using lidar and quickbird data to model plant production and quantify uncertainties associated with...

Remote Sensing of Environment 113 (2009) 2366–2379

Contents lists available at ScienceDirect

Remote Sensing of Environment

j ourna l homepage: www.e lsev ie r.com/ locate / rse

Using LiDAR and quickbird data to model plant production and quantify uncertaintiesassociated with wetland detection and land cover generalizations

Bruce D. Cook a,⁎, Paul V. Bolstad a, Erik Næsset b, Ryan S. Anderson c, Sebastian Garrigues e,Jeffrey T. Morisette f, Jaime Nickeson g, Kenneth J. Davis h

a NASA Goddard Space Flight Center Biospheric Sciences Branch, Code 614.4 Greenbelt, MD, 20771, USAb Department of Ecology and Natural Resource Management (INA) Norwegian University of Life Sciences (UMB) P.O.Box 5003, NO-1432 Ås, Norwayc College of Forestry and Conservation The University of Montana 437 Science Complex, Missoula, MT, 59812, USAe Service “Analyse et Produits Image”, CNES DCT/SI/AP 18 avenue Edouard Belin-BPI 1219 31409 Toulouse Cedex 4, Francef U.S. Geological Survey Fort Collins Science Center 2150 Centre Ave., Building C, Fort Collins, CO, 80526, USAg NASA Goddard Space Flight Center Terrestrial Information Systems Branch, Code 614.5 Greenbelt, MD, 20771, USAh Department of Meteorology The Pennsylvania State University 503 Walker Building, University Park, PA, 16802-5013, USA

⁎ Corresponding author. NASA Goddard Space FlightSystems Branch, Code 614.5 Greenbelt, MD, 20771, USA. Tel614 6695.

E-mail address: [email protected] (B.D. Cook).

0034-4257/$ – see front matter. Published by Elsevierdoi:10.1016/j.rse.2009.06.017

a b s t r a c t
a r t i c l e i n f o
Article history:Received 28 May 2008Received in revised form 22 June 2009Accepted 28 June 2009

Keywords:Primary productionLeaf area index (LAI)Light-use efficiencyCarbon-use efficiencyModerate Resolution ImagingSpectroradiometer (MODIS)Digital hemispheric photographyEddy covariance

Spatiotemporal data from satellite remote sensing and surface meteorology networks have made it possible tocontinuously monitor global plant production, and to identify global trends associated with land cover/use andclimate change. Gross primary production (GPP) and net primary production (NPP) are routinely derived from theModerate Resolution Imaging Spectroradiometer (MODIS) onboard satellites Terra and Aqua, and estimatesgenerally agree with independentmeasurements at validation sites across the globe. However, the accuracy of GPPandNPP estimates in some regionsmay be limited by the quality ofmodel input variables and heterogeneity at finespatial scales. We developed new methods for deriving model inputs (i.e., land cover, leaf area, andphotosynthetically active radiation absorbed by plant canopies) from airborne laser altimetry (LiDAR) andQuickbird multispectral data at resolutions ranging from about 30 m to 1 km. In addition, LiDAR-derived biomasswas used as a means for computing carbon-use efficiency. Spatial variables were used with temporal data fromground-based monitoring stations to compute a six-year GPP and NPP time series for a 3600 ha study site in theGreat Lakes region of North America. Model results compared favorably with independent observations from a400 m flux tower and a process-based ecosystem model (BIOME-BGC), but only after removing vapor pressuredeficit as a constraint on photosynthesis from theMODIS global algorithm. Fine-resolution inputs capturedmore ofthe spatial variability, but estimateswere similar to coarse-resolution datawhen integrated across the entirelandscape. Failure to account for wetlands had little impact on landscape-scale estimates, becausevegetation structure, composition, and conversion efficiencies were similar to upland plant communities.Plant productivity estimates were noticeably improved using LiDAR-derived variables, while uncertaintiesassociated with land cover generalizations and wetlands in this largely forested landscape were consideredless important.

Published by Elsevier Inc.

1. Introduction

Recent advances in remote sensing with light detection and ranging(LiDAR) have provided natural resource scientists and practitioners withan unprecedented opportunity to derive height, biomass and three-dimensional structural attributes of plant communities across large,heterogeneous landscapes (e.g., Nelson et al., 2004; Næsset, 2004; Lefskyet al., 2005). Apowerful extensionof this technology is the fusionof LiDARandmultispectral datasets to characterize the structure, composition, and

Center Terrestrial Information.:+1 301 614 6689; fax:+1 301

Inc.

functional attributes of terrestrial vegetation (e.g., Popescu et al., 2004;Coops et al., 2004). Merging structural data from LiDAR and spectralinformation frommultispectral sensors simplifies landcover classificationusing schemes such as the IGBP (International Geosphere-BiosphereProgramme),whosebroadvegetative classes are definedby the fractionalcover of trees and shrubs and percentage of evergreen and deciduousfoliage (Loveland et al., 2000; Thomlinson et al., 1999). Fine spatialresolution multispectral imagery (e.g., Quickbird, IKONOS) is particularlyuseful for evaluating uncertainties that may exist in coarse resolutionglobal satellite products (Morisette et al., 2003; Steinberget al., 2006), andto verify the underlying theory and performance of algorithms that areused to derive these products (e.g., Chen et al., 2004). In addition, fine-resolution data can be used to determine the appropriate resolution ofinput variables that are needed to achieve accuracy at either stand- orregional-scales (e.g., Potter et al., 2007; Ahl et al., 2005).

mailto:[email protected]

http://dx.doi.org/10.1016/j.rse.2009.06.017

http://www.sciencedirect.com/science/journal/00344257

2367B.D. Cook et al. / Remote Sensing of Environment 113 (2009) 2366–2379

The Moderate Resolution Imaging Spectroradiometer (MODIS) is amoderate-resolution (≥250 m) multispectral sensor onboard NASA'searth observing satellites, Terra and Aqua. We demonstrate here howseveral land products derived from MODIS can be evaluated using fine-resolution LiDAR and Quickbird data. Among these derived products areland cover; leaf area index (LAI); fractional photosynthetically activeradiation absorbed by vegetation (fPAR, 400 to 700 nm); gross primaryproduction (GPP), i.e., carbon fixed by photosynthesis; and net primaryproduction (NPP), i.e., conversion of fixed carbon to plant biomass. Landcover, LAI and fPAR are upstream products that are used to compute GPPand NPP through the use of climate-constrained light- and carbon-useefficiency models (Running et al., 2004). Several studies have demon-strated strong relationships between LiDARdata and LAI based on canopygap fraction (Morsdorf et al., 2006; Solberg et al., 2006; Thomas et al.,2006;Riañoet al., 2004; Lovell et al., 2003), but fewhavedemonstrated itsutility for computing fPAR and modeling photosynthesis using light-useefficiency equations (Chasmer et al., 2009). Methods for quantifyingcanopy transmittance and daily integrated fPAR from direct measure-ments of canopy structure with LiDAR are lacking (e.g., Parker et al.,2001), and accuracy assessments are needed to compare LiDAR-derivedfPAR and remote sensing methods that rely on land cover, multi-anglereflectance, and radiative transfer models (e.g., Shabanov et al., 2003).

The performance of light-use efficiency models and quality of GPPestimates is intimately linked to the accuracy of input variables,particularly fPAR (Zhao et al., 2005; Turner et al., 2006a). Dungan andNemani (2006) evaluated the MODIS MOD17A2 algorithm using theTaylor series method for propagating uncertainty, and determined thatthe influence of fPAR on GPP was greater than the light-use efficiencyparameter (ε) and incident PAR; their findings suggest that improve-ments in the accuracy of fPAR measurements will have the greatestimpact on reducingGPPuncertainty.Most regional- andglobal-scale fPARalgorithms have been developed using satellitemultispectral sensors andrely on radiative transfer equations or empirical relationships withvegetative indices (e.g., Huang et al., 2008; Shabanov et al., 2000; Sellerset al., 1992). TheMODIS fPAR algorithm relies on the former for improvedaccuracy, but reduced reflectance in mixed canopies and deciduousforests limits its use in theGreat Lakes region (Yang et al., 2006; Shabanovet al., 2005, 2007). The MODIS backup algorithm is used in cases wherereflectance measurements are insufficient for obtaining accurate retrie-vals from radiative transfer equations. The backup algorithm is based onempirical relationships betweenNDVI and fPAR, but is considered tohavegreater uncertainty and lower quality (Wang et al., 2001; Yang et al.,2006). MODIS validation studies at this site and other locations withsimilar vegetation have demonstrated fPAR overestimation (Turner et al.,2006a; Steinberg et al., 2006; Ahl et al., 2005), and interannual variabilitytends to be obscured (Turner et al., 2006b).

Global monitoring of photosynthetic activity and primary productionis important to natural resource planners and climate scientists alike,since Earth's climate and ability to sustain consumer demand is linked toplant growth and CO2 uptake potential of the terrestrial biosphere (IPCC,2007; Vitousek et al., 1986). MODIS algorithms for estimating plantproduction rely on light- and carbon-use efficiency equations that are lessdetailed than process-based ecosystem models (e.g., Running and Hunt,1993), since biophysical responses to environmental drivers depend onglobally-averaged lookup table parameters for broad land cover classes(Heinsch et al., 2003). On average, MODIS GPP and NPP estimatesgenerally agree with independent measurements at validation sitesacross the globe (Heinsch et al., 2006; Zhao et al., 2005); however, localbiases and uncertainties need to be addressed beforeMODIS products canbe recommended for use in smaller regions (e.g., Pan et al., 2006). Inmixed forested landscapes of the Great Lakes region of North America,MODIS tends to overestimate LAI and fPAR inputs, resulting in GPP andNPP overestimation (Turner et al., 2006a; Heinsch et al., 2006; Ahl et al.,2005). Estimates of NPP are highly sensitive to the total amount of foliage,because the MODIS algorithm is dependent on allometric relationshipsbetween the mass of leaves and other tissues to estimate maintenance

and growth respiration. Wetlands are another source of uncertainty inproductivity estimates, since wetland area is underestimated in theMODIS primary land cover product (IGBP; Pflugmacher et al., 2007) andthere is no analogue to IGBP wetlands in the University of Maryland(UMD) classification scheme used by MODIS GPP and NPP algorithms(Friedl et al., 2002; Hansen et al., 2000; Heinsch et al., 2003). Also, whilewetland parameters exist in MODIS lookup tables (see Appendix), it isuncertain how well these global parameters represent wetlands of theGreat Lakes Region.

The purpose of this study was to 1) develop methods for improvingIGBP land cover and LAI and fPAR estimates using combined airborneLiDAR data and Quickbird imagery; 2) evaluate the model logic andparameters in theMODIS GPP algorithm using independent observationsfrom a 400 m flux tower; 3) develop biome-specific relationships forcomputing carbon-use efficiency and NPP from LiDAR-derived biomass;4) determine how a wetlands class might change landscape-scaleestimates of photosynthesis and production; and 5) quantify the effectof land cover generalizationonGPPandNPPestimates in aheterogeneouslandscape in the Great Lakes region of North America. We hypothesizedthat fine-resolution inputs would better capture the heterogeneity andspatially variability within landscape. We also hypothesized that thedifference between fine- and coarse-resolution estimates, when inte-grated across entire landscapes, could be used to quantify uncertainties inregional GPP and NPP that are associatedwith land cover generalizationsand the failure to account for wetlands and wetland processes.

2. Materials and methods

2.1. Site description

This studywas conducted in a highly fragmented, forested landscapenearPark Falls,Wisconsin,USA,whichexperiences a continental climatewithwarm,wet summers and coldwinters. At the center of the 6×6 kmstudy area is a 400 m broadcasting tower (45.9470 °N, 90.2732 °W)instrumented to measure local meteorology and landscape-scale fluxesof CO2, H2O vapor, and heat by eddy covariance (Davis et al., 2003;Berger et al., 2001). Observations from this tall tower are unique due tothe large flux footprint, permitting continuous surface flux measure-ments over a heterogeneous landscape that typifies much of the regionsurrounding theGreat Lakes (Desai et al., 2008). Thisflux tower is part ofthe Chequamegon Ecosystem-Atmosphere Study (ChEAS), AmeriFlux,and FLUXNET networks, and data are publicly accessible through ChEAS(http://cheas.psu.edu) and the US Department of Energy's CarbonDioxide Information Analysis Center (http://cdiac.ornl.gov).

The landscape surrounding theflux tower also serves as one ofNASA'sEarth Observing System (EOS) land validation core sites (Nickeson et al.,2007; Morisette et al., 2003; http://landval.gsfc.nasa.gov), includingproducts derived from the Moderate Resolution Imaging Spectroradi-ometer (MODIS), amultispectral sensor onboard NASA's earth-observingsatellites, Terra and Aqua. MODIS provides global estimates of land cover,LAI, fPAR, GPP, and NPP, and field campaigns have been conducted in thesurrounding area to determine the accuracy and spatial variability ofthese products (e.g., Turner et al., 2006a; Ahl et al., 2004; Burrows et al.,2002). Eddy covariance data and previous MODIS validation studies atthis site allowed estimates from this study to be compared and reconciledwith the flux tower time series; geostatistical interpolation (Burrowset al., 2003); and a gridded, process-based ecosystem model (BIOME-BGC; Turner et al., 2005).

The study area has a mean elevation 455 m above sea level, and localrelief varies by b20 m. Although subtle, this difference in elevation ispartially responsible for the complex mosaic of wetland and uplandecosystems that is characteristic of the region. Early- to mid-successionalupland forests dominate the landscape, but approximately one-third ofthe study area is composed of structurally and physiologically distinctlowlandplant communities (Anderson, 2007;Ahl et al., 2004; Ewers et al.,2002; Mackay et al., 2007). Upland stands are generally characterized by

http://cheas.psu.edu

http://cdiac.ornl.gov

http://landval.gsfc.nasa.gov

Fig. 1. Location of digital hemispheric photographs (DHP) taken from within a 30 msquare field plot (dotted line). Arrows indicate the path taken during collection.

2368 B.D. Cook et al. / Remote Sensing of Environment 113 (2009) 2366–2379

mixed northern hardwood species (Acer saccharum, Tilia americana,Fraxinus pennsylvanica, Betula papyrifera); early- to mid-successionalaspen-fir (Populus tremuloides, Populus grandidentata, Abies balsamea);and conifers (Pinus resinosa, Pinus banksiana, Picea glauca). Lowlands aregenerally characterized by wetland shrub and sedge species in fens andalong streambanks (Alnus rugosa, Salix spp., Carex spp.); deciduoushardwood species in retired and seasonal drainageways (Fraxinus nigra,Ulmus rubra, Acer rubrum); ericaceous shrubs and moss in open bogs(Chamaedaphne calyculata, Ledum groenlandicum, Sphagnum spp.); andwetland conifers in drier peatlands and bog edges (Thuja occidentalis,Larix larcina, Picea mariana).

Approximately 78% of the 3600 ha study area is actively managedby the Chequamegon–Nicolet National Forests for multiple-use goals,including wood and fiber production, preservation of fish and wildlifehabitat, and outdoor recreation. Forest composition managementobjectiveswithin the study area range fromearly successional aspen toconifer stands, and prescribed management practices involve clearcutand shelterwood harvests, thinning, fire, and reforestation (USDAForest Service, 2004). Private land holdings are largely forested, withonly a small portion classified as grassland (mostly residential yardsand small fields used for hay production). Less than 1% of the entirestudy area was barren, developed, or occupied by open water.

2.1.1. LiDAR data collection and processingSmall footprint airborne laser scanning data were acquired on 20

August 2005 and 17 April 2006 during “leaf-on” and “leaf-off” periodsfor deciduous plants. Collection and initial processingwas performed byAirborne1, Inc. (El Segundo, CA). First and last returns for each laserpulse were acquired with an ALTM 2025 (Optech, Inc., Ontario, Canada)using a scan frequency of 30 Hz, repetition frequency of 25 kHz, scanangle of 14 °, and a 50% flight line side lap for an average density of 2.1and 1.6 pulses m−2 for the leaf-on and leaf-off collections, respectively.The sensor was flown at a mean altitude of 1350 and 900 m during theleaf-on and leaf-off collections, respectively, resulting in anapproximatespot size diameter of 17 and 11 cm.

TerraScan software (TerraSolid, Ltd., Jyväskylä, Finland) was used todevelop a digital terrain model (DTM) that was used to separate first andlast returns into ground and feature objects (Axelsson, 2000). Groundcontrol points along non-obscured roadways throughout the study area(n=34) were within 15 cm of leaf-off DTM elevations, and differencesbetween leaf-on and leaf-off ground elevations for the entire scene wereb60 cm. Delaunay triangulation of ground hits was used to compute thebase elevation of feature returns, and feature heights were computed bydifference.

2.1.2. Quickbird collection and processingQuickbird imagery (DigitalGlobe, Inc., Longmont, CO) for the study

areawas acquired on 5 Aug 2005 and 17 April 2006, coincidingwith theleaf-on and leaf-off LiDAR collection. Quickbird multispectral imageshave three visible bands (blue, 520–600; green, 520–600; red, 630–690 nm) and one near-infrared (NIR) band (760–900 nm)with a spatialresolution of 2.4 m. Orthorectified and radiometrically corrected datawere provided by the vendor, and atmospheric correction and re-flectancewas computedwith amodifiedMODTRAN algorithm, FLAASH(Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes; ITTVisual Information Solutions, Boulder, CO). Normalized DifferenceVegetation Index (NDVI) was computed from red and NIR leaf-onreflectance values to distinguish openwater and barren land fromgreenvegetation.

To separate deciduous and coniferous vegetation, an unsuper-vised ISODATA classification was performed with ENVI software (ITTVisual Information Solutions, Boulder, CO) using multispectral bandsfrom leaf-on and leaf-off images. The complete land cover classifica-tion scheme and accuracy assessment are discussed in the followingsections.

2.1.3. Plot-scale PAI estimates from hemispheric photographsDigital Hemispheric Photographs (DHP) were collected for this study

to compute fractional cover (fCover or P(0)) by woody plants, plant areaindex (PAI), andΩ for numerous forest types and age classes. Upland andwetland sample plots (n=102) were selected at random from standswithin the Chequamegon–Nicolet National Forest. Thirteen DHP per plotwere collected on a systematic grid (Fig. 1), providing coverage of an areathat approximated the minimum mapping unit for this study (28.8 m).Coordinates for the plot centerwere recordedwith a Trimble (Sunnyvale,CA) Geo-XT or Geo-XH GPS equipped with an external Hurricaneantenna. GPS data were differentially corrected with a reported accuracyof b50 cm to ~2.5 m.

Upward looking DHP were acquired with a Nikon digital camera(Tokyo, Japan) equipped with an 180 ° fisheye lens. Approximately 82%of the imageswere collectedwith a Coolpix 5400 camera and FC-E9 lens(5 M pixel), while the remaining images were acquired with a Coolpix950 or 900 camera and FC-E8 lens (~3 M pixel). Prior to imageprocessing, each camera was calibrated to determine the image opticalcenter and projection function. Images were processed with CAN-EYEsoftware, version 5.0,which ismade available by the InstitutNational dela Recherche Agronomique, Unité Climat, Sol et Environnement (INRA-CSE, Avignon, France; http://www.avignon.inra.fr/CAN-EYE/). CAN-EYEprovides indirect estimates of PAI based on gap fraction measurementsandΩ computed by the Lang and Yueqin (1986) logarithm gap fractionaveraging method, and has been evaluated for use with the VALERIproject (VAlidation of Land European Remote sensing Instruments) atthe Larose Forest in Canada (Abuelgasim et al., 2006) and in croplands(Garrigues et al., 2008). The software facilitates DHP processing byperforming simultaneous processing of images from the same plot;allowing masking of undesired objects in the image (e.g., sun glare,photographer); and providing a user-interface for supervised classifica-tion of sky and vegetation.

2.1.4. Gap fraction methodsGap fraction theory was used to derive biophysical variables from

hemispheric photographs and LiDAR data. Gap fraction models predictthe probability of a probe, such as a light beam,missing all the foliage asa function of gaps between canopy elements at an angle relative to theprobe. This relationship can be expressed as a modified Poisson model,

http://www.avignon.inra.fr/CAN-EYE/

Fig. 2. Leaf-on and leaf-off LiDAR data from a deciduous forest field plot. Similarities inthe a) vertical distribution of feature returns and b) pulse penetration demonstrate theLiDAR footprint effect and instrument sensitivity to both foliage and woody material.


using Markov chains to account for clumping of canopy elements (e.g.,overlying layers of leaves):

P θð Þ = exp−G θ;αð ÞX θð ÞPAI

cos θ

� �ð1Þ

where P(θ) is the gap fraction for zenith angle θ; G(θ,α) is the projectedarea of plant elements with angle α on a plane perpendicular to θ;Ω(θ)is the foliar clumping index or Markov chain factor; and PAI is the plantarea index, which is defined as the hemi-surface area for all foliage(Neumann et al., 1989). The product of PAI andΩ(θ) is also referred to as“effective” PAI (PAIeff=Ω(θ) PAI), and is thequantity oftenmeasuredbyoptical instruments (Chen and Cihlar, 1995). Gap fraction models suchas this have been reviewed elsewhere in the literature (e.g., Weiss et al.,2004; Gower et al., 1999), and form the theoretical basis for manycommercial instruments (Jonckheere et al., 2004; Bréda, 2003; Wellesand Cohen, 1996) and the methods used to analyze hemisphericphotographs of the canopy (Weiss et al., 2004; Leblanc et al., 2005; Chenet al., 2006).

2.1.5. Landscape-scale PAI from LiDARIndirect estimates of fCover, PAI, and Ω from DHP were used to

1) correct LiDAR fCover for footprint effects; 2) parameterize a verticalprojection function [G(0)] for estimating effective PAI (PAIeff) frompenetration of near-nadir (θ=0 to 7 °) LiDAR pulses; and 3) estimateabsolute PAI for the entire landscape.

Each LiDAR pulse has a finite dimension, meaning that gaps smallerthan pulse footprints are potentially missed if partially interception ofthe beam is not considered. Plot-level studies suggest that the idealgap frequency beam width should not exceed 1 mm (Caldwell et al.,1983), which is considerably less than LiDAR footprints of N11 cm inthis study. However, not every feature a pulse encounters will result ina detectable return, which ultimately depends on factors such as pulsepower and repetition frequency, instrument sensitivity and detectionalgorithm, and feature absorption and scattering (Hopkinson, 2007;Chasmer et al., 2006a,b;Wagner et al., 2004). Field observations can beused to compensate for these effects. Indirect estimates of leaf-onfCover, i.e., P(0), from DHP were greater than LiDAR fCover (i.e.,ground:total ratio for first returns) for all the field plots, a bias that hasbeen observed by others (Morsdorf et al., 2006; Parker et al., 2004;Lovell et al., 2003). This bias may be more evident in deciduous andmixed forests where the gap-size distribution is relatively small andrelated to gaps between foliage elements, as opposed to open coniferstandswhere the gap-size distribution is relatively large and related togaps betweenwhole trees. In this study, linear least squares regressionwas used to adjust LiDAR estimates of fCover to match DHP fCover.Differences between LiDAR and DHP fCover computed for view anglesof 0 to 7 ° and 0 to 10 °, respectively, were assumed to be insignificant.

A vertical projection function, G(0), for all the field plots wascomputed by inverting Eq. (1) and fitting a linearized least squaresregression to LiDAR gap fraction (i.e., ground:total ratio for firstreturns) and DHP-derived PAIeff. In deciduous stands, partial inter-ception of the LiDAR pulse by woody elements during the leaf-offperiod resulted in vertical return distributions that closely resembledthe leaf-on period (Fig. 2a). However, somewhat greater canopypenetration was achieved during the leaf-off collection (Fig. 2b),resulting in a better model fit and improved PAIeff predictions. Sinceboth models relied on empirical regression, the leaf-off data wereused to estimate PAIeff for the entire scene at a resolution of 28.8 m;this resolution approximated the field plot size and facilitatedaggregation of Quickbird pixels (2.4 m) without resampling. Bothfootprint size and the large number of returns from woody materiallimit the use of LiDAR for detecting subtle variations in PAI, such asyear-to-year changes within the same stand. However, evidence fromother studies suggest that small footprint LiDAR is capable ofquantifying and delineating insect defoliations that result in a large

loss of foliage (Solberg et al., 2006). LiDAR was unable to resolve thePAI of grasslands, so a value of 1 m2 m−2 was used for these sites onthe basis of direct field measurements (Burrows et al., 2002).

CAN-EYE estimates of clumping from the near-nadir viewangle,Ω(0),were positively correlated with DHP-derived fCover (r=0.75). This wasexpected, since gap fraction should decrease and clumping increase as astand matures (i.e., canopy closure and depth increase). Linear leastsquares regression was used to describe this relationship (data notshown),whichwasusedwith LiDAR-derived fCover andPAIeff to estimatelandscape-scaleΩ(0) and absolute PAI.

2.1.6. LAI time seriesLeaf phenology and LAI expression was monitored from 2000–2005

in an intermediate aspen stand (45.8127 °N, 90.0638 °W) and maturenorthern hardwood stand (45.80587 °N, 90.07969 °W) using contin-uous, indirect measurements of intercepted photosynthetically activeradiation (PAR, 400 to 700 nm) from a single Quantum sensor (LiCOR,Lincoln, NE) at the center of 30×30m plot and direct measurements oflitter fall (n=14 trapswith a 0.092m2 footprint). A simplified version ofEq. (1), the Beer-Lambert law (Monsi and Saeki, 2005), was used tocompute an annual light extinction coefficient (k) as a function of PARtransmission:

Iz = I0 = exp −k × LAIð Þ ð2Þ

where I0 and Iz are the average daily photon flux densities for PARincident at the topof the canopy andbelow thecanopyat aheightof 2m,respectively; and LAI is a direct measure of leaf area index from non-harvested litter fall. Transmittance during the two weeks prior to andfollowing leaf-on was equated with an LAI of zero to account for lightabsorption by non-photosynthetic canopy elements, and mean trans-mittance during July and August was equated with LAI without anycorrection. This weighted approach was used to adjust estimates for agreater fraction of light intercepted bywood during spring leaf-out andautumn leaf fall; in contrast, fully expanded leaves in closed canopieswith high LAI were expected to shade most woody elements (Kuchariket al., 1998; Gower et al., 1999). Light extinction coefficients were


calculated for 2000 and 2002 when litter fall was collected, and meanvalues were used for other years. A daily time series for 2000–2005wascomputed for each site using k and continuous measurements of PARtransmittance, and results were smoothed by a 7-d moving average.

We used observed daily LAI at the aspen and upland hardwoodsites to compute a time series of LAI relative to that observed duringthe collection of LiDAR data. Temporal variations in LAI across thelandscape were obtained by multiplying LiDAR-derived PAI byrelative LAI for similar land cover types. Deciduous PAI for each28.8 m pixel was estimated as the product of total PAI and thepercentage of corresponding Quickbird pixels (n=144) classified asdeciduous (see below). Aspen accounted for ~23% of the landscape,and was computed separately from other deciduous species toaccount for a severe and widespread forest tent caterpillar defoliationthat occurred during 2001 (Weber et al., 2001). Aspen dominatedstands were manually delineated by field observations and photo-interpretation (Maxa & Bolstad, 2009), and temporal variations inaspen LAI were represented by observations from the intermediate-aspen monitoring station. Temporal variations in LAI for all otherdeciduous species were represented by observations from the uplandhardwoods monitoring station. Evergreen LAI was equated withLiDAR-derived PAI and held constant year-round.

2.1.7. LiDAR-derived APAR and independent stand measurementsAbsorbed Photosynthetically Active Radiation (APAR) was calcu-

lated every 30 min for direct and diffuse sources, and totals wereintegrated daily. Global incident PAR was measured above vegetationwith a Quantum sensor (LiCOR, Lincoln, NE) near the tall tower, andmissing values were filled with observations from other flux towers inthe ChEAS network. Diffuse incident PAR was measured during limitedperiods with a Type BF3 Sunshine Sensor (Delta-T Devices, Ltd.,Cambridge, England) and rotating shadowband fitted with a Quantumsensor (Ascension Technology, Inc., Waltham, MA). In the absence of acontinuous record from both sensors, an index of cloudiness (CI) wasused as a proxy for the ratio of diffuse:global PAR (Fig. 3):

CI = 1− I0Imax

� �ð3Þ

where I0/Imax is the ratio of observed and maximum incident PAR(Bird and Hulstrom, 1981).

DirectAPARderived fromLiDARwascalculatedas theproductofdirectPAR and directional fPAR from Eq. (1), where fPAR=1−P(θ). Based onCAN-EYE results, we expected G(θ,α) and LAIeff to vary with the solarzenith angle over the course of a day. However, a consistent relationshipwith cover type was absent, and there were nomeans of estimating fPARwith changes in θ fromnear-nadir LiDARdata. Consequently,we assumeda spherical (i.e., random) leaf angle distribution, where G=1/2cosθ(Campbell and Norman, 1989). For diffuse APAR derived from LiDAR, G

Fig. 3. Relationship between diffuse:global PAR and a cloudiness index (CI) for twodifferent sensors that measure diffuse PAR: a non-mechanical sunshine sensor (Delta-TBF3) and a rotating shadowband (see text for details). Measurements of CI were availablethroughout the study, while the availability of diffuse PAR measurements was limited.

was integrated for all zenith angles (θ=0–90 °; Baldocchi and Hutchison,1986).

We attempted to validate spatiotemporal APAR estimates bycomparingmodel output to single point subcanopy PARmeasurementsthat were continuously measured from 2000 to 2005 in a mixed upland(45.9425 °N, 90.2744 °W) and red pine dominated stand (45.9618 °N,90.2679 °W) located within the study area. Data from these sites wereindependent from plot data used to develop LiDAR-PAI relationshipsand PAR transmission data used to create the LAI time series.

2.1.8. Land cover classification and grid cell sizeThe MODIS GPP algorithm (MOD17A2; Heinsch et al., 2003) uses a

parameter lookup table that relies on a broad land cover classificationobtainedbymoderate-resolution satellite imagery. TheMODIS land coverproduct (MOD12Q1; Strahler et al., 1999) is based on the classificationscheme of the International Geosphere-Biosphere Programme (IGBP;Loveland et al., 2000). These IBGP classes are relabeled to analogousclasses in the University ofMaryland (UMD) land cover scheme for use inMOD17A2. IGBP land cover for this study was derived from structuralattributes from LiDAR (i.e., canopy height and fCover) and multispectralinformation from Quickbird images (i.e., NDVI and reflectance) andsimilarly relabeled to equivalent UMD classes (Hansen et al., 2000).

A decision tree was used to perform IGBP classifications withoutneighborhood aggregation at four different resolutions: 28.8, 57.6,288, and 1008 m (Fig. 4). These resolutions are analogous to existingsatellite sensors such as Landsat, AWiFS, and MODIS, and facilitatedaggregation of Quickbird pixels (2.4 m) without resampling. Waterand barren land were identified using both an unsupervisedclassification of Quickbird reflectance values (see above) and anNDVI b0.5. Wetlands and uplands were manually delineated by fieldobservations and photointerpretation, which had an overall accuracyof 86% (Maxa & Bolstad, 2009). Forests, savannas, shrubland, andgrasslandwere separated based on fCover at heights corresponding toIGBP definitions (Fig. 4). Accuracy of LiDAR-derived heights wasassured by control points and fCover calibration with DHP throughoutthe study area (see above). Evergreen, mixed, and deciduous forestswere separated based the percentage of Quickbird pixels classified asconiferous or deciduous, also according to IGBP definitions (Fig. 4).

2.1.9. GPP model logic and parametersThe MODIS GPP algorithm (MOD17A2; Heinsch et al., 2003) is

based on a climate constrained light-use efficiency model (Monteith,1972; Nemani et al., 2003), which takes the form of a multiplicativeequation (Heinsch et al., 2006):

GPP = APAR × fPAR × emax × m Tminð Þ × m VPDð Þ ð4Þ

where ↓PAR is the photosynthetically active radiation (PAR) incidentat the top of the canopy (MJ); fPAR is the fraction of photosyntheticradiation absorbed by vegetation; εmax is the maximum photosyn-thetic light-use efficiency (g C MJ−1 PAR) when environmentalconditions are nearly optimal; and m(Tmin) and m(VPD) are multi-pliers between zero and unity that reduce the photosyntheticefficiency term based on responses to daily minimum air temperature(Tmin) and mean daytime vapor pressure deficit (VPD), respectively.

Biome specific εmax and parameters for computing m(Tmin) andm(VPD) werederived fromBIOME-BGCsimulations (RunningandHunt,1993) over the global domain for use with upstream MODIS products(land cover, fPAR, and LAI) and interpolated surface meteorology fromNASA's GlobalModeling and Assimilation Office (GMAO) (Heinsch et al.,2003; Zhao et al., 2005). The most up-to-date MOD17A2 parameters(Collection 4.8; Table A1) available at the time of this studywere used inthemodel. Landcover classified as openwater, barren, or urbanwas only1% of the landscape (see below) and assigned a GPP of zero.

These model parameters can be independently verified by collectingusing flux tower estimates of GPP, surface meteorology, and APAR to

Fig. 4. Decision tree used for land cover classification.


invert the MOD17A2 algorithm. Previous studies in a nearby maturehardwood forest and shrub wetland (Cook et al., 2008, submitted forpublication) have shown close agreement between parameters from theMOD17A2 lookup table and field observations. Similarly, we evaluatedmodel performance andparameter values by comparing parameters from

an inversion of tall-tower fluxes with parameter estimates weighted bycover type. Absorbed PAR was averaged across the entire landscape, be-cause periods in time when reliable flux footprints can be calculated arelimited (Wang et al., 2006). Also, flux tower data were not available from18 August 2002 through 10 May 2003 due to instrumentation and data

Fig. 5. Laser pulse penetration data from field plots fit to a linearized gap fraction model(Eq. (1)) using effective plant area index (PAIeff) computed from digital hemisphericphotographs (DHP).

Fig. 6. Continuous observations of leaf area index (LAI) in an intermediate-aspen anduplandhardwood stand from2000 through 2005 (see text for details). Units are relative to2005 growing season LAI, which represents the period when LiDAR data was collected.

Fig. 7. Comparison of absorbed photosynthetically active radiation (APAR) predicted byLiDAR data using gap fraction methods and APAR measured in a a) mixed upland andb) red pine stand with a single PAR sensor in the center of a 28.8 m grid cell. Daily datafrom 2000–2005 is shown together with a 1:1 line. Daily variations were associatedwith differences in incident PAR and seasonal changes in leaf area.


storage problems. Methods for modeling GPP from flux tower observa-tions and inverting MOD17A2 are described elsewhere by Cook et al.(2004).

2.1.10. NPP and Carbon-Use Efficiency (CUE)Net primary production (NPP), defined as the carbon fixed as plant

biomass, can be related to total photosynthesis through a term referred toas carbon use efficiency (CUE=NPP/GPP). It has been proposed that CUEis approximately constant for forests (Waringet al., 1998), but thismaybean artifact of computing respiration as a proportion of net productivity(Medlyn and Dewar, 1999). DeLucia et al. (2007) surveyed the literaturefor independent estimates of GPP and NPP and reported greater CUEvariability between forest types and ages.

One method of quantifying CUE in forests is by comparing biometricestimates of NPP and flux tower estimates of GPP (Curtis et al., 2005).Standing biomass and live woody aboveground NPP (ANPP) in forests inthis landscape were quantified by the authors in a previous study usingLiDAR and field biometry methods (Anderson, 2007). From that study,belowground live woody biomass increment was estimated as 20% ofANPP (Gower et al., 2001), and was added to ANPP along with estimatesof detritus (i.e., leaf litter and fine root turnover) to approximate a 6-yearmeanNPP. Aboveground detrituswas equatedwith leafmass, whichwascomputed by dividing the annual maximum LAI by the specific leaf areafrom the MODIS biome property lookup table (Table A1). Leaf mass ofdeciduous species was expected to turnover annually, while evergreenneedleleaf leaf mass was assumed to turnover every four years (Heinschet al., 2003). Belowground detritus was estimated using a generalizedrelationship between litter fall and belowground carbon allocation (RaichandNadelhoffer, 1989). No attemptwasmade to estimate detritus inputsdue to mortality.

CUEwas linearly related to total biomass for each of the forest classes,and these relationships were used to estimate forest CUE across thelandscape. NPP data for non-forested sites (only ~10% of the entirelandscape) was not available, so a mean CUE for these cover types (0.27)was taken fromaBIOME-BGC simulation at this site (Turner et al., 2006a).

3. Results

3.1. LAI and APAR

LiDAR pulse penetration was well described by PAIeff from DHP andgap fraction theory (Eq.(1); Fig. 5). The near-nadir leaf projectionfunction,G(0), obtained by linear regression for all cover typeswas0.48,suggesting a spherical leaf angle distribution (Campbell and Norman,1989). A G(0) closer to 1 would have suggested a more horizontaldistribution, and a G(0) approaching 0 would suggest a more verticaldistribution.Modelfit errorwas likely tobe associatedwith the accuracyof computing PAIeff fromDHP (Jonckheere et al., 2004; Chen et al., 2006;

Leblanc et al., 2005), and differences in plant architecture. There wereinsufficient data to develop statistical regressions for each of the majortree species and age classes, and the implementation of such a methodwould have been impractical since most grid cells ≥28.8 m containedseveral species and age classes (Anderson, 2007).

Continuous monitoring of LAI in the aspen and upland hardwoodstands recorded similar trends in peak growing season values from 2001

Fig. 8. Interannual and seasonal variations in clear-sky photosynthetic light-useconversion efficiency (ε) estimated by inversion of Eq. (4) with flux tower observationsand LiDAR-derived APAR.

Table 1Changes in land cover distribution and productivity estimates due to grid cell size.

Resolution (m)

28.8 57.6 288 1008

Land cover (%)Evergreen needleleaf forest (ENF) 16 14 7 0Deciduous broadleaf forest (DBF) 9 8 5 0Mixed forest (MF) 5 5 3 0Woodland (WL) 31 37 59 97Wooded grassland/shrubland (WG) 6 7 8 0Open shrubland (OS) 1 1 0 0Grassland (G) 3 2 0 0Wetland (WET) 28 26 17 3Barren/Water 1 1 0 0GPP (g C m−2 y−1) 912±316a 924±282 936±180 929±68NPP (g C m−2 y−1) 384±178a 390±166 404±127 420±66

a Mean±SD for all grid cells (n=[6048/resolution]2).


through 2005 (Fig. 6). The record also documented a severe and wide-spread outbreak of forest tent caterpillars during 2001, which resulted inthe complete defoliation of aspen stands and apartial defoliation of standscontaining less desirable tree species (Cook et al., 2008). Despite a reflushof new leaves following defoliation, LAI was suppressed throughout the2001 growing season. Greater than average LAI was observed during thetwo years following the defoliation, which may have been a response toincreased soil nutrient availability following defoliation and mineraliza-tion of frass (Lovett et al., 2002; Russell et al., 2004) or may exemplifyinterannual variability related to other resources (Cowling and Field,2003). Regardless of the exact cause, these temporal variations in LAIallowed us to test the GPP model under a range of canopy conditions.

A comparison of daily APAR derived from LiDAR and independentsubcanopy stand measurements from 2000 through 2005 revealed astrong linear relationship (Fig. 7a,b), indicating that model estimatesadequately captured daily and seasonal changes in APAR. An exactcomparison of APAR values was not possible, since the Quantumsensor area of influence was not identical to APAR averaged over a28.8 m grid cell. Modeled APAR was less than observed APAR in themixed upland and red pine stands by 13 and 7%, which may suggestDHP measurement error or failure to account for G(θ,α) and Ω(θ)variability (Eq. (1)). However, data from only two points on thelandscape did not indicate a negative bias across the entire scene.

3.2. GPP model performance and parameters

The MOD17A2 algorithm was inverted using tall-tower estimates ofGPP, surface meteorology, and LiDAR-derived APAR to evaluate modelperformance and parameter values. Photosynthetic conversion efficien-

Fig. 9. Linear ramp function describing the constraint on light-use conversion efficiency,m(Tmin), by inversion of Eq. (4) with flux tower observations and LiDAR-derived APAR.

cies (ε) were computed for each day of the year, and an annual εmax of1.09±0.07 g C MJ−1 APAR (mean±SD) was computed for 2000–2005under clear skies and low VPD (Fig. 8). This landscape-averagedparameter derived from flux tower data was similar to the biomeproperty lookup table values (Table A1) weighted for all cover types(1.03 g C MJ−1 APAR). Response to minimum daily air temperature waswell described by the MOD17A2 linear ramp function (Fig. 9), and theparameters that describe this function are similar to values in the biomeproperty lookup table (Table A1). In contrast, ε did not exhibit a responseto mean daytime vapor pressure deficit (VPD). This may be explained bythe temperate climate and low VPD (b1.6 kPa), relatively moist uplandsoils (Martin and Bolstad, 2005; Cook et al., 2004), and the influence ofvegetation in low lying areas, where water is available to plants year-round and wetland-adapted plants are less apt to exert stomatal control.Identical findings were found in a nearbywetland (Cook et al., submittedfor publication), and only aweak VPD responsewas observed in a nearbyupland hardwood forest (Cook et al., 2008). Based on these observations,VPD was removed as constraint on ε in the GPP model (Eq. (4)).

3.3. NPP and carbon-use efficiency

Net primary production (NPP)was 384±26 g Cm−2 y−1 (mean±SD) when averaged across the landscape for years 2000 through 2005(Table 1). Spatial variability of 28.8 m grid cell data was comparable toBIOME-BGC simulations at a similar resolution (25 m) during 2000(Fig. 10; Turner et al., 2006a).Modeled NPP across the landscape rangedfrom 0 to 800 g C m−2 y−1, and the frequency distribution was nearlysymmetric and mesokurtic (Table 1).

Carbon use efficiency (CUE) was linearly related to total biomassfor each of the forest and woodland classes (Fig. 11a–d). Model fitswere best for the forest classes defined by the proportion of evergreen

Fig. 10. Frequency distributions of annual NPP for 2000 computed in this study (28.8 mgrid cells) and BIOME-BGC simulations (25 m grid cells; Turner et al., 2006a,b).

Fig. 11. Relationships between total standing biomass and carbon use efficiency (CUE) for forest and woodland classes (see Table 1 for UMD class abbreviations).


and deciduous trees (i.e., evergreen needleleaf, deciduous broadleaf,andmixed; R2≥0.77), and data weremore variable for the woodlandsclass containing an unspecified mix of evergreen and deciduousspecies (R2=0.67). Aboveground woody biomass was the primaryfactor determining forest ANPP (Anderson, 2007); this translated intoa positive, linear relationship between biomass and CUE, becauseAPAR and GPP estimates in developed canopies of forests andwoodlands were seemingly independent and invariant of standingbiomass (data not shown). Previous studies at this site also confirmthat annual foliage production and APAR are not good predictors ofANPP (Burrows et al., 2003; Ahl et al., 2004).

If GPP estimates from light-use efficiency models are correct and NPPscales as a linear function of biomass, CUE data suggests that forestedstands with less total biomass use a greater proportion of fixed carbon tosupport autotrophic respiration. At first glance, these results appear tocontradict theories that CUE is lowest in old-growth forests with greater

Fig. 12. Attributes of IGBP land cover classes in the study area. Themedian value is representeand minimum values by the vertical “whiskers” (see Table 1 for class abbreviations).

total biomass andmaintenance respiration costs (e.g., Brownet al., 2004).However, much of this landscape is characterized by early successionalvegetation, not old-growth forests. Also, much of the biomass in semi-mature stands is associated with biologically inactive woody biomass,which contributes far less to growth and maintenance respiration thanliving tissue (i.e., leaves, live roots and sapwood). Even in young trees,universal relationships with whole-plant respiration are strongest fornitrogen, not biomass (Reich et al., 2006). Together, these observationsunderscore the need to better understand CUE and age-related trends(DeLucia et al., 2007).

3.4. Land cover and attributes

Overall accuracy of the unsupervised classification of leaf-on andleaf-off Quickbird data was assessed by comparing 28.8 M pixelsclassified as evergreen needleleaf forest, deciduous broadleaf forest,

d by the thick horizontal line; the upper and low quartiles by the box; and themaximum

Fig. 13. Influence of the IGBP wetland class on cumulative estimates of a) daily GPP andb) annual NPP at a grid cell resolution of 28.8 m. Estimates are shown for all possibleIGBP classes including wetlands (IGBP+Wetlands, solid line), and for IGBP classesexcluding wetlands (IGBP-Wetlands, dotted line). Estimates of GPP from the flux towerwithin the study area (gray line) are shown for comparison.


mixed forest, barren, or water (n=135) with field observations andphotointerpretation of satellite imagery. Overall accuracy was 93%, anderrors of omission and commission were b20% for individual classes.

Fine-scale land coverwas dominated bywoodlands (31%), wetlands(28%) and evergreenneedleleaf forests (16%).Other forests andwoodedareas accounted for an additional 20%, and non-forested lands and openwater comprised the remaining5%. Site-specific attributes related to theproductivity of these land cover types are summarized in Fig. 12a–d.Forest classes have the greatest fCover by definition, so it was notsurprising that they demonstrated the greatest fPAR and GPP also.Woodlands, shrublands, and grasslands followed forest GPP inrespective fashion, based on their ability to capture light for photo-synthesis. Wetland plant species and vegetation structure was highlyvariable, ranging from low-stature sphagnumbogs andwetmeadows tolowland hardwood and conifer-dominated forests. Wetland fPAR andGPP were highly variable as a result, spanning the entire range of fPARand GPP values that characterized upland vegetation types.

Carbon use efficiency from independent estimates of GPP and NPPwas not constant for forest andwoodland classes (Fig. 12c). Valueswerewithin theoretical range for higher plants (Amthor, 2000), andremarkably similar to those summarized from the literature by DeLuciaet al. (2007). Evergreenneedleleaf forestswere characterized by a lowerCUE thandeciduous broadleaf forests, presumably because of additionalleaf maintenance costs outside of the growing season. As expected, CUEof mixed forest and woodland classes were characterized by a CUEbetween evergreen needleleaf and deciduous broadleaf forests. As wasthe case for GPP, wetland CUE was highly variable and represented therange of values for upland deciduous and coniferous vegetation.

Combined differences in GPP and CUE contributed to NPP attributesfor each of the land cover classes (Fig. 12d).While evergreen needleleafforests and woodlands captured more light energy through photo-synthesis, a lower CUE resulted in less carbon fixed as biomass than indeciduous broadleaf forests. Sparsely vegetatedwoodlands, shrublands,and grasslands were characterized by a relatively low NPP (medianb100 g C y−1), andwetlandswere characterized by anNPP covering thefull range of all the classes (0 to ~800 g C y−1).

3.5. Influence of a generalized wetland class

Wetlands in the Great Lakes Region often occupy an area smaller thancan be resolved by coarse-resolution sensors (≥250 m) designed forglobal coverage and frequent revisits (e.g., MODIS, AVHRR). Detectionwith fine-resolution multispectral imagery (e.g., Landsat, SPOT) hasproved challenging also, because surfaces can be obscured by vegetation;soils may be saturated without presence of standing water; manywetlands are only seasonally inundated; and broadband spectralsignatures of wetland and upland vegetation are similar (Ozesmi andBauer, 2002). By excludingwetlands from the classification processes, wewere able to evaluate the effect of land cover generalization andmisclassification on landscape-scale GPP and NPP estimates. In theabsence of a wetlands class, wetlands were assigned to one of the uplandclasses based on vegetation structure and composition. However, thestructure of wetland and upland vegetation was quite similar and thedistribution of land cover classes was not drastically altered (data notshown). Consequently, cumulative estimates of GPP and NPP using thefull IGBP classification and one that excluded wetlands differed by only~100 g C m−2 over six years (Fig. 13a,b). Landscape-scale estimates ofGPP by eddy covariance and light-use efficiency model estimates usingeither classification differed by b100 g C m−2 during the same timeperiod (Fig. 13a).

3.6. Effect of land cover generalizations and coarse-resolution fPAR

Land cover distributions shifted considerably when the classificationprocedure was performed with different sized grid cells (Table 1). Ingeneral, this landscape fits the IGBP description for a woody savanna,

which is analogous to a woodland under the UMD scheme, and thepercentage of land cover in this class increased as the resolution becameincreasingly coarse. At the approximate resolution ofMODISGPP andNPPproducts (1 km), all but one of the 36 pixels was classified as woodland(Fig. 14).

Changes in land cover affected landscape-scale estimates of GPP andNPP through MOD17A2 parameters (Table A1) and CUE relationships(Fig. 11a–d) used in the models. Effective PAI was spatially aggregatedby averaging, and was independent of land cover distributions. Thecombined effect of land cover generalization and coarse-resolutioncanopy data on spatial GPP and NPP estimates is illustrated in Fig. 14,and summarized for the entire landscape in Table 1. A broad range ofGPPvalues exist atfinest resolution (28.8m), ranging fromzero for openwater, barren and urban land to ~1800 g C m−2 y−1 for mature forests.The effect of land cover generalizations on landscape-scale GPP wererelatively small (b2%), because MOD17A2 biome property lookup tableparameters are similar for the dominant forest and woodland covertypes (Table A1) and model estimates are more sensitive to dailyincident PAR and fPAR (i.e., APAR). Grid size and land cover effects onNPP were somewhat larger, but differences between fine- and coarse-resolution estimates were still b10% (Table 1). Production estimatesincreased at coarser resolutions, because evergreen needleleaf forestsand non-forested land with low CUE were lumped into generalizedwoodland and wetland classes. Land cover effects on NPP wereminimized by the fact that CUE for the generalized woodland classwas close to the average value for all classes.

4. Discussion

This paper presents a LiDAR-based approach for estimating fPAR foruse as an input in light-use efficiency models of GPP, and introducesbiome- and biomass-dependent CUE parameters as a method forestimating regional-scale NPP. Photosynthesis and productivity esti-mates produced by both of these methods are subject to uncertaintiesassociated with land cover generalizations and biome-specific modelparameters. Species composition, edge effects, soil properties, and otherdeterminants of plant growth are important at fine spatial resolutionsalso (Jarvis, 1995), but an evaluation of these additional factors wasbeyond the scope of this study. The following discussion addressesadvantages offeredby LiDAR-derivedmodel inputs, andpotentialmodel

Fig. 14. Effect of coarse-resolution land cover and spatially aggregated fPAR on GPP and NPP estimates (g C m−2 y−1; see Table 1 for UMD class abbreviations).


uncertaintieswhenheterogeneous land cover is generalized in forests ofthe Great Lakes region in North America.

Active sensors such as LiDAR are less likely to be affected by issuessuch as reflectance saturation, and three-dimensional data providesstructural information that can be used to improve multi-layer radiativetransfer models (e.g., Ni-Meister et al., 2001; Kotchenova et al., 2003).Data from this study suggests that gap fraction theory and generalizedequations that assume spherical leaf angle distributions can be used topredict fPAR andAPAR for a broad-range of forest types in theGreat Lakesregion (Fig. 7a, b). In addition, fCover andgap size information fromLiDARenables estimation of clumping (Ω) of canopy elements and absolute PAI(this study; Chen and Cihlar, 1995). However, footprint effects must firstbe removed if gap fraction and gap size equations are used with discrete,small footprint LiDAR data. This study and others (e.g., Morsdorf et al.,2006; Parker et al., 2004; Lovell et al., 2003) demonstrate that LiDAR-derived fCover is related to pulse size (i.e., minimum detectable gap size)and returns triggered by partial interception of the beam by foliage andbranches. The latter is responsible for similar canopy heights and pulsepenetration during leaf-on and leaf-off conditions (Fig. 2a, b; Næsset,2005), which limits the ability of discrete, small footprint LiDAR to sensefoliage independent of woody elements. Parameters that describe plantcanopy structures can be extracted fromdigital hemispheric photographs(Leblanc et al., 2005; Jonckheere et al., 2004; Chen et al., 2006) to correctfor footprint effects, as was demonstrated in this study, using empiricalrelationships that depend on canopy condition (i.e., leaf-on or leaf-off)and details associated with data acquisition (i.e., flight parameters, laserscanner and instrument settings).

Algorithms such as MODIS MOD17A2 retrieve instantaneous fPARestimates during the satellite overpass, not daily integrated fPAR. Thus,light-use efficiency model predictions also can be improved using a gapfraction equation (Eq. (1)) to compute and integrate fPAR at selectintervals throughout a day. In addition, partitioning of incident PAR into

direct and diffuse sources can account for potential cloudiness effects(Cook et al., 2008, submitted for publication; Turner et al., 2003;Kotchenova et al., 2003). Althoughnear-nadir observations fromairborneLiDAR provide no information about variations in the projection function(G) or clumping (Ω)with changes in solar zenith angle (θ), a spherical leafangle distribution and invariant Ω(θ) produced reasonable results forplant canopies in this study (Fig. 7a, b). Future studies are needed todetermine whether LiDAR returns from a broader range of scan anglesmight beuseful for estimatingP(θ) andother canopy structural properties(Holmgren et al., 2003; Lovell et al., 2003).

LiDARdatawereused topredict the quantity of absorbed PARbyplantcanopies for the GPP algorithm, and to estimate a parameter thatdescribes the conversion of fixed carbon to biomass in theNPP algorithm.Several studies have demonstrated that LiDAR height and densitymetricscan be used to predict standing biomass (e.g., Næsset & Gobakken, 2008;Lim et al., 2003; Patenaude et al., 2004), and data from this site indicatedthat biomass is a critical factor for predicting both woody biomassincrement (Anderson, 2007) and carbon-use efficiency (CUE; Fig. 11a–d).Biome-specific relationships between biomass and CUE constrain theparameter for similar vegetation types (Fig. 12a–d), while at the sametime accounting for stand differences in growth and maintenancerespiration. These observed relationships and LiDAR-derived biomassreduce uncertainties when biomass and plant respiration are estimatedfrom allometric relationships between leaf area and total plant biomass(e.g., MODIS MOD17A3), and produce results that are similar to process-based ecosystem models (Fig. 10; Turner et al., 2006a).

Structural information from LiDAR and spectral information fromQuickbird provide complimentary sources of remotely sensed data, andfacilitated land cover classification for a broad range of resolutions(Fig. 14). We hypothesized that the exclusion of a wetlands class andgeneralizations that accompanied coarse-resolution land cover wouldcontribute to large uncertainties in GPP and NPP estimates. However, the


influence of fPAR (a function of surface vegetation) and lightconditions (a function of incoming PAR and cloudiness) exerted amuch greater influence on GPP than light-use efficiency para-meters. Similar results were observed in a nearby shrub wetland(Cook et al., submitted for publication), where it was demonstratedthat water table effects on GPP were indirectly expressed primarilythrough the amount of annual foliage that was produced. Greaterdifferences between biome-based CUE parameters translated intogreater differences in NPP, but the effect was muted by the overalldominance of forest and woodland classes in this setting (Table 1;Fig. 14). However, we would expect grid cell size and landcover generalizations to have a greater impact on productionestimates in regions where there is broader range of cover typesand CUE, e.g., cropland and grasslands interspersed with forests andwoodland.

To summarize, light- and carbon-use efficiency models produceestimates of GPP and NPP that agree with independent estimatesfrom micrometeorological methods and process-based ecosystemmodels, and predictions were relatively insensitive to errorsassociated with land cover generalizations and the inability todetect wetlands that characterize much of the Great Lakes region.LiDAR provides detailed information about canopy structure, whichpreviously has been obtained through indirect methods such asmulti-directional reflectance and inversion of radiative transfermodels. LiDAR data can be used with gap fraction theory toimprove inputs for modeling of plant growth and production, andcomplements multispectral imagery for determining terrestrialland cover.

Acknowledgements

This research was funded in part by the National Institute forClimatic Change Research (NICCR) and Terrestrial Carbon Processes(TCP) programs of the US Department of Energy (DoE); US NationalAeronautics and Space Administration (NASA) in support of the NorthAmerican Carbon Program (NACP) andMid-Continent Intensive (MCI)campaign; US National Science Foundation (NSF); and University ofMinnesota Initiative for Renewable Energy and the Environment(IREE). Any opinions, findings, and conclusions or recommendationsherein are those of the authors and do not necessarily reflect the viewof DoE, NASA, NSF, or IREE. The authorswish to thank Tom Steele, GaryKellner, and Karla Ortman at the Kemp Natural Resources Station,University of Wisconsin, who provided technical support and accom-modations throughout this project; and to Tim Brass, Steve Burns, andAndy Rasmussen, who demonstrated tremendous attention to detailwhile collecting large quantities of field data. From Bruce, a specialthanks and appreciation goes toWuYang for her constant support andencouragement while writing this manuscript.

Appendix A

Table A1MODIS MOD17A2/A3 (collection 4.8) biome parameters used in this study.

Biomea

ENF DBF MF WL WG OS G WET

εmax

(g C MJ−1 APAR)1.058 1.158 1.158 1.030 0.930 0.764 0.720 1.044

Tmin maximum(°C)

8.31 9.94 9.50 11.39 11.39 8.80 12.02 7.94

Tmin minimum(°C)

−8.00 −6.00 −7.00 −8.00 −8.00 −8.00 −8.00 −8.00

Specific leaf area(m2 kg−1)

23.8 28.2 26.5 34.0 36.0 14.5 41.5 26.2

a See Table 1 for UMD biome abbreviations.

References

Abuelgasim, A., Fernandes, R. A., & Leblanc, S. G. (2006). Evaluation of national andglobal LAI products derived from optical remote sensing instruments over Canada.IEEE Transactions on Geoscience and Remote Sensing, 44, 1872−1884.

Ahl, D. E., Gower, S. T., Mackay, D. S., Burrows, S. N., Norman, J. M., & Diak, G. R. (2004).Heterogeneity of light use efficiency in a northern Wisconsin forest: implications formodeling net primary production with remote sensing. Remote Sensing of Environment,93, 168−178.

Ahl, D. E., Gower, S. T., Mackay, D. S., Burrows, S. N., Norman, J. M., & Diak, G. R. (2005).The effects of aggregated land cover data on estimating NPP in northernWisconsin.Remote Sensing of Environment, 97, 1−14.

Amthor, J. S. (2000). The McCree-de Wit-Penning de Vries-Thornley respirationparadigms: 30 years later. Annals of Botany, 86, 1−20.

Anderson, R.S., 2007. LiDAR-Based estimates of biomass, woody productivity andlandcover in the Chequamegon National Forest, Wisconsin. M.S. thesis, Universityof Minnesota, Saint Paul, Minnesota, USA.

Axelsson, P., 2000. DEM generation from laser scanner data using TIN models. TheInternational Archives of the Photogrammetry and Remote Sensing, Amsterdam,The Netherlands, Vol. XXXIII, Part B4/1, pp. 110–117.

Baldocchi, D. D., & Hutchison, B. A. (1986). On estimating canopy photosynthesis andstomatal conductance in a deciduous forest with clumped foliage. Tree Physiology, 2,155−168.

Berger, B. W., Davis, K. J., Yi, C., Bakwin, P. S., & Zhao, C. L. (2001). Long-term carbondioxide fluxes from a very tall tower in a northern forest: flux measurementmethodology. Journal of Atmospheric and Oceanic Technology, 18, 529−542.

Bird, R. E., & Hulstrom, R. L. (1981). A Simplified clear sky model for direct and diffuseinsolation on horizontal surfaces. SERI Technical Report SERI/TR-642-761 Golden,CO: Solar Energy Research Institute.

Bréda, N. J. J. (2003). Ground-based measurements of leaf area index: a review ofmethods, instruments and current controversies. Journal of Experimental Botany,54, 2403−2417.

Brown, J. H., Gillooly, J. F., Allen, A. P., Savage, V. M., & West, G. B. (2004). Toward ametabolic theory of ecology. Ecology, 85, 1771−1789.

Burrows, S. N., Gower, S. T., Clayton, M. K., Mackay, D. S., Ahl, D. E., Norman, J. M., et al.(2002). Application of geostatistics to characterize leaf area index (LAI) fromflux tower to landscape scales using a cyclic sampling design. Ecosystems, 5,667−679.

Burrows, S. N., Gower, S. T., Norman, J. M., Diak, G., Mackay, D. S., Ahl, D. E., et al. (2003).Spatial variability of aboveground net primary production for a forested landscapein northern Wisconsin. Canadian Journal of Forest Research, 33, 2007−2018.

Caldwell, M. M., Harris, G. W., & Dzurec, R. S. (1983). A fiber optic point quadrat systemfor improved accuracy in vegetation sampling. Oecologia, 59, 417−418.

Campbell, G. A., &Norman, J.M. (1989). The description andmeasurement of plant canopystructure. In G. Russell (Ed.), Plant Canopies: Their Growth, Formand FunctionSociety forExperimental Biology, 31. (pp. 1−19). Cambridge University Press.

Chasmer, L. E., Barr, A., Hopkinson, C., McCaughey, H., Treitz, P., Black, A., et al. (2009). Scalingand assessment of GPP from MODIS using a combination of airborne LiDAR and eddycovariance measurements over jack pine forests. Remote Sensing of Environment, 113,82−93.

Chasmer, L. E., Hopkinson, C., Smith, B., & Treitz, P. (2006). Examining the influence ofchanging laser pulse repetition frequencies on conifer forest canopy returns. Pho-togrammetric Engineering and Remote Sensing, 72, 1359−1367.

Chasmer, L. E., Hopkinson, C., & Treitz, P. (2006). Investigating laser pulse penetrationthrough a conifer canopy by integrating airborne and terrestrial LiDAR. CanadianJournal of Remote Sensing, 32, 116−125.

Chen, J. M., & Cihlar, J. (1995). Plant canopy gap-size analysis theory for improvingoptical measurements of leaf-area index. Applied Optics, 34, 6211−6222.

Chen, J. M., Govind, A., Sonnentag, O., Zhang, Y., Barr, A., & Amiro, B. (2006). Leaf area indexmeasurements at Fluxnet-Canada forest sites. Agricultural and Forest Meteorology, 140,257−268.

Chen, X., Vierling, L., Rowell, E., & DeFelice, T. (2004). Using LiDAR and effective LAI data toevaluate IKONOS and Landsat 7 ETM+vegetation cover estimates in a ponderosa pineforest. Remote Sensing of Environment, 91, 14−26.

Cook, B. D., Bolstad, P. V., Martin, J. G., Heinsch, F. A., Davis, K. J., Wang, W., et al.(2008). Using light-use and production efficiency models to predict forestproduction and carbon exchange during canopy disturbance events. Ecosystems,11, 26−44.

Cook, B. D., Bolstad, P. V., Heinsch, F. A., Davis, K. J., Wang, W., Teclaw, R. M., et al.,(submitted for publication). Cloudiness and water table measurements improveMODIS GPP predictions in a shrub wetland. Journal of Geophysical Research.

Cook, B. D., Davis, K. J., Wang, W., Desai, A., Berger, B. W., Teclaw, R. M., et al. (2004).Carbon exchange and venting anomalies in an upland deciduous forest in northernWisconsin, USA. Agricultural and Forest Meteorology, 126, 271−295.

Coops, N. C., Wulder, M. A., Culvenor, D. S., & St-Onge, B. (2004). Comparison of forestattributes extracted from fine spatial resolutionmultispectral and LiDAR data. CanadianJournal of Remote Sensing, 30, 855−866.

Cowling, S. A., & Field, C. B. (2003). Environmental control of leaf area production:implications for vegetation and land-surface modeling. Global Biogeochmical Cycles,17, 1007. doi:10.1029/2002 GB001915.

Curtis, P. S., Vogel, C. S., Gough, C. M., Schmid, H. P., Su, H. B., & Bovard, B. D. (2005).Respiratory carbon losses and the carbon-use efficiency of a northern hardwoodforest, 1999–2003. New Phytologist, 167, 437−456.

Davis, K. J., Bakwin, P. S., Yi, C., Berger, B.W., Zhao, C., Teclaw, R.M., et al. (2003). The annualcycles of CO2 andH2O exchange over a northernmixed forest as observed from a verytall tower. Global Change Biology, 9, 1278−1293.

http://dx.doi.org/10.1029/2002�GB001915


DeLucia, E. H., Drake, J. E., Thomas, R. B., & Gonzalez-Meler, M. (2007). Forest carbon useefficiency: is respiration a constant fraction of gross primary production? GlobalChange Biology, 13, 1157−1167.

Desai, A. R., Noormets, A., Bolstad, P. V., Chen, J., Cook, B. D., Davis, K. J., et al. (2008).Influence of vegetation type, stand age and climate on carbon dioxide fluxes acrossthe Upper Midwest, USA: implications for regional scaling of carbon flux. Agricul-tural and Forest Meteorology, 148, 288−308.

Dungan, J.L., and Nemani, R., 2006. Evaluating increases in accuracy achieved by NACP.Eos Trans. AGU, 87(52), Fall Meet. Suppl., Abstract B51C-0326.

Ewers, B. E., Mackay, D. S., Gower, S. T., Ahl, D. E., Burrows, S. N., & Samanta, S. S. (2002).Tree species effects on stand transpiration in northern Wisconsin. Water ResourcesResearch, 38, 81−811.

Friedl, M. A., McIver, D. K., Hodges, J. C. F., Zhang, X. Y., Muchoney, D., Strahler, A. H., et al.(2002). Global land covermapping fromMODIS: algorithms and early results. RemoteSensing of Environment, 83, 287−302.

Garrigues, S., Shabanov, N. V., Swanson, K., Morisette, J. T., Baret, F., & Myneni, R. B.(2008). Intercomparison and sensitivity analysis of leaf area index retrievals fromLAI-2000, AccuPAR, and digital hemispherical photography over croplands. Agri-cultural and Forest Meteorology, 148, 1193–1209.

Gower, S. T., Krankine, O., Olson, R. J., Apps, M., Linder, S., & Wang, C. (2001). Netprimary production and carbon allocation patterns of boreal forest ecosystems.Ecological Applications, 11, 1395−1411.

Gower, S. T., Kucharik, C. J., & Norman, J. M. (1999). Direct and indirect estimation of leafarea index, ƒAPAR, and net primary production of terrestrial ecosystems. RemoteSensing of Environment, 70, 29−51.

Hansen, M. C., DeFries, R. S., Townshend, J. R. G., & Sohlberg, R. (2000). Global land coverclassification at 1 km spatial resolution using a classification tree approach. Inter-national Journal of Remote Sensing, 21, 1331−1364.

Heinsch, F. A., Reeves, M. C., Votava, P., Kang, S., Milesi, C., Zhao, M., et al. (2003). User'sGuide: GPP and NPP (MOD17A2/A3) Products, NASA MODIS Land AlgorithmMissoula,Montana: University of Montana.

Heinsch, F. A., Zhao, M., Running, S. W., Kimball, J. S., Nemani, R. R., Davis, K. J., et al.(2006). Evaluation of remote sensing based terrestrial productivity from MODISusing AmeriFlux tower eddy flux network observations. IEEE Transactions onGeoscience and Remote Sensing, 44, 1908−1925.

Holmgren, J., Nilsson, M., & Olsson, H. (2003). Simulating the effects of LiDAR scanningangle for estimation of mean tree height and canopy closure. Canadian Journal ofRemote Sensing, 29, 623−632.

Hopkinson, C. (2007). The influence of flying altitude and beam divergence on canopypenetration and laser pulse return distribution characteristics. Canadian Journal ofRemote Sensing, 33, 312−324.

Huang, D., Knyazikhin, Y., Wang, W., Deering, D. W., Stenberg, P., Shabanov, N., et al.(2008). Stochastic transport theory for investigating the three-dimensional canopystructure from space measurements. Remote Sensing of Environment, 112, 35−50.

Intergovernmental Panel on Climate Change (IPCC) (2007). Climate change 2007: thephysical science basis. In S. Solomon, D. Qin, M. Manning, Z. Chen, M. Marquis, K. B.Averyt, M. Tignor, & H. L. Miller (Eds.), Contribution of Working Group I to the FourthAssessment Report of the Intergovernmental Panel on Climate Change (pp. 996pp).Cambridge, UK: Cambridge University Press.

Jarvis, P. G. (1995). Scaling processes and problems. Plant, Cell and Environment, 18,1079−1089.

Jonckheere, I., Fleck, S., Nackaerts, K.,Muys, B., Coppin, P.,Weiss,M., et al. (2004). Reviewofmethods for in situ leaf area index determination. Part I: Theories, sensors, andhemispheric photography. Agricultural and Forest Meteorology, 121, 19−35.

Kotchenova, S. Y., Shabanov, N. V., Knyazikhin, Y., Davis, A. B., Dubayah, R., & Myneni, R. B.(2003). Modeling LiDAR waveforms from time-dependent stochastic radiativetransfer theory for remote estimations of forest structure. Journal of GeophysicalResearch, 108, 4484. doi:10.1029/2002JD003288.

Kucharik, C. J., Norman, J. M., & Gower, S. T. (1998). Measurements of branch area andadjusting leaf area index indirect measurements. Agricultural and Forest Meteorology,91, 69−88.

Lang, A. R. G., & Yueqin, X. (1986). Estimation of leaf area index from transmission ofdirect sunlight in discontinuous canopies. Agricultural and Forest Meteorology, 37,229–243.

Leblanc, S. G., Chen, J. M., Fernandes, R., Deering, D. W., & Conley, A. (2005).Methodology comparison for canopy structure parameters extraction from digitalhemispherical photography in boreal forests. Agricultural and Forest Meteorology,129, 187−207.

Lefsky, M. A., Hudak, A. T., Cohen, W. B., & Acker, S. A. (2005). Geographic variability inLiDAR predictions of forest stand structure in the Pacific Northwest. Remote Sensingof Environment, 95, 532−548.

Lim, K., Treitz, P., Baldwin, K., Morrison, I., & Green, J. (2003). LiDAR remote sensing ofbiophysical properties of tolerant northern hardwood forests. Canadian Journal ofRemote Sensing, 29, 658−678.

Loveland, T. R., Reed, B. C., Brown, J. F., Ohlen, D. O., Zhu, J., Yang, L., et al. (2000).Development of a global land cover characteristics database and IGBP DISCoverfrom 1-km AVHRR Data. International Journal of Remote Sensing, 21, 1303−1330.

Lovell, J. L., Jupp, D. L. B., Culvenor, D. S., & Coops, N. C. (2003). Using airborne andground-based ranging LiDAR to measure canopy structure in Australian forests.Canadian Journal of Remote Sensing, 29, 607−622.

Lovett, G. M., Christenson, L. M., Groffman, P. M., Jones, C. G., Hart, J. E., & Mitchell,M. J. (2002). Insect defoliation and nitrogen cycling in forests. BioScience, 52,335−341.

MacKay, D. S., Ewers, B. E., Cook, B. D., & Davis, K. J. (2007). Environmental drivers ofevapotranspiration in a shrub wetland and an upland forest in northernWisconsin.Water Resources Research, 43, W03442 10.1029/2006W R005149.

Martin, J. G., & Bolstad, P. V. (2005). Annual soil respiration in broadleaf forests ofnorthern Wisconsin: influence of moisture and site biological, chemical, andphysical characteristics. Biogeochemistry, 73, 149−218.

Maxa, M., & Bolstad, P. (2009). Mapping northern wetlands with high resolutionsatellite images and lidar. Wetlands, 29, 248–260.

Medlyn, B. E., & Dewar, R. C. (1999). Comment on the article by R. H. Waring, J. J.Landsberg and M. Williams relating net primary production to gross primaryproduction. Tree Physiology, 19, 137−138.

Monsi, M., & Saeki, T. (2005). On the factor light in plant communities and itsimportance for matter production. Annals of Botany, 95, 549−567.

Monteith, J. (1972). Solar radiation and productivity in tropical ecosystems. Journal ofApplied Ecology, 9, 747−766.

Morsdorf, F., Kotz, B., Meier, E., Itten, K. I., & Allgower, B. (2006). Estimation of LAI andfractional cover from small footprint airborne laser scanning data based on gapfraction. Remote Sensing of Environment, 104, 50−61.

Morisette, J. T., Nickeson, J. E., Davis, P., Wang, Y., Tian, Y., Woodcock, C. E., et al. (2003).High spatial resolution satellite observations for validation of MODIS landproducts: IKONOS observations acquired under the NAS Scientific data purchase.Remote Sensing of Environment, 88, 100−110.

Næsset, E. (2004). Practical large-scale forest stand inventory using a small footprintairborne scanning laser. Scandinavian Journal of Forest Research, 19, 164−179.

Næsset, E. (2005). Assessing sensor effects and effects of leaf-off and leaf-on canopyconditions on biophysical stand properties derived from small-footprint airbornelaser data. Remote Sensing of Environment, 98, 356−370.

Næsset, E., & Gobakken, T. (2008). Estimation of above- and below-ground biomassacross regions of the boreal forest zone using airborne laser. Remote Sensing ofEnvironment, 112, 3079–3090.

Nelson, R., Short, A., & Valenti, M. (2004). Measuring biomass and carbon in Delaware usingan airborne profiling LiDAR. Scandinavian Journal of Forest Research, 19, 500−511.

Nemani, R. R., Keeling, C. D., Hashimoto, H., Jolly, W. M., Piper, S. C., Tucker, C. J., et al.(2003). Climate-driven increases in global terrestrial net primary production from1982 to 1999. Science, 300, 1560−1563.

Neumann, H. H., Hartog, G. D., & Shaw, R. H. (1989). Leaf area measurements based onhemispheric photographs and leaf-litter collection in a deciduous forest duringautumn leaf-fall. Agricultural and Forest Meteorology, 45, 325−345.

Nickeson, J., Morisette, J., Privette, J., Justice, C., & Wickland, D. (2007). Coordinatingearth observing system land validation. EOS Transactions, 88(7), 81−82.

Ni-Meister, W., Jupp, D. L. B., & Dubayah, R. (2001). Modeling LiDAR waveforms inheterogeneous and discrete canopies. IEEE Transactions on Geoscience and RemoteSensing, 39, 1943−1958.

Ozesmi, S. L., & Bauer, M. E. (2002). Satellite remote sensing of wetlands. WetlandsEcology and Management, 10, 381−402.

Pan, Y., Birdsey, R., Hom, J., McCullough, K., & Clark, K. (2006). Improved estimates ofnet primary productivity from MODIS satellite data at regional and local scales.Ecological Applications, 16, 125−132.

Parker, G. G., Harding, D. J., & Berger, M. L. (2004). A portable LiDAR system for rapiddetermination of forest canopy structure. Journal of Applied Ecology, 41, 755−767.

Parker, G. G., Lefsky, M. A., & Harding, D. J. (2001). Light transmittance in forest canopiesdetermined using airborne laser altimetry and in-canopy quantum measurements.Remote Sensing of Environment, 76, 298−309.

Patenaude, G., Hill, R. A., Milne, R., Gaveau, D. L. A., Briggs, B. B. J., & Dawson, T. P. (2004).Quantifying forest above ground carbon content using LiDAR remote sensing. Re-mote Sensing of Environment, 2004, 368−380.

Pflugmacher, D., Krankina, O. N., & Cohen, W. B. (2007). Satellite-based peatlandmapping: potential of the MODIS sensor. Global and Planetary Change, 56,248−257.

Popescu, S. C., Wynne, R. H., & Scrivani, J. A. (2004). Fusion of small-footprint LiDAR andmultispectral data to estimate plot-level volume and biomass in deciduous andpine forests in Virginia, USA. Forest Science, 50, 551−565.

Potter, C., Gross, P., Genovese, V., & Smith, M. L. (2007). Net primary productivity offorest stands in New Hampshire estimated from Landsat and MODIS satellite data.Carbon Balance and Management, 2. doi:10.1186/1750-0680-2-9.

Raich, J. W., & Nadelhoffer, K. J. (1989). Belowground carbon allocation in forestecosystems: global trends. Ecology, 70, 1346−1354.

Reich, P. B., Tjoelker, M. G., Machado, J. L., & Oleksyn, J. (2006). Universal scaling ofrespiratory metabolism, size and nitrogen in plants. Nature, 439, 457−461.

Riaño, D., Valladares, F., Condés, S., & Chuvieco, E. (2004). Estimation of leaf area indexand covered ground from airborne laser scanner (LiDAR) in two contrasting forests.Agricultural and Forest Meteorology, 124, 269−275.

Running, S. W., & Hunt, E. R. (1993). Generalization of a forest ecosystem process model forother biomes, BIOME-BGC, and an application for global-scale models. In J. R. Ehleringer& C. Field (Eds.), Scaling Physiological Processes: Leaf to Globe (pp. 141−158). San Diego,California: Academic Press.

Running, S. W., Nemani, R. R., Heinsch, F. A., Zhao, M., Reeves, M., & Hashimoto, H. (2004). Acontinuous satellite-derivedmeasure of global terrestrial primary production.BioScience,54, 547−560.

Russell, C. A., Kosola, K. R., Paul, E. A., & Robertson, G. P. (2004). Nitrogen cycling inpoplar stands defoliated by insects. Biogeochemistry, 68, 365−381.

Sellers, P. J., Berry, J. A., Collatz, G. J., Field, C. B., & Hall, F. G. (1992). Canopy reflectance,photosynthesis, and transpiration, III. A reanalysis using improved leaf modelsand a new canopy integration scheme. Remote Sensing of Environment, 42,187−216.

Shabanov, N. V., Huang, D., Yang, W., Tan, B., Knyazikhin, Y., Myneni, R. B., et al. (2005).Analysis and optimization of the MODIS leaf area index algorithm retrievalsover broadleaf forests. IEEE Transactions on Geoscience and Remote Sensing, 43,1855−1865.

http://dx.doi.org/10.1029/2002JD003288


Shabanov, N. V., Huang, D., Knjazikhina, Y., Dickinson, R. E., & Myneni, R. B. (2007).Stochastic radiative transfer model for mixture of discontinuous vegetationcanopies. Journal of Quantitative Spectroscopy & Radiative Transfer, 107, 236−262.

Shabanov, N. V., Knyazikhin, Y., Baret, F., & Myneni, R. B. (2000). Stochastic modeling ofradiation regime indiscontinuousvegetation canopies.Remote Sensing of Environment,74, 125−144.

Shabanov, N. V., Wang, Y., Buermann,W., Dong, J., Hoffman, S., Smith, G. R., et al. (2003).Effect of foliage spatial heterogeneity in the MODIS LAI and FPAR algorithm overbroadleaf forests. Remote Sensing of Environment, 85, 410−423.

Solberg, S., Næsset, E., Hanssen, K. H., & Christiansen, E. (2006). Mapping defoliation during asevere insect attack on Scots Pine using airborne laser scanning. Remote Sensing ofEnvironment, 102, 364−376.

Steinberg, D. C., Goetz, S. J., & Hyer, E. J. (2006). Validation ofMODIS FPAR products in borealforests of Alaska. IEEE Transactions on Geoscience and Remote Sensing, 44, 1818−1828.

Strahler, A., Muchoney, D., Borak, J., Friedl, M., Gopal, S., Lambin, E., et al. (1999).MODISLand Cover Algorithm Theoretical Basis Document (ATBD), Version 5.0: MODIS LandCover and Land-Cover Change Boston, Massachusetts: Boston University.

Thomas, V., Finch, D. A., McCaughey, J. H., Noland, T., Rich, L., & Treitz, P. (2006). Spatialmodelling of the fraction of photosynthetically active radiation absorbed by aboreal mixedwood forest using a LiDAR-hyperspectral approach. Agricultural andForest Meteorology, 140, 287−307.

Thomlinson, J. R., Bolstad, P. V., & Cohen, W. B. (1999). Coordinating methodologies forscaling land cover classifications from site-specific to global: steps towardvalidating global map products. Remote Sensing of Environment, 70, 16−28.

Turner,D. P., Ritts,W.D., Cohen,W.B.,Maeirsperger, T. K., Gower, S. T., Kirschbaum,A.A., et al.(2005). Site-level evaluation of satellite-based global terrestrial gross primary produc-tion and net primary production monitoring. Global Change Biology, 11, 666−684.

Turner, D. P., Ritts, W. D., Cohen, W. B., Gower, S. T., Running, S. W., Zhao, M., et al.(2006). Evaluation of MODIS NPP and GPP products across multiple biomes. Re-mote Sensing of Environment, 102, 282−292.

Turner, D. P., Ritts, W. D., Zhao, M., Kurc, S. A., Dunn, A. L., Wofsy, S. C., et al. (2006).Assessing interannual variation in MODIS-based estimates of gross primaryproduction. IEEE Transactions on Geoscience and Remote Sensing, 44, 1899−1907.

Turner, D. P., Urbanski, S., Bremer, D., Wofsy, S. C., Meyers, T., Gower, S. T., et al. (2003). ACross-biome comparison of daily light use efficiency for gross primary production.Global Change Biology, 9, 383−395.

USDA Forest Service (2004). Chequamegon-Nicolet National Forests, 2004 Land andService Management Plan Washington, DC: US Government Printing Office.

Vitousek, P. M., Ehrlich, P. R., Ehrlich, A. H., &Matson, P. A. (1986). Human appropriationof the products of photosynthesis. BioScience, 36, 368−373.

Wagner, W., Ullrich, A., Melzer, T., Briese, C., & Kraus, K. (2004). From single pulse to full-waveform airborne laser scanners: potential and practical challenges. Proceedings ofthe XXth International Society for Photogrammetry and Remote Sensing Congress (12–23July 2004, Istanbul, Turkey), Commision III, vol. XXXV-B3. (pp. 201−206).

Wang, W., Davis, K. J., Cook, B. D., Butler, M. P., & Ricciuto, D. M. (2006). DecomposingCO2 fluxes measured over a mixed ecosystem at a tall tower and extending to aregion: a case study. Journal of Geophysical Research-Biogeosciences, 111(G02005).doi:10.1029/2005JG000093.

Wang, Y., Tian, Y., Zhang, Y., El-Saleousb, N., Knyazikhina, Y., Vermote, E., et al. (2001).Investigation of product accuracy as a function of input andmodel uncertainties: casestudy with SeaWiFS and MODIS LAI/FPAR algorithm. Remote Sensing of Environment,78, 299−313.

Waring, R. H., Landsberg, J. J., & Williams, M. (1998). Net primary production of forests:a constant fraction of gross primary production? Tree Physiology, 18, 129−134.

Weber, S., Scanlon,K., & Janick, N. (2001).Approximate area affected by forest tent caterpillar[map]. 1;1,500,000 : Wisconsin Department of Natural Resources December 2001.http://www.dnr.state.wi.us/org/land/Forestry/FH/FTC/OutbreakHistory/2001.htm(13 July 2006).

Weiss, M., Baret, F., Smith, G. J., Jonckheere, I., & Coppin, P. (2004). Review of methodsfor in situ leaf area index (LAI) determination. Part II: Estimation of LAI, errors andsampling. Agricultural and Forest Meteorology, 121, 37−53.

Welles, J. M., & Cohen, S. (1996). Canopy structuremeasurements by gap fraction analysisusing commercial instrumentation. Journal of Experimental Botany, 47, 1335−1342.

Yang, W., Huang, D., Tan, B., Stroeve, J. C., Shabanov, N. V., Knyazikhin, Y., et al. (2006).Analysis of leaf area index and fraction of PAR absorbed by vegetation productsfrom the Terra MODIS sensor: 2000–2005. IEEE Transactions on Geoscience andRemote Sensing, 44, 1829−1842.

Zhao, M., Heinsch, F. A., Nemani, R. R., & Running, S. W. (2005). Improvements of theMODIS terrestrial gross and net primary production global data set. Remote Sensingof Environment, 95, 164−176.

http://dx.doi.org/10.1029/2005JG000093

http://www.dnr.state.wi.us/org/land/Forestry/FH/FTC/OutbreakHistory/2001.htm

using lidar and quickbird data to model plant production and quantify uncertainties associated with...

Documents