 2005 5 1.316723 [7] CA-Ca2* Canada -125.291 49.8705 ENF 2000 2005 6 0.623031 [1] US-Wi7* USA -91.0589 46.64583 GRA 2005 2005 1 0.917994 - US-Wi8* USA -91.2462 46.72083 DBF 2002 2002 1 1.441195 - US-Me4* USA -121.622 44.4992 ENF 2000 2000 1 0.651052 [21] CA-TP3* Canada -80.3483 42.7068 ENF 2003 2005 3 1.14903 [22] US-ARb* USA -98.0402 35.5497 GRA 2005 2006 2 1.792967 [2] US-Me3* USA -121.608 44.3154 ENF 2004 2005 2 1.168232 [13] US-SO2* USA -116.623 33.3739 CSH 2000 2006 7 0.495293 [21] CA-Ca1* Canada -125.334 49.86725 ENF 2000 2005 6 1.319532 [23] CA-NS3* Canada -98.3822 55.9117 ENF 2001 2005 5 0.61117 [24] CA-NS5* Canada -98.485 55.8631 ENF 2001 2005 5 0.696321 [24] CA-Ojp* Canada -104.692 53.9163 ENF 2000 2005 6 0.577636 [1] US-Me2* USA -121.557 44.4523 ENF 2003 2005 3 1.08697 [7] US-Ne1* USA -96.4766 41.1651 CRO 2001 2005 5 2.827378 [8] US-Bn1* USA -145.378 63.9198 ENF 2003 2003 1 0.775621 [25] CA-SF3* Canada -106.005 54.0916 CSH 2003 2005 3 0.520906 [26] CA-Obs* Canada -105.118 53.9872 ENF 2000 2005 6 0.636964 [1]

Note: IGBP is based on the MODIS MOD12 land cover type 2 classification for year 2003. CRO (Croplands); CSH (Closed Shrublands); DBF (Deciduous Broadleaf Forest); EBF (Evergreen Broadleaf Forest); ENF (Evergreen Needleleaf Forest); GRA (Grassland); MF (Mixed Forest); OSH (Open Shrublands); WSA (Woody Savannas).

Figure S1. Map of the predicted leaf nitrogen content per dry mass (g g-1). The interpolation model can explain 47% of the variance among 287 data measurement points (adjusted R2 = 0.47).

Figure S2. Map of the predicted potential leaf area per dry mass (SLA, m2 kg-1). The interpolation model can explain 43% of the variance among 154 data points.

Figure S3. Scatterplot matrix showing relations between optimum light use efficiency (LUEopt, g C MJ-1), leaf nitrogen content per dry mass (%) and specific leaf area (SLA, m2 kg-1). Leaf nitrogen content shows a linear relationship with LUEopt; dashed lines denote the standard errors.

Figure S4. Location of all 62 flux tower sites including 17 sites used for testing the geospatial model.

Figure S5. Relationship between leaf nitrogen content and optimum light use efficiency. Leaf nitrogen can explain 43% of the spatial variance among flux tower sites; grey shading denotes standard errors.

Figure S6. Components of the fitted GAM model showing the relationship of each independent variable to tower observed LUEopt. Dashed lines denote the standard errors.

Figure S7. The response of soil water content (kg m-2) and leaf nitrogen content per dry mass (%) on the fitted LUEopt model (g C MJ-1).

Figure S8.The fitted GAM model diagnostic plots. The upper left plot shows that the data distribution is close to normality. The upper right plot indicates that with increasing mean, variance is approximately constant. The histogram of the residuals are close to normal distribution. The lower right plot shows the relationship between the fitted and measured values.

Table S2. List of estimated mean (Est, g C MJ-1), standard error (SE) and p values for the coefficients of the fitted generalized additive model (GAM) indicating that leaf nitrogen content and soil moisture are the most important prediction factors for LUEopt, followed by cropland land cover type, percent tree cover, grassland cover type, and aspect eastness in relative importance.

Parameter Est SE P value Intercept -1.758570 0.238923 1.18e-08

Nitrogen 0.815683 0.115470 2.86e-08 GRA 0.259882 0.105598 0.018834 CRO 0.590601 0.139332 0.000152 Percent tree 0.006143 0.001683 0.000835

Eastness 0.075437 0.041209 0.075533

Note: The smoothed soil moisture term showed 3.361 estimated degrees of freedom with p-value of 2.54e-05.

Figure S9. The variogram fitted to the residuals of the final GAM model with logged link for LUEopt showing the semivariance between the pairs modeled within a 300 km range. The nugget value used for the model is 0.01 with sill of 0.035.

Table S3. Biome property lookup table (BPLUT) for the MODIS MOD17 GPP algorithm.

Note: CRO (Croplands); CSH (Closed Shrublands); DBF (Deciduous Broadleaf Forest); EBF (Evergreen Broadleaf Forest); ENF (Evergreen Needleleaf Forest); GRA (Grassland); MF (Mixed Forest); OSH (Open shrublands); WSA (Woody Savannas). VPD (Pa) and T (˚C) dimensionless scalars range from 0-1. The T constraint is 1 (no constraint) when daily minimum air temperature is above Tmax and 0 (fully constrained) when air temperature is below Tmin. The VPD constraint is 0 above VPDmax and 1 below VPDmin. For values between min and max, T and VPD scalars are estimated using: fT= T - Tmin / Tmax -Tmin fVPD = 1- (VPD – VPDmin ) / (VPDmax – VPDmin)

Scalar ENF EBF DBF MF CSH OSH WSA SAV GRA CRO Tmin -8.00 -8.00 -6.00 -7.00 -8.00 -8.00 -8.00 -8.00 -8.00 -8.00 Tmax 8.31 9.09 9.94 9.50 8.61 8.80 11.39 11.39 12.02 12.02 VPDmin 650 800 650 650 650 650 650 650 650 650 VPDmax 4600 3100 1650 2400 4700 4800 3200 3100 5300 4300

Table S4. Annual sum of daily GPP (g C m2 yr-1) modeled using the MOD17 LUE model and baseline prescribed biome specific LUEmax (g C MJ-1) inputs and new spatially explicit LUEopt (g C MJ-1) inputs for the tower (FLUXNET) observation sites. * represents the validation sites.

Site.ID year number of Obs. days Tower Obs. GPP GPP (LUEmax) GPP (LUEopt) CA-Ca3 2004 362 1441.92 1496.6 1532.62 CA-Gro 2004 299 923.64 1111.99 1191.63 CA-Let 2003 283 662.05 474.79 687.58 CA-NS2 2004 207 442.42 642.11 424.65 CA-Qcu 2003 290 310.25 581.24 319.95 CA-SF1 2004 179 721.31 809.7 701.5 CA-SJ3 2005 295 552.76 999.76 638.58 CA-TP4 2004 321 1333.22 1097.7 1428.19 CA-WP1 2005 326 769.94 1188.93 908.08 US-Arc 2006 252 861.79 676.07 1140.81 US-Aud 2005 255 350.46 370.55 553.87 US-Bar 2005 264 1072.61 1044.55 1274.65 US-Bkg 2005 336 749.06 535.09 846.98 US-Blo 2003 357 1020.5 1528.56 1256.35 US-Bo1 2005 337 1525.16 727.96 1308.36 US-Fmf 2006 252 708.64 907.6 993.69 US-FPe 2005 297 507.54 347.96 414.77 US-FR2 2005 364 1397.05 1014.54 1661.71 US-Goo 2004 350 1743.61 1324.82 1625.96 US-Ha1 2003 296 1516.14 1052.52 1526.37 US-Ho1 2000 366 1599.04 1196.37 1338.12 US-Ho2 2003 316 1329.83 1052.31 1248.72 US-IB1 2006 229 1360.98 478.95 937.44 US-Ivo 2004 295 165 497.75 154.27 US-KS2 2003 289 1686.32 1791.12 1669.01 US-Los 2003 216 518.69 914.15 816.32 US-LPH 2003 326 1208.79 1084.86 1259.07 US-Me1 2004 202 268.21 434.06 359.57 US-MMS 2004 251 1568.

publications · 2020. 2. 18. · improving ecosystem productivity modeling through spatially...

Documents