Satellite-Based Modeling of Gross Primary Production in an Evergreen Needleleaf Forest Xiangming Xiaoa*, David ~Ollin~Er~,John Abera, Mike Goltzc, Eric A

Home , Eddy covariance

Available online at www.sciencedirect.com Remote Sensing s.,ENcE@.,.EcTe of Environment ELSEVIER Remote Sensing of Bnvironment 89 (2001) 5 19-534 www.elsevier.codloeate/rse

Satellite-based modeling of gross primary production in an evergreen needleleaf forest Xiangming xiaoa*, David ~ollin~er~,John Abera, Mike Goltzc, Eric A. avids son^, Qingyuan Zhanga, Berrien Moore 111"

Received 3 March 2003: rcce~ved~n revrsed folni 7 h'ovembe~2003, accepted I1 November 2003

Abstract

The eddy covariance technique provicics valuable information on net ecosystem exchange (NEE) of COz, between the atmosphere and tcrrcstrial ccosystcms, ccosystcm respiration, and gross primary production (GPP) at a varicty of C02 cddy flux towcr sitcs. In this paper, we devclop a new, satcllitc-based Vcgctation Photosynthesis Model (VPM) to estimate thc scasonal dynamcs and interannual variation of GPP of cvergrecn nccdlelcaf forcsts. The VPM modcl uscs two improvcd vcgetation tndice? (Enhanced Vegetation Indcx (EVI), Land Surfacc Water Indcx (LSWI)). Wc used tnultr-year (1998 200 I) lmagcs from the VEGETATION sensor onboard the SPOT4 satellite and COz flux data from a C02 eddy flux tower site in Howland, Maine, USA. The seasonal dynamics of GPP predicted by the VPM yodel agreed well with observed GPP in 1998-2001 at the IIowland Forest. These results demonstrate the potential of the satellite-driven VPM model for scaling-up GPP of forests at the C02 flux tower sites, a key component for the study of the carbon cycle at reg~onaland global scales. O 2003 Elsevier Inc. All rights reserved.

K~vwolzls.VBGETATIOK sensor. Vegctat~on~ndcx. Gross ecosystem exchange of COz; Howland forest

1. lntrocluction a1 , 1999; Lau ct a1 , 2000, 2002: Schuizt: et al., 1999). Evergreen needleleaf forests can act as carbon sinks or carbon The boreal arest is the largest terrestrial biome on Earth sources, depending upon climate and land use history. C02 and is composed of a small number ofplant species. Although flux data collected at flux tower sites provide invaluable relatively simple in vegetation structure, boreal forests play information on ecosystem processes, and can be used to an important role in the global cycles of carbon, water and improve process-based ecosystem models (Law et al., nutrients as well as the climate system. Estimates of net 2000). Eddy flux towers at forest sites provide integrated primary productivity of boreal forests vary widely (Melillo et flux measurements over large footprints that range from a few a).. L 903, Schulze et al.. 1009). In recent years, a number of to many hectares, depending upon tower height and weather field studies have used eddy covariance techniques to provide conditions. To scale-up C02fluxes from flux tower sites is an information on seasonal dynamics and interannual variation important challenge in the study of the carbon cycle at of net ecosystem exchange (NEE), ecosystem respiration (R) regional and global scales. and gross primary production (GPP) of evergreen needleleaf Satellite remote sensing provides consistent and system- forests across the world (Goulden et al.. 1997; Ilollingcr ct atic observations of vegetation and ecosystems, and has played an increasing role in characterization of vegetation structure and estimation of gross primary production (GPP) or net primary production (NPP) of forests (Rehrenfeld et * Conesponding author. Tel.: -t1-603-862-38 IS: fax: +I -603-862- 0188. al.. 2001 : Field et al.. 1995. 1998: Potter et al.. 1993: Prme 1;-mrril adc1re.s~:xiang11ling.xino~~~1n11.cd~1 (X. Xiao). & Goward. 1995; Ru~rnyet al., 1994, 1999; R~rnn~nget al..

0034-42.574- see tiont matter 33 2003 Elsevier Inc. All lights resewed. doi: 10.101bij.rse.2003.11.008 1999, 2000). These satellite-based studies have used the addition, at the canopy level, vegetation canoples are light-use efticiency (LUE) approach to estimate either GPP composed of photosynthetically active vegetation (PAV, (Prince & Ciouard, 1995; Running et al, 1999, 2000) or mostly green leaves) and non-photosynthetically active NPP (F~cldet al.? 1995; Potlcr ct al.. 1993), and the vegetation (NPV, mostly senescent foliage, branches and formulations of these Prod~rctionEfficiency Models (PEMs) stems). NPV has a significant effect on FAPAR at the are the following: canopy level. For example, in forests with a leaf area index of <3.0, NPV (stern surface) increased canopy FAPAR by GPP = c, x FAPAR x PAR (1) 10-40% (Asner et al., L098). At the leaf level, individual green leaves also have some proportion of NPV (e.g., primarylsecondaryitertiary veins), dependent upon leaf age NPP = E, x FAPAR x PAR (2) and type. Thus, FAPAR by a forest canopy must be partitioned into two components: where PAR is the incident photosynthetically active radiation (MJ m- ') in a time period (day, month), FAPAR is the fraction of PAR absorbed by vegetation canopy, and E, is the light use efficiency (LUE, g C MJ- ' PAR) in GPP calculation, and E,, is the light use efficiency in NPP calculation. Only the PAR absorbed by PAV (i.e., FAPARpAV) is used The time step of the PEM models ranges from daily for photosynthesis, therefore, any model that takes the (Running et al., 2000) to monthly (Field et al., 19951, conceptual partition of PAV and NPV of forest canopy into dependent upon image composites of the satellite. 6, or E,, consideration is likely to improve estinlation of the amount is usually estimated as a function of temperature, soil of PAR absorbed by the forest canopy (PAV) for photosyn- moisture and/or water vapor pressure deficit (Field et al., thesis and quantitication of light use efficiency (E, or 6,) of 1995; Prince & Cioward. 1995; Running et al., 2000). vegetation over time. FAPAR is closely related to Normalized Difference A new generation of advanced optical sensors has re- Vegetation Index (NDVI), which is calculated as a normal- cently come into operation, for instance, the VEGETATION ized ratio between red (jllcd)and near infrared (p,,,,) bands (VGT) sensor onboard the SPOT-4 satellite (launched in (Ihcker, 1979): March 1998) and the Moderate Resolution Imaging Spec- troradiometer (MODIS) sensor onboard the Terra satellite (launched in December 1999). These new sensors have more spectral bands, in conlparison to the AVHRR sensor that has only red and near infrared bands for vegetation In remote sensing analysis, FAPAR is usually estimated study (calculation of NDVI). The VGT sensor onboard the as a linear or nonlinear function of NDVI (Prince & SPOT-4 satellite has four spectral bands: blue (0.43-0.47 Cioward. 1005: Ruirny et al.. 1904: Running et a]., 2000): pm), red (0.61-0.68 ym), near infrared (NIR, 0.78-0.89 pm) and shortwave inhred (SWIR, 1.58- 1.75 pm). Data availability of these spectral bands offers an opportunity to develop improved vegetation indices and incorporate them where the coefficients a and 12 vary, dependent upon the into new satellite-based models for improving estimation of NDVI data set used by the PEM models (Prrnce & Goward, GPP of vegetation at regional to global scales. 1995). FAPAR is also closely related to leaf area index (LAI). In this study, our objective is to develop and validate a A number & process-based global NPP models do not new satellite-based Vegetation Photosynthesis Model explicitly calculate FAPAR, but compute a leaf area index (VPM) that estimates GPP of evergreen needleleaf forests (Rutmy el al.. 1999). FAPAR can be estimated as a function of over the plant-growing season, using the improved vegeta- LA1 and light extinction coefficient (li) (Ruiniy et al., 1999): tion indices that can be derived from the new generation of advanced optical sensors (eg, VGT). Our approach is to FAPAR = 0.95(1 - e-"LA') (5) combine the multi-year (1 998-200 1) image data from the VGT sensor onboard the SPOT-4 satellite with C02 flux These PEM models are largely based on the quantitative data from an eddy flux tower site at Howland, Maine, USA. LAI-FAPAR and NDVI-FAPAR relationships, and have The C02 eddy flux tower site is located near Howland, been applied at reegional to global scales, using monthly Maine (45.20407"N and 68.74020°W, 60-m elevation). The NDVI data fi-om AVHRR sensors (Ikld ct a]., 1995: Potter vegetation of this 90-year-old evergreen needleleaf forest is et al., 1993; P~inceK: Gonard. 1995) and SeaWiFS sensor about 41% red spruce (Pinus i.rthetrs Sarg), 25% eastern (Bchrenfcld ct al., 2001). It is known that NDVI suffers hemlock (Twga cunndei~sis(L.) Cam). 23% othcr conifers several lirnitatrons. including sensitivity to atmospheric and 11?6 hardwoods (Ilolliugcr et al., 1999). The leaf area cond~tions,sensitwity to soil background. and saturation index (LAI) of the forest stand is about 5.3 ni'/m2. Plant- of NDVI values in multi-layered and closed canopies. In growing season usually starts around mid-April ( - day X Xino et al. / Hernote Sensing of'E~wirorrma~t89 (2004) 519-534

100) and lasts about 180 days. Eddy flux measurements of (Huete et al., 2002). In an earlier study that compared VGT- COz, H20and energy at the site have being conducted since derived NDVI and EVI for Northern Asia over the period of 1996 (Hollinger et al.. 1999) and is part of the AmeriFlux 1998-200 I, the results indicated that EVI is less sensitive to network (11ttp://public.oni1.g~~~~ian~c~ifl1~.~/1~ataii~1dc~.cfi~1).residual atmospheric contanlination due to aerosols from Availability of CO? flux data from a CO? flux tower site extensive fires in 1998 (Xiao ct ill.. 2003). of evergreen needleleaf forest makes it possible (1) to Significant effort and progress have been made in devel- evaluate the relationship between the improved vegetation oping advanced vegetation indices that are optimized for indices and photosynthetic activities of vegetation, and (2) retrieval of FAPAR from individual optical sensors (Ciobron to assess satellite-based models that estimate the seasonal et al., 1999, 7000: Govaerts et al., 1999). Detailed infornu- dynamics of GPP of forests at the spatial and temporal tion on mathematical formulae and parameters of these scales that are relatively consistent between satellite obser- vegetation indices was given elsewhere ((iobron et al.. vations and flux tower measurements. Any improvement in 1000). The i~nplenlentationof these vegetation indices representation of seasonal dynamics of GPP of forests by requires the top-of-atmosphere (TOA) bidirectional retlec- the satellite-based models will enrich our understanding of tance factors (BRFs) data as input data, and blue band is net CO? exchange between the forest ecosystems and the used to rectify red and NIR bands (Gobron et al., 2000). atmosphere over time. This study is one of many steps These vegetation indices have been optimized for the towards our long-term goal for development and application Medium Resolution Imaging Spetrometer (MERE), the of the satellite-based VPM model to quantify the spatial Global Imager (GLI) and the VEGETATION sensors. patterns and temporal dynamics of GPP of evergreen boreal forests across the globe at 1-km spatial resolution. 2.2. Vegetcztiorz irtdices that use NIR UII~SWIR bands

In comparison to numerous studies that use red and NIR 2. A brief dcscription of vegetation indices spectral bands (e.g., in calculation of NDVI), a limited number of studies have explored the SWIR spectral bands A number of vegetation indices have been developed for (e.g., 1.6 and 2.1 ym) for vegetation study. It was reported broad-waveband optical sensors (e.g.. Landsat, AVHRR) that the SWIR band (1.6 pm) was sensitive to plant water over the last few decades, and can be generalized into three content ('I'ucker. 1980). In order to calculate leaf water categories: (I) vegetation indices that use only red and NIR content, field and laboratory work are needed to measure spectral bands, including NDVI; (2) vegetation indices that fresh weight (FW) and dry weight (DW) as well as specific use blue, red and NIR spectral bands; and (3) vegetation leaf area (SLA, cm'lg) of leaves. Leaf water content is indices that use NIR and SWIR spectral bands. Here we usually described by (1) foliage moisture content (FMC, %, briefly review the last two categories of vegetation indices. calculated as (FW - DW) x 100/FW, or (FW - DW) x 1001 DW) and (2) equivalent water thickness (EWT, glcm', 2. I. Vkgetation indices thot zm blue, red and NIR hands calculated as (FW - DW)/Leaf Area). A number of studies have suggested that a combination of NlR and SWIR bands The blue band is primarily used for atmospheric comc- have the potential for retrieving leaf and canopy water tion, and has also been used in developing improved content (EWT, glcm7), based on Landsat image data (I-lunt vegetation indices that use blue, red and near infrared bands. & Kocli. 1989). hyperspectral image data (Gao. 1996: For instance, to account for residual atmospheric contami- Se~xinoet al., 2000b) and VGT data (Ccccato ct al., 2001. nation (e.g., aerosols) and variable soil background reflec- 2002a,b). The Moisture Stress Index (MSI), which is tance, the Enhanced Vegetation Index (EVI) was developed calculated as a simple ratio between SWIR (1.6 pm) and (Iluetc et 81.. 1997, 2002; Justice et al., 1998). EVI directly NIR (0.82 ym) spectral bands, was proposed to estimate leaf normalizes the reflectance in the red band as a hnction of relative water content (%) and equivalent water thickness the reflectance in the blue band (Iiuete et al.. 1997): (EWT. dcm') of different plant species (flunt & Roch, 1 1989):

where G=2.5, C,= 6, C2=7.5, and L= I (Huctc ct al., 1997). The above MSI index (a simple ratlo between SWIR and EVI is linearly correlated with the green leaf area index NIR bands) could be used as a first approximation to (LAI) in crop fields, based on airborne multispectral data retrieve equivalent water thickness (&m2) at leaf level, (Boegh et al.. 2002). Evaluation of radiometric and bio- based on laboratory measurements, the radiative transfer physical perfomlance of EVI calculated from the Moderate model (PROSPECT) and a sensitivity analysis (Ceccatu et Resolution Imaging Spectroradiometer (MODIS) sensor al., 17001). In analyses of the 10-day con~positeof VGT data indicated that EVI remained sensitive to canopy variations that are fieely available to users through the website (http.' frec.vgt.vito.bc), another water index was calculated as the /~limatc data normalized difference between the NIR (0.78-0.89 pm) and SWIR (1.58- 1.75 pm) spectral bands (Xiao ct al.. 7002c), here it is called "Land Surfice Water Index (LSWI)": flux tower

Val1 atton L.WI = Pnir - Psivir Llte ature Pnir f Pswir

Analyses of multi-temporal VGT data have shown that LSWI is useful for improving classification of cropland and Fig. 1. The schematic diagram of the Vegetation Photosynthesis Model (VPM). EVI-Enhanced Vegetation Index; LSWI-Land Surfacc Wdter 2( forests (Xau al.. 2002a,h,c). This water index is similar in Index: FAPARpAFthe kction of photosynthetically active radiation mathematic Som~ulationto the Normalized Difference Water (PAR) absorbed by the photosynthetic active vegetation (PAV) in the Index (NDWI) that uses reflectance values in the 0.86 and canopy: Twal.,. PKa,, and W,,,,,,---scalars for temperature, leaf phenology 1.24 pm spectral bands of hyperspectral data (Ch, 1996): and canopy water content, respectively; GPP-.--grossprimary production of terrestrial ecosystems; s,-maximum light use eficieney (pol C02/pmol PPFD). VGT-VEGETATION sensor onboard the SPOT-4 satellite: 1'0.86 - /)I24 MODIS-Moderate Resolution Imaging Speetroradionieter onboard the NDWI = (10) P0.86 + P1.24 NASA Terra and Aqua satellites.

Recently, Ceccato et al. (20023.b) proposed the Global Vegetation Moisture Index (GVMI) to retrieve equivalent where PAR is the photosynetically active radiation (ymol water thickness (g/m2) at canopy level, using images from photosynthetic photon flux density, PPFD), and E is the light the VGT sensor: use efficiency (pmol CO2Ipmol PPFD). Light use efficiency (E) is affected by temperature, water, and leaf phenology:

(/'nir(rccti~ic<~)f 0. 1) - (Pswir + 0.02) GVMI = (11) (~o~~(rcct~ficd)f O. ) + (Pswr + 0.02)

where pn,r(rect,ticd)is the reflectance values of the rectified where is the apparent quantum yield or maxim~tmlight NIR band, which are derived from a complex procedure that use efficiency (pmol C02/ymol PPFD), and T,,,I,l, WScap,, involves blue spectral band and uses the apparent reflec- and P,c,l,r are the scalars for the effects of temperature, tance as seen at the top-of-atmosphere (VGT-P product, water and leaf phenology on light use efficiency of vegeta- http:ilwww.vgt.vito.be) as input data (Ciobron et al., 2(!01)). tion, respectively. Field data collected at shrub steppe, shrub savannah, tree T,c,i,, is estimated at each time step, using the equation savannah and woodland in Senegal (West Afiica) during developed for the Terrestrial Ecosystem Model (Raich et al., 1998-2000 were used to evaluate the potential of GVMI for 1991): retrieval of EWT at canopy level (Ceccato ct al.. 20023). The comparison between GVMI and NDVI shows that GVMI provides information related to canopy water content (EWT), while NDVI provides infonnation related to vegetation greenness (Ceccato et al.. 2002;1). where TminrT,,,, and 7;,,,, are minimum, maximum and optimal temperature for photosynthetic activities, respectively. If air temperature falls below T,,,,,,, T,,,I,, is set to be zero. The effect of water on plant photosynthesis (JVScd13has 3. Description of the satcllite-based Vegetation \ Photosynthesis Model (VPM) been estimated as a function of soil moisture andlor water vapor pressure deficit (VPD) in a number of PEM models 3.1. Oivrview qf the VPM model (Field et al.. 1995: Pnnce & Goward, 1995: Running tt al., 2000). For instance, in the CASA (Carnegie, Stanford, Based on the conceptual partition of NPV and PAV Ames Approach) model, soil moisture was estimated using within a canopy (see Eq. (G)), we proposed a new satel- a one-layer bucket model (Malmstrom et al.. 1997). Soil lite-based Vegetation Photosynthesis Model (VPM) for moisture represents water supply to the leaves and canopy, estimation of GPP over the photosynthetically active period and water vapor pressure deficit represents evaporative demand in the atmosphere. Leaf and canopy water content of' vegetation ( FI~.1 ): is largely determined by dynamic changes of both soil moisture and water vapor pressure deficit. As the tirst order GPP = tg x FAPARp,\Vx PAR (12) of approximation, here we proposed an alternative and X Xino et (11. / Rrmote Seiumg qf Env~rorl~rte~~t89 (-7004) 519-534

600 1 I Howland Forest 20 500 2.

2. um 7X 6 400 G - 10 p 2 - N- N- E' E 300 ...m .al L - abl - k' 0 gi p: 200 b m K 2 bl 9 100 -1 0

Time (10-day interval)

Fig. 2. The seasonal dynamics of plioto~ynthet~cdllyactive rad~atlon(PAR) and mean alr tenlperdture dur~ng1998-2001 at the eddy flux tower site of Howland Forest, Mane, USA.

simple approach that uses a satellite-derived water index Evergreen needleleaf trees in temperate and boreal zones (see Erl. (9)) to estimate the seasonal dynamics of WSca1,, have a green canopy throughout the year, because foliage is retained for several growing seasons. Canopy of evergreen needleleaf forests is thus con~posedof green leaves at various ages. Fixed turnover rates of foliage of evergreen needleleaf forests at canopy level were used in some where LSWI,,, is the maximum LSWI within the plant- process-based ecosystem models (Aber &.FcJerer, 1991; growing season for individual pixels. When multi-year I.aw et al.. 2000). In this version of the VPM model, LSWI data are available, we will calculate the mean LSWI therefore, a simple assumption of P,c,I,, is made for ever- values of individual p~xelsover multiple years at individual green needleleaf forests: temporal points (daily, weekly or 10-day), and then select the maximum LSWI value within the photosynthetically active period as an estnnate of LSWI,,,,,. , In the VPM model, P,Ldl,,is included to account for the Photosynthetic activity of vegetation canopy is in part effect of leaf age on photosynthesis at canopy level. Leaf determined by the amount of PAR the PAV absorbs for age affects the seasonal patterns of photosynthetic capacity photosynthesis. To accurately estimate FAPARp,\\, in forests and net ecosystem exchange of carbon in a deciduous forest is a challenge to both radiative transfer modeling and field (Wilson et al.. 7001). In comparing daily light use efficiency measurements. In this version of the VPM model, FAPAR from four Con tlux tower sites (an agriculture field, a pAv within the photosynthetically active period of vegetation tallgrass prairie, a decirluous forest and a boreal forest), is estimated as a linear hnction of EVI: the results support inclusion of parameters for cloudiness and the phenological status of the vegetation (Ttu-ner et d., FAPARpAv= EVI (19) 2003). In the VPM model. calculation of P,,,Ia, IS dependent 4 upon life expectancy of leaves (deciduous versus ever- 3.2. Esfiination oj'inodel parameters for evergreen boreal green). For a canopy that is dominated by leaves with a jbrests life expectancy of 1 year (one growing season, e.g., deciduous trees), P,,:ll,, is calculated at two different phases (note The EO values vary with vegetation types, and informa- that detailed d~scussionof P,,,,,, of deciduous forests will be tion about EO for individual vegetation types can be obtained presented in another paper.): from a survey of the literature (Ruimy et nl., 1995) andlor analysis of gross ecosystem exchange of CO? and photo- I + LSWI synthetic photon flux density (PPFD) at a C02 eddy flux Pscalar = 7, tower site (Goulden rt al., 1997). Estimation of the During bud burst to leaf full expansion ( 16) parameter is largely determined by the choice of either a linear or nonlinear model (e.g., hyperbolic equation) be- PScaI,,.= 1 After leaf fill1 expansion (17) tween GPP and inctdent PPFD data (generally at half-hour 524 X. Xiuo et nl. / Re~iioteSensing qf'Eiiviro~mrent89 (2004) 519-534

time-step) over a year (Frolking et a]., 1998; Ruimy et al., sets (Kuinq et al., 1995), it was reported that in a linear 1995, 1990): model, GPP= 0.020 x PPFD (i.e., c0 30.020 pmol C02/ pmol PPFD or - 0.24 g Clmol PPFD), but in a nonlinear hyperbolic function, GPP=0.044 x PPFD x 43.35/(0.044 x PPFD + 43.35) (i.e., E~ = 0.044 ymol COz/pmol PPFD or 0.528 g Clmol PPFD). In the VPM model, EO values u x PPFD x GEE,,, - NEE = - R derived from the hyperbolic function are used. a PPFD GEE,,,, x + In order to obtain 80 value for the VPM model, a literature survey was conducted to gather published infor- where cc is apparent quantum yield (as PPFD approaches to mation on EO for evergreen needleleaf forests, in those O), and /lis the slope of the linear fit. publications values were estimated using the nonlinear For instance, in a review study that exanlined the rela- hyperbolic function (Eq. (21)). The Boreal Ecosystem- tionship between GPP and PPFD tiom 126 published data Atmosphere Study (BOREAS) conducted C02 flux mea-

Howland Forest

----8-- 1998

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Time (10-day interval)

Howland Forest [3

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Time (10-day interval)

Fig. 3. The seasonal dynam~csof nct ecosystem euchange of C02 (NEE) and gross prlmary product~on(GPP) at the eddy flux tower of Howland Forest, Mame. USA. X. Xiuo et RI. /Remote Sensing qf'Enviro~me~~89 (2004) 519-534 525 surement at a few evergreen needleleaf forest sites in to the 0.48 g Clmol PPFD used here from a boreal forest Canada. During 311 611 994.- 1013 111 996, the eddy covari- tower site in Canada (Cioulden ct a]., 19971, based on an ance technique was used to measure net ecosystem ex- approximate conversion of 2.05 -2.17 between MJ (1 0") change of C02 between the atmosphere and a black and mol PPFD (Aber et al., 1996: Wiss 61 r*;ornian, 1985). spruce (Picell mariann) forest in central Manitoba, Canada In calculation of T,,,l,, (see Eq. (14)), T,,,, To,, and T,, (Goulclen et al., 1997). The site (55.879ON. 98.484"W) is values vary among different vegetation types (Aher & dominated by 10-m-tall 120-year-old black spruce, with a Federer. 1992; Raich et al., 199 1). For evergreen needleleaf minor layer of shrubs and continuous feather moss. Through forest, we use 0, 20 and 40 "C for T,,,;,, T,,,, and T,,,, examination of the relationship between GPP and incident respectively (Aber 6r Federer, 1992). Photosynthesis of photosynthetically active photon flux density (PPFD), it was conifers in temperate to boreal zones is limited by low reported that apparent quantum yield for the tower site is temperahtres (DeLucia & Smith, 1987). To better capture ~~=0.040pnol C02/pmol PPFD (Gouldcn ct 31.; 1997). the effect of air temperature, in calculation of TScai,,instead . Sinlilarly, an intensive field campaign (IFC) was conducted of using the daily mean air temperature that is calculated as in midsummer or peak growing season (IFC-2 = July 26 to the average value between daily maximum temperahire August 8, 1994) at the 19 nmriritia forest (6ld black spruce, (generally daytime) and daily mininlum temperature (night or OBS site, located at 53.85"N and 105.12" W) in Canada time), we used the average daytime temperature, which was (Sullivan et 31.. 1997). Vegetation at the site consists calculated as the average between daily mean temperature primarily of a I? nmriaiza overstory (up to 12 m tall and and daily maximum temperature (Aber Rr. Fellercr, 1992). 155 years of age) with some tamarack and Pinus hattksiczna present. The apparent quantum yields calculated from the data measured in IFC-2 were 0.041 It 0.003 for upper 4. Sitc-specific data for simulation and validation of the canopy and 0.035 + 0.002 (pmol C021pol PPFD) for VPM model lower canopy, respectively (Sulli~wet al.. 1997). In this study, we used 8" = 0.040 ymol C02/pmol PPFD, or 0.48 g 4.1. Description of site-spec!cific,field data Clmole PAR for evergreen needleleaf forest ((ioulden et al., 1997). The eo = 0.040 pmol C02/pmol PPFD value was also Daily climate (maximumiminimum temperature, precip- used in the 3-PG model that uses leaf area index to calculate itation) and photosynthetically active radiation (nlollday FAPAR of a Pims ponderosa forest (Law GI at., 2000). In PPFD) data during 1996 2001 at Wowlapd Forest were the standard MODIS-based GPP/NPP algorithm (MOD17) available for this study. The annual mean air temperature that uses NDVI to calculate FAPAR (Running et al.. 1.999. during 1996-2001 was about 6.7 "C, while the annual 2000), the E" value of evergreen needleleaf forest is 1.008 g mean daytime air temperature during 1996-2001 was about CIMJ (approxinlately 0.46-0.49 g Clmol PPFD), very close 9.2 "C. In order to be consistent with the 10-day con~posite

r I

Time (10-day interval)

Fig. 4 The seasonal tlynam~csof su~facerctlectance values of four spectral bands of VEGETATION sensor durmg 1908-2001 at the eddy flux tower site of Howland Forest, Mame, USA. 526 X. X~uoet nl. /Remote Sensing qf E~tvirorr~~~cnt89 (2004) 519-534

satellite images we used (see Section 3.2), we calculated the the eddy flux tower site at Howland Forest (Fig. 41, based 10-day mean temperature from the daily temperature data, on the geographical information (latitude and longitude) of and the 10-day sum of PAR from the daily PAR data, the tower, and then calculated vegetation indices (Fig. 5). In respectively (Fig. 2 ). order to estimate LSWI,,,, for the Howland Forest site, we Daily flux data of NEE, GPP and ecosystem respiration calculated the mean seasonal cycle of LSWI for all 10-day (R) at Howland Forest during 1996-2001 were generated periods in the 4-year data set (1998-2001). The resulting from the half-hourly tlux data. Half-hourly values were mean seasonal data at the 10-day interval represents a calculated from the covariance of the tluctuations in vertical wind speed and C02 concentration measured at 5 Hz (Holl~ngeret al., 1999). Half-hourly flux values were excluded from further analysis if the wind speed was below 0.5 rn s I, sensor variance was excessive, rain or snow was falling, for incomplete half-hour sample periods, or instru- ment malfunction. At night flux values were excluding from further analysis if the friction velocity (uh) was below a threshold of 0.25. To obtain annual estimates of C02 exchange, values missing from the half-hourly record of annual NEE were modeled by con~biningestimates of canopy photosynthes~sand nochtmal respiration. Daytime C02 exchange rates were obtained from Michaelis-Menten models of PPFD with coefficients fitted on a monthly basis. Howland Forest Missing nocturnal C02exchange values were obtained from y ,O, , , , , , , , , second order Fourier regressions between Julian day and Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Oec nocturnal respiration. Ellled half-hourly NEE data were used Time (10-day interval) to estimate respiration and GPP in the following way. All data points with PAR values less than 5 pmol m-- ' s - ' were used to estimate dark respiration rate. For each year, all "dark" NEE values were regressed against measured soil temperature. The resulting regression equation was then used with measured soil temperatures to predict respiration during "light" periods (PAR>5 pol m-2 s- I). GPP was then estimated as NEE minus estimated respiration for all "light" periods (using conventton of opposite signs for GPP and respiration). We calculated the 10-day sums of GPP and NEE from the daily GPP and NEE data, in order to be consistent with the 10-day composite satellite images we Howland Forest used (I'ig. 3). 024 , , , 9 1 5 3 , , , 1 I Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec 4.2. Description of itmge.~,from the VEGETATION se~rsor Time (10-day interval)

We used 10-day composite images from the VEGETA- TION (VGT) optical sensor onboard the SPOT-4 satellite that was launched in March 1998. The VGT sensor provides daily images for the globe at I-kin spatial resolution. Standard 10-day synthetic products (VGT-SIO) are generated by selecting a pixel with the maximum Normalized DifTerence Vegetation Index (NDVI) value in a 10-day period, and are freely available to the public (11ttp:I frcc.vgt.vito.be). There are three 10-day composites within a month: days 1 - 10, 11 -20, and 21 to the end of month. We acquired the VGT-SlO data over the period of Howland Forest k / April 1 - 10, 1998 to December 2 1 3 1, 2002 for the globe. 0 1 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Details on methods for pre-processing and calculation of T~me(10-day interval) vegetation indices from VGT-S I0 data were presented elsewhere (X~aoet al., 2003~.1001). In this study, we Fig. 5. The seasonal dynamics of vegetation indices during 1908-2001 at extracted spectral bands from one I -km pixel that covers the eddy flux tower site of Howland Forest, Maine, USA. 120 1.o "typical year", and we then selected the maximum LSWI -8- GPP - EVI Howland Forest value (0.4 1; July 2 1-3 1) within April 1 -November 10 as an estimate of LSWI ,,,,, (i.e., LSWI,,, = 0.4 1).

5. Results

5.1. Seusonul dynamics of EVI mild ND VI jbr evergreen needleleu[ fi~rest

The seasonal dynamics of EVI differs tiom that of NDVI during late April--early November in terms of both Tme (10-day interval) magnitude and phase (I:ig. 5). During summer (June, July and August) of 1998-2001, the maximum NDVI value 12C u ranges in the order of 0.80 0.85, and were much higher .- Howiand Forest $ l0C than the maximum EVI values (in the order of - 0.5). a When calculating FAPAR using LA1 for the Howland X 2 80 Forest site (see Eq. (5), with k=0.5, LAI=5.3 m2/m2), - the resultant FAPAR is about 0.88. There were relatively m f 60 N large differences between NDVI and EVI over the plant- E growing season at 10-day interval in 1998-2001 (Figs. 5 3 -0 40 and (7). a During the plant-growing season, EVI reached its peak in 8 20 I 0 EVI VS. GPP early summer and then declined gradually (Figs. 5 and 6). The observed decrease of EVI after reaching its peak in 0 I early summer may be caused by many factors, including Enhanced Vegetation lndex (EVI) complex interactions between atmosphere and leaflcanopy as well as leaf optical property. In addition to the correction 120 u 0 NDVI vs GPP term (6 x pEd- 7.5 x pl,luc) in the EVI equation, which .- -GPP = -24.1 + 113 3 X NOVI, ?=0.20.N=88 corrects residual atmospheric contamination above the can- a$ loo opy, changes in leaf properties may also contribute to the Howland Forest 2 80 decline of EVI over time. Evergreen needleleaf trees consist T- of leaves with various ages of years. As a needleleaf gets m 2 60 N old, it changes in its size (e.g., leaf thickness), dry weight. E chlorophyll content and nitrogen content. Based on a G 0 40 comparative assessment of needle anatonly of red spruce -n (Rock et al.. 1994), needleleaf thickness of 1st year leaves 4 20 (726 + 44 pm) is about I I% smaller than that of 2nd year leaves (803 + 46 pm), and there is less intercellular air space 0 ( (%) in the 2qd year leaves. Although chlorophyll and Normalized Difference Vegetation lndex (NDVI) nitrogen concentrations (mglg DW) may be relatively stable over seasons, an increase in leaf thickness results in a larger Fig. 6. A co~nparison between vegetat~on Indices and gross piimary volume of needleleaf, which leads to dilution effect of production (GI'P) at the eddy flux tower site of Howland Forest. Maine, chlorophyll and nitrogen in the needleleaf (mglcm". The USA. The data within the period of April 1 to Kovember 10 during 1998- changes in leaf size (e.g., thickness), intercellular air space, 2001 were used. The simple linear regression niodcls behveen GPP and $ EVI or NDVI have a P.:0.0001. dry weight. and the dilution effect might together affect retlectance, transmittance and absorption of needleleaf, for instance, the 2nd year needles of red spruce have slightly The conlparisons between vegetation indices (EVI, higher retlectance values in blue band but little change of NDVI) and GPP show that the seasonal dynamics of EVI reflectance values in red band, in comparison to the first- followed those of GPP better than NDVI in tenns of phase year needles (Rock et al., 1991.). The adjusting factor (L= 1) and amplitude of GPP (Fig. 6). When using all the obser- for soil and vegetation background in the correction term vations within April 1 to November 10 during 1998 2001, (6 x pmd- 7.5 x pbluc+ L) of EVI equation also plays a EVI has a stronger linear relationship with GPP than NDVI large role in the seasonal dynamics of EVI. After reaching (Fig. 6). The NDVI curve seems to be out of phase with its peak in early summer, NIR values declined gradually, GPP in the early and late part of the plant-growing season. resulting in lower EVI values (Fg. 3). After its peak in early summer, EVT gradually declined over Table I respectively. The "spring trough" corresponds to the begin- Time-integrated soms of gross primary production (g ~hn', GPP), ning of photosynthetically active period of evergreen nee- photosynthet~callyactive radiation (mol/ni2, PAR), and prcc~p~tation(nun, PPT) at Howland Forest, Maine, USA dleleaf forest. LSWI increased through spring and reached its peak in late July. The 10-day periods that have minimum Year Towcr data VPM model PAR and clirnatc LSWI in fall season were November 1- 10 (1998), October GPPOIX11-12, GPPohs 64-11) GPPolcd(4-11) PARLII PPT4-11 2 1-3 1 (1 999). November 1- 10 (2000) and October 1 1-20 (2001). The "fall trough" corresponds to the ending of photosynthetically active period. Seasonally integrated GPP over the period of April 1 to November 10 accounts for 91% (1 998), 88% (1 999), 9 1 '% (2000) and 92% (200 1) GPP,I,, .12, is the obsen~dGPP from January to Dcccmber. GPPOb5(J I I,, GPP,,d (4 I 1,. PA& I I and PPT4 I I are the obsenwl GPP, predicted of the annually integrated GPP from Janualy I to December GPP, PAR and PPTover the per~odof Apnl I to Kovcmber 10, rcspcctwcly. 3 1, respectively (Table 1). The "spring trough" and "fall trough" of LSWI time series within a year were also late summer and fall seasons while NDVI had little change observed in an earlier study that examined multi-temporal during the same period (Fig. 6). LSWI data for deciduous broadleaf forests and evergreen needleleaf forests in Northeastern China (Xiao et al., 2002~). LSWI time series data in 1998 -2001 had distinct seasonal dynamics within the plant-growing season (Fig. 5). We also calculated the Moisture Stress Index (MSI, see Eq. Among the three vegetation indices (LSWI, EVI, and (8)) for all 10-day composites of VGT data in 1998-2001 NDVI), the seasonal dynamics LSWI is unique and charac- and the comparison between MSI and LSWI (Fig. 7) shows terized by a "spring trough" and a "fall trough" (big. 5). At that there is a close relationship between MSI and LSWI for Howland Forest site, forest stands in the winter and early evergreen needleleaf forest at Howland Forest site. The spring is largely composed of snow, wet soil and vegetation. results from a modeling study that used the PROSPECT A snow pack of up to 2 m could exist from December to radiative transfer model confinned the relationship between March. Snow has much higher reflectance in visible and equivalent water thickness (EWT, g/cm2) and the MSI, and near infrared bands, in comparison to vegetation. The high MSI could therefore be used as a first approximation to LSWI values in winter and early spring are attributed to retrieve vegetation water content at leaf leuel (Ceccaro ct al., snow cover in the forest stands. As snow melted in late 2001 ). At canopy level, EWT,,,,,,, (g/m2) is a product of spring. LSWI declined. The 10-day periods that had mini- EWTlCdt(g/ni2) and leaf area index (LAI, m2/m') (Ceccato mum LSWI in spring season were May 1 - 10 (1 998), April et nl., 7007b). NO field measurements of leaf and canopy 11-20 (1 999), April 1- 10 (2000) and May 1 - 10 (200 I), water content at Howland Forest site during 1998-2001

0.8

0.7 - Howland Forest

Land Surface Water lndex (LSWI)

FI~.7. A cotnpallson between the Land Su~tjceWater Index (LSWI) and Mo~stureStress lndex (hlSI) dunng 19%-2001 at the eddy flux tower slte of Howland Foreat, Mame. LSA. X. Xicio er al. /Remote Seasing of Enviro~ra~ent89 (2004) 519-534 529

were available for con~parisonbetween LSWI and vegeta- needleleaf forests have only slightly changes in LA1 and tion water content. However, field sampling of fresh weight SLW over the plant-growing season, and therefore, single (FW) and dry weight (DW) of spruce and hemlock needles LA1 and SLW values were often used in estimation of GPP at Howland Forest site were conducted on six dates (5119, 61 of evergreen needleleaf forests by some process-based 6, 719, 7/16, 817, 9/11) in 2002. Foliage moisture content ecosystem models (Abcr & Federer, 1992; Law ct al., (FMC, %) at Howland Forest was calculated using fresh 2000). As a simple approximation, we used LAI=5.3 m'l weight (FW) and dry weight (DW) of leaves (= 100 x m' and SLW=280 g/m2 (Aber & Federer, 1992) to estimate (FW - DW)/FW). Because no field measurements of spe- EWT of evergreen needleleaf forest for the six sampling cific leaf weight (SLW, g1cm2) and leaf area index (LAI) dates at Howland site (Fig. X), and the resultant EWT varied were conducted on those sampling dates in 2002, we cannot from 0.018 g/cm2 (511912002) to 0.048 g/cm2 (7/9/2002), accurately calculate EWT on those sampling dates. Unlike within the EWT range reported in a study that examined the the grassland and savannah vegetation that have large relationship between MSI and EWT (Ccccato ct al.. 2001). seasonal changes in SLW and LA1 over the plant-growing The temporal dynamics of LSWI within the plant-growing season (('cccato ct al., 2002a1, mature stands of evergreen season at the Howland site are sensitive to changes of FMC

0 50 70

Howland Forest ,

Year 2002 (10-day interval)

Land Surface Water Index (LSWI) in 2002

Fig. 8 A comparison benvcen Land Surface Water Index (LSWI) and vcgetatlon water content of torcsts In 2002 at the flux tower slte of Howland Forest. Mane, USA. The crror bars for foltagc rrlolstulc contcnt (FMC. %) and cqu~valentwater th~ckncss(EWT. gicd are thc standard dev~at~onof mcaswerrlcnts at lnd~v~dualsamplmg dates. Field measurements of fresh and dty we~ghtwere conducted on leaves of nvo domlnant spccles (red spmce and eastern hemloch) at the Howland site. 120. v v B Howland Forest o Howland Forest

.-c 60-

0, , , , , . , , , , , , Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Year 1998 (10-day mterval) Year 1999 (1 0-day interval)

I 12a1 y\iHowland Forest Howland Forest

0 0 Ooo 0 , - 0, , . , ...... , I Feb Mar Apr May Jun Jut Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Year 2000 (10-day ~ntervai) Year 2001 (10-day ~nterval)

Fig. 9. A co~njxarisonofthe seasonal dynamics between the observzd gross primary production (GPP) and predicted CiPP overthe photosynthetically active period (April I to Kovember 10) during 1998-2001 at the eddy flux tower site of Howland Forest, Maine, USA. Solid circle-GPP,,,,,. and open cir~le-GPP,,~.

and EWT in 2002 (Fig. 8). For hture tield work at the Howland site, add~tionaltield measurement of SLW and LA1 should be carried out together with measurements of foliage fresh weight and dry weight, which would lead to improve retrieval of EWT through satellite-based water indices, as demonstrated in recent studies (Ceccato et al.. 2001, 1002a,b). As suggested by the limited field data of FMC in 2002 (Fig. 8) and the close relationship between LSWI and MSI during 1998-2001 (Fig. 7) at the Howland site, LSWI might be a useful indicator for canopy water content of evergreen needleleaf forest.

5.3. Seasonal djmmics of predicted C02JIzrxes from the VPM nzud?l

The VPM model was nm using the site-specific data of I d temperature, PAR and vegetation indices in 1998-200 1. The seasonal dynamics of predicted GPP (GPPpred)tiom the

Howland Forest VPM model was compared with the observed GPP (GPPnb,) data at 10-day interval over the period of April I -Novem- ber 10 (k'i~.9). The seasonal dynamics of GPPprcdover the

Observed GPP (q cim2) In 10-day from tower data plant-growing season agreed reasonably well with those of GPPOb,. The simple linear regression model also shows a f ~g 10 A cornpartson between the observed gross pumdry product~on good agreement between GPPprcdand GPPnh, during the (GPP) and p~edlctedGPP ovel the photosynthet~callyactwe per~od(Apnl I plant-growing season In 1998-200 1 (Fig. 10). Seasonally to November 10) dunng 1998-2001 at the eddy flux towel vte of Howland integrated GPPPKd(g ~lm') over the period of April 1 to hest, Mame I he s~mpleIme& regrewon models GPI',,d=O 91 X GP- GI-'P,b,, 1'-095, Y=8X, fJ< 00001. or GPP,,,d= - ll 63+1 08 xGP- November 10 is lower than seasonally integrated GPP,,b,. G1'Pob,. 1'-0 70, iV- 88. 1' -0 0001 ranging from - 20% in 2000 to - 3% in 1999 (Table I). 6. Discussions FAPARNpv are important, and a laboratory-based study was conducted to estimate canopy photosynthetic and non- The multi-year simulations of the VPM model have photosynthetic components from spectral transmittance shown that in general, there is a good agreement between (Serrmo et al.. 2000a). EVI is a semi-empirical mathe- GPPPrcdand GPP,,,,, over the photosynthetically active matic transformation of observed reflectance from individ- period during 1998-2001. However, there still exist large ual spectral bands (blue, red and NIR) of optical sensors differences between GPPub, and GPPprcdin a few 10-day (Fluete et al., 2002). The seasonal dynamics of EVI agreed periods (Fig. 91, for instance, smaller GPPPxd in early April, well with the observed GPP of evergreen needleleaf forest and October-November during 1998-2000. Those large in 1998-2001, but NDVl had poor correlation with GPP discrepancies between GPP,b, and GPPPxd may be attrib- of evergreen needleleaf forest during the plant-growing uted to three sources of errors. The tirst source is the season (Fig. 6). Another study also reported that canopy sensitivity of the VPM model to PAR and temperature. NDVI did not correlated with leaf net CO? uptake of Air temperature in October-November is relatively low, mature evergreen chaparral shrubs in 1998-1999 (Stylin- T,,,I,, may be over-corrected (smaller values), resulting in ski ct al., 2002). One interpretation of the observed lower light use efficiency (c). Selection of T,,,,, is likely to decrease of EVI during late summer and fall seasons is have some impact on 7;,,1,r, particularly in both early spring that less amount of PAR is absorbed by the PAV for and late fall seasons. In this study, we used T,,,,,,=O "C photosynthesis due to the aging process of leaves (possibly (Abcr St 1;ederer. 1992), while another process-based eco- an increase of NPV within leaves, changes in leaf structure system model used T,,,,,,=- 2.0 "C for temperate forest and pigments). Ongoing measurements in the California (Raich et al., 100 1). The second source is the error (over- chaparral suggest that evergreen shrubs undergo large estimation or underestimation) of observed GPP (GPPub,). seasonal changes in their leaf carotenoidchlorophyII ratios GPPOb, is calculated from field-measured NEE (NEEobs) (Sinis & Ganion. 2002). More field and laboratory studies and ecosystem respiration (R,,,, and RIllgI,,):NEEob,= GP- across leaf, canopy and landscape levels are needed to PPobs- (Rday+ Rnlgllr).For a given amount of NEE as better understand and quantify the relationship between measured by the eddy-covariance method, an error in improved vegetation indices (e.g., EVI) and FAPARpAvof estimation of R,,,, would result in an error in estimation of forests in the plant-growing season. In addition, progress GPP. The two major steps that must be taken to derive GPP has recently been made in using radiative transfer model- are the gap filling of NEE and estimation of daytime ing approach to develop advanced vegetation indices for ecosystem respiration. Both of these steps require subjective estimation of FAPAR, using the top-of-atmosphere (TOA) decisions and are currently the subject of a great deal of reflectance data from the VGT sensor (Gobron et al.. discussion (Falge et al.. 2001 ah). The third source is the 2000). When the TOA reflectance data from the VGT time-series data of vegetation indices derived from satellite sensor become freely available to users, it will be of images. We used the 10-day VGT composites that have no interest to compare those advanced vegetation indices BRDF correction or normalization, and thus, the effect of (Gobrot1 et al.. 2000) with the semi-empirical EVI, using angular geometry on surface reflectance and vegetation CO? flux data from the flux tower sites. indices remained. The NDVI-based conipositing method The second hypothesis in the VPM model is to use a used to generate the 10-day composite VGT image may satellite-based water index for estimating the water scalar also affect the time-series data of vegetation indices (EVI (Ws:,,,I,r) in calculation of light use efficiency (8). This and LSWI). Use of daily cloud-free VGT data may improve alternative approach differs from other PEM models that prediction of GPP by the VPM model. Further investiga- use soil moisture andor water vapor pressure deficit to tions are needed to quantify the relative role of individual adjust the water scalar (W5,,1,r) in calculation of light use sources of error in evaluation of the VPM model using CO? efficiency (E) (Field et al.. 1995: Pimce & Goward, 1995; flux data from tlux tower sites. Running et al.. 2000). To what degree canopy water content can be retrieved from satellite images is an important In the VPM model. we propose two smple but inno- \ vative ideas that could result in significant improvement in research question for remote sensing science I('han1pagne estimating seasonal dynamics of gross primary production et al.. 1003: Penuelas et al.. 1907; Sims 61 Ciamon. 2003). of evergreen needleleaf forests at large spatial scales. The While the best wavelength for prediction of canopy water first hypothesis in the VPM model is to use an improved content from ground-based data (no atn~osphericinterfer- vegetation index (e.g., EVI in this study) to estimate the ence) were 960 and 1180 nm, the best wavelength for fraction of PAR absorbed by photosynthetically active satellite remote sensing of canopy water content (with vegetation (PAV) for photosynthesis (FAPARpAV).which atmospheric interference) would be 1150--1260 and , clearly differ from the other PEM models that use NDVI to 1520 -1540 nm (Sims St Gamon, 1003). The availabdity estimate FAPAR (Potter cl al., 1993; Pr~ncec9i Goward, of time-series data of SWIR and NIR bands from the new 1995: ftunnmg et al.. 2000). Quantitatwe partition of generation of optical sensors (e.g., VGT, MODIS) offers vegetation canopy into PAV and NPV components, and new opportunity for quantifymg canopy water content at consequently partition of FAPAR into FAPARp*v and large spat~alscales through both the vegetation indices 532 X. Xfao rt al. / Renrofe Sensing of'Environn~ent89 (2004) 519-534 approach (Ccccalo ct al., 7002b) and the radiative transfer satellite-based modeling of GPP may have significant modeling approach (Zarco-'l'ejada ct al., 2003). Earlier implications on the study of carbon cycle processes of studies have shown that the Moisture Stress Index (MSI) evergreen needleleaf forests in both temperate and boreal and the Global Vegetation Moisture Index (VGMI) are zones. As the VGT sensor provides daily images of the sensitive to changes in equivalent water thickness (g/cm2) globe, there is a potential to use the VPM model and climate at leaf and canopy levels (Ceccato el al.. 2001, 3-001a,b: data (temperature and PAR) to quantify the spatial patterns Hunt & Rock, 1989). A water index calculated as a and temporal dynamics of GPP of boreal forests at 1-km normalized difference between MODIS band 6 (1628- spatial resolution. 1652 nm) and band 2 (841-876 nm) was compared to in situ top layer soil moisture measurement from the semiarid Senegal and the results showed a strong correlation between Acknowledgements the water index and soil moisture in 2001 (Fensholt & Sandhol, 2003). The preliminary comparison between We thank the International Paper for providing access to LSWI and foliage moisture content of evergreen needleleaf the research site in Howland, Maine, USA. John Lee, forest at Howland Forest, Maine (Pig. 8) has shown the Holly Hughes, and Jeremiah Walsh provided expert seasonal changes of leaf water content and sensitivity of assistance with the multi-yew C02 flux and climate data LSWI in the plant-grow~ng season. More fieldwork are set for the Howland Forest site (l~ttp:/~public.uml.gov/ needed to collect multiple-year data of leaf and canopy arnt.1-iflux.i'Datalindex.ch). We are grateful to the four water content of evergreen needleleaf forests over the plant- reviewers, and their comments and suggestions on the growing season, in support of the effort to retrieve canopy earlier versions of the manuscript greatly improved the water content through both the empirical and radiative manuscript. The C02 flux research at the Howland site transfer modeling approaches. In addition to improving was supported by the Office of Science (BER), US. quantification of the accuracy, adequacy and precision of Department of Energy, through the Northeast Regional water indices (e.g., LSWI), more field and laboratory work Center of the National Institute for Global Environmental are also needed to study the effect of leaf and canopy water Change under Cooperative Agreement No. DE-FC03- content on photosynthesis of evergreen needleleaf forests, 9OER6 1010. The modeling study was supported by so that our hypothesis about the relationship between leaf research grants from NASA Interdisciplinary Science water content and photosynthesis in the VPM model could Program (NAG5-10135) and Land Cover and Land Use be fully tested at the canopy level and over the plant- Change Program (NAGS- I 1 1 60). growing season.

References 7. Summary

Aber, J. D., S( Federer. C. A. (1992). A generalr7ed, lumped-paratneter The Wowland Forest site is representative of an ecotonal model of photosynthesis. evapotranspiration and net primary production boreal-northem hardwood transitional forest (Hollinger et in temperate and boreal forest ecosystems. Oecologio, 92, 463-474. al.. 1994). The eddy covariance measurements have shown Aber, J. D., Relch. P. B., & Goulden, M. L. (1996). Extrapolating leaf COz that evergreen needleleaf forest at the site had distinct exchange to the canopy: A generalized model of forest photosynthesis compared with measurements by eddy correlatton. Oecologia, 106, seasonal dynamics and moderate interannual variation in 257 265. GPP during 2998 2001. The response functions of C02 and Asncr. G. P., Wessman. C. A., & Archer, S. (1998). Scale dependence of ' water vapor exchange at Howland Forest due to climate absorpt~onof photosynthettcally active radiation in terrcstrral ecosys- variation are similar to those of other spruce forests (Goul- tems. Eculu~~calAppli~ctlruns, 8, 1003 - 102 1. den et al., 1997; Iiollinger zt al., 1909). The satellite-based Behrenfeld, M. J., Randelson. J. T.. McCla~n,C. R., Feldman, G. C., Los. S. O., Tucker, C. J.. Falkowskt, P. G.. F~eld,C. B., Frown, R., Esaras. VPM model uses two improved vegetation indices (EVI and W. 8.. Kolber. D. D, & Pollack. h. H. (2001). B~ospher~cprlmaly LSWI) that can be generated only from the new generation production during an ENS0 ttmsttlon. Sc~ence,291. 2594-2507 of optical sensors (e.g., VGT), which has the potential to Hoegh, 6.. Soegaard, H.. Broge, N.. Hasager, C'. B ,Jensen. N. 0.Schelde. provide major improvement over the current satellite-based K., & Thomsm, A. (2002). Airbonie mult~spectraldata for quant~@~ng Production Eftic~encyModels that only employ NDVI. The leaf area index. nitrogen concentration, and photosynthetic efliciency in agriculture. Remote Se~uingof Environment. XI, 179- 193. VPM model is capable of tracking seasonal dynamics and Ceccato. P., Flasse. S., & Gregoire, J. M. (2002a). Designing a spectral interannual variations in GPP of evergreen needleleaf forest index to estimate vegetation water content from remote sensing data: at a sub-monthly (10-day in this study) temporal resolution. Part 2. Validation and applications. Remote Sensing cfEnviroinne~rt,82. Additional studies are needed to continue validat~onof the 198--207. VPM model across various forest tower sites, that is, cross- Ceccato, P., Flasse. S., Tarantola. S., Jacquernoud. S., & Grcgoirc. J. M. (2001). Detecting vegetation lcaf water content using reflectance in the biome con~parisonand cross-site comparison within a optical domain. Rentute Sr,ui~~gof'Eirr.ironmc.nt, 77. 22 ---33. biome type. Calculation of GPP 1s the first step in the study Ceccato, P., Gobron, N., Flasse, S., Pin% D., & Tarantola. S. (2002b). of carbon cycle of terrestrial ecosystems. Our progress in Designing a spectral index to estimate vegetation water content frotn rcmote senslng data: Part 1. Theoret~calapproach. Reniote Senrzng u/ companson of vegetatlon ~ndicesglohal set of TM Images for EOS- Envtroninent, 82, 188 - 197 MODIS. Remote Sensrng of Envrron~nent,59, 440-451. Champagne, C. M., Stacnz. K., Bannan. A,, McNairn, H., & Degusc, J.-C. Hunt, E. R., & Rock, B. N. (1989). Dctect~onof changes In leaf watcr- (2003). Val~dat~onof a hyperspectral curve-fittmg model for estlmatlon content using near-~nfrared and mddle-inbred reflectances. Reinote of plant water content of agr~culturalcanopies. Remote Sensrrtg oj'En- Srnsntg oJ Enviroi~ment,30, 43-54. vrrorrn~ent,87, 148 - 160 Justwe, C. 0.. Vermote, E.. Townshend, J. R. G, Defies, R., Roy, D. P., DeLucia, E;. H., & Sm~th.W. K. (1987). Air and soil temperature lim~ta- Hall, D. K.. Salomonson, V. V., Pnvette, J. L., R~ggs,G.. Strahler. A,. tions on photosynthesis in Englemann spruce dunng summer. Canad~un Lucht, W., Myneni, R. B., Knyazikhin, Y., Runnmg, S. W., Nemani, Jour?~rrloj Fore~rResearch. 17, 527-533. R. R., Wan, Z. M., Huete, A. R., van Leeuwen, W., Wolfe, R. E., Falge, E.. Baldocchi, D., Olson, R.. Anthom, P., Aubinet, M., Bernhofer, C.. Gglio, L.. Muller. J. P., Lewis, P.. & Barnsley. M J. (1998). The Burba, G ,Ceulemans, G., Clement. I<.. Ilohnan, H., Granier, A., Gross, Moderate Resotut~onImagrng Spectroradrometer (MODIS). Land re- P, Gru~lwald,T, Hollinger, U. .lensen, h' 0. Katul, G.. Keronen, 1'. mote senalng for global change ~eseareh /LEE 7?airsactronr on Cko- Kowalsk~.A.. Lai. C. T., Law. B E.. Meyers. T., Moncneff, J.. Moors, swetre and Rsatote Se~rsrrrg,36. 1228-1249. . E., Munger, J. W., Pdegnard, K., Ranmk. U., Rebmann, C., Suyker, A., Law, B. E., Falge, E.. Gu, L., Baldoechi, D. D.. Bakwln, P.. Be~b~g~er.P.. Tenhunen, J., Tu, K., Venna, S., Vesala, T., W~lson,K., & Wofsy, S. Davis, K., Dolman, A J , Falk, M., Fuentes. J. D., Goldste~n,A., (200la). Gap fillmg strategies for long tenn energy flux data wts. Agri- Granter, A., Grelle, A,. Holhnger, D., Janssens, 1. A., Janis. P., Jensen. cultural and Fore.st hleteo~alo~g,107. 7 1 - 77 N. 0, Katul. G , Mahl~,Y., Matteucci, G., Meyen, T. Monson, R . Falge, E., Baldocch~,D., Olson, R.. Anthon~,P.. Aubmct, M , Bemhofer, Munger, W., Occhcl. W, Olson, R.. Pilcgaard, K., Paw, K. T., Thor- C., Burba, G.. Ceulemans. R., Clcment. R., Dol~nan,H , Gran~er.A, geirsson, H.. Valent~n~,R., Venna, S., Vesala, T., W~lson,K., & Wofiy, Gross. P., Grunwald. T.. Hollnigcr. D , Jensen, N 0 , Kah~l,G., Ker- S. (2002). Envnontnental controls over carbon d~oxidcand water vapor onen, P., Kowalshr, A,, Lnl, C. T., Law, B. E.. Meyers, T., Moncr~efE exchange of te~restrcalvcgetatcon. A~pcrdtt~ruland Forat Meteorology. H., Moors, E., Munge~,J W, Pdegaard, K., Ratunk, U , Rebmann. C, 113. 97- 120. Suyker, A., Tenhunen, .I. Tu, K., Vemia, S., Vesala, T., Wdson, K., & Law, B. E., Wanng. R. H . Anthom, P. M., & Aber, J. D. (2000). Meawe- Wofsy, S. (2001b). Gap fill~ngstrategies for defens~bleammal sums of lnents of gross and net ecosystem productivity and water vapour ex- net ecosystem exchange 4grrcultural and hbrest Meteorology. 107, change of a Pinur porirlerosn ecosystem. and an evaluation of two 43-69 generalized models. Global C%airge Biology, 6. 155- 168. Fensholt, I<.,& Sandhol. 1 (2003). Derivation of a shortwave rnfrared Malmstrorn, C M.. Thompson. M. V., Juday, G P.. Los, S. O., Randerson, water stress ~ndexfiorn MOIIIS near- and shortwave mfrared data In J. T.. & Field, C. B. (1997). Interannual variation in global-scale net a sennand environment. Remote Senwzg ojWrvirc,n~trmt, 87, 11 1 - 12 1. pnmary productm: Testmg model estlrnates. Glohnl Brogeochemical F~eld,C. B., Bchrenfeld, M. J , Randerson. J. 7.. & Falkowskt, P. (1998). Cjrles, 11, 367-392. Prnnary product~onof the b~osphere.integmtmg terrestr~aland oceanlc Melillo, J M.. Mcguire, A. D , Kicklighter, D. W., Moore, B , Vorosmarty, components. Sctmce, 281, 237 240. C. J., & Schloss. A. L. (1993). Global chmate-change and terrestnal net F~eld,C. B., Randerson, J. T., & Malrnstrom, C. M. (1995). Global net primary production. A'rrture, 363, 234- 240. prrmary prod~~ct~on-co~nb~n~ngecology and remote-senstng. Remote Penuclas, J., Pinol, J.. Ogaya, R.. & Filclla, I. (1997).*Estnnation of plant Sensing oj Enwro~imerrt,51, 74 88. water concentration by the retlectance water ~ndexWI (R900/R970). Frolkmg, S. E., Bub~et.J. L., Moore. T. R., Ball. T., Belhsano, L. M., InternotionfllJournal oj'Rernote Sensmng, 18, 2869-2875. Bhardwaj, A., Carroll, P., CIIII, P. M.. Latleur, P. M., McCaughey, 1. H., Potter, C. S., Randerson, J. T., Field, C. B., Matson, P. A., Vitousek, P. M., Roulet, N. T.. Suyker, A E., Verma, S. B.. Waddington. J. M , & Mooney. H. A., & Klooster, S. A. (1993). Terrestrial ecosystem pro- Wh~ting,G. J (1998). Relat~onslnpbetween ecosystem p~oduct~v~ty duction-a process model-based on global satellite and surface data. and photosynthet~callyactwe rad~at~oufor northem peatlands. Glohol Global Biogeockeniztol Cvcles, 7, 8 1 1 -84 1. Bmgeoc-liemrcnl Cjrler. 12. 1 1 5 - 126. Piince, S. D., & Goward, S. N. (1995). Global primary product~on A Qao. H C (1990) NDWI-A nonnalved difference water mdex for Ie- remote sensmg approach. Journal qf Biogeography, 22, 815-835. mote senslng of vegetat~onItquid water from space. Remote Smsmg of Ra~ch,J. W., Rastetter, E. B., Mehllo, J. M., K~cklrghter.D. W., Steudler. Environment, 58, 257-266 P A., Peterson, B. J., Grace, A. L, Moore, 8.. & Vorosmarty. C. J. Ciobron. N. P~nty,B., Verstraetc, M., & Govaerts, Y. (1999). The MERE (1 991). Potentla1 net pnmary productivity In South-America-applica- Global Vegetation Index (MGVI): Descnpt~onand prel~nnnaryappl~ca- tlon of a global-model. Ecol~,Kitul.4pplzcattonv,1, 399-429. tion. Iritenzcrtional Journal of Remote Sensrng, 20. 19 17 - 1927. Rock, B N., W~llrarns,D. L., Moss. D. M., Lauten, G. N., & Ktm, M. Cobron, N.. Pinth B.. Verstraete, M M., & Wtdlowski, J. L. (2000). ( 1994). H~gh-spectral-resolut~onfield and laboratory optical reflectance Advanced vegetation indices opt~m~zedfor up-coming scnsors. Design. measurements of red spruce and castcrn hcmlock necdles and branches. performancc, and applreatrons IEtE ljansactron~on Geoscrence and Remote Sensing of Envrronmml, 47, 176- 189. Renrote Sensrng, 38. 2489-2505. Rum~y,A,. Dcd~eu.C; , & Saug~er,B (1996). TURC A diagnoshc model Goulden, M L. Daube, B C, ran, S. M.. Sutton, D J, Bazzaz. A, of contmental gross primacy product~v~tyand net pllrnary product~vity Munger. J. W, 6t WoKsy, S. C. (1997) Phys~olog~calresponses of a Glohul Bzogeuc/ie~n~a/Cycles, 10. 269- 285. black spice forest to weather. Journal oj Geopliytrral Rerea~ch-Artnos- Rumy, A. Jan~s,P G , Baldocch~,U. D.. & Saug~et; B. (1995). C02 plierer. 102, 28987-28996 fluxes over plant canopes and solar radiation. A revlew Advances 111 Govaerts. Y. M , Ve~

Running. S. W., Thornton, P. E., Nemani, R., & Glassy, J. M. (2000). Tucker, C. J. (1980). Remote-sens~ngof leaf water-content in the near- Global terrcstnal gross and net prlmary productiv~tyfrom the Earth ~nhred.Remote Sens~ngof Envilanm~iI,10, 23-32. Obscrving System. In 0 E. Sala, R. B. Jackson, H. A. Mooncy, Sr Turner, D. P., Urbansk~.S., Brerner, D., Wofsy. S. C., Meycrs. T., Gower, R. W. Hoowarth (Eds.). Methods in ecosle?n scrence (pp. 44 57). S. T., & Gregory, M. (2003). A cross-biomc comparison of daily light New York: Sprmgcr. use efficiency for gross primary production. Globul Change Biolu~q: Schulze, E. D., Lloyd, J., Kelhher, F. M., Wirth. C, Rebmann, C., Luhker, 9, 383 395. B., Mund. M., Knohl, A,, Milyukova, 1. M.. Schulze, W.. Ziegle~,W, Weiss. A,, & Nonnan, J. M. (1985). Partitioning solar-radiation Into direct Varlagm, A. B., Sogachev, A. E. Valentini. R., Dole. S.. Grigonev, S., and d~fhse,visible and near-infrared coniponents. Agriculrurol nnd Kolle, O., Panfyorov. M. I., Tchebakova, N., & Vygodskaya. N. N. hresr Meteo~olugj:34. 205-21 3. (1999). Productivity of forests In the Eurosiberian boreal region and W~lson,K. B.. Baldocchi, D. D., B Hanson, P. J. (2001). Leaf age affects their potentla1 to act as a carbon sink-a synthes~s.Glohnl C%unge the seasonal patterns of photosynthetic capaclty and net ecosystem eu- Biology. 5, 703-722. change of carbon in a dec~duousforest. Plant, Cell and Environment, Senano. L., Gamon, J. A,, & Penuelas, J. (20003). Estimat~onof canopy 24. 571 -583. photosynthet~cand nonphotosynthetlc components from spectral t~ans- Xiao, X., Boles, S., Frolklng. S.. Salas, W., Moore, B.. Li, C., He, L., & mlttance. E~ology,XI, 3 149- 3 162. Zhao, R. (2002a). Landscape-rcale charactenmtlon of cropland in Chi- Senano, L., Ustin, S. L., Roberts, D. A., Gainon, J. A., & Pcnuelas, J na using Vegetation and Landsat TM images. Internotronul .k~urliulof (2000b). Deriving water contcnt of chapad vegetation &om AVlRlS Remote Sens~ng,23, 3579-3594. dat,l. Rmlote Sensrng of Envirc,nment, 74. 570 581. Xiao, X., Boles, S., Frolkmg, S., Salas, W., Moore, B., Li, C., He, L., Sr Si~ns,D A., & Gamon. J. A. (2002) Relationships bctwcen Icafp~gment Zhao, R. (2002b). Observation of flooding and ncc tnnsplanting of contcnt and ~pectralreflectance across a wide wngc of spccics, leaf paddy rice fields at the slte to landscape scales in China using VEGE- structures and devclopmcntal stages. Remole Srnsrng of Environn~mt. TATION sensor data. Inte~nntzorrc~lJoirrnul of Remote aSensi~~g,23. 81, 337-354. 3009-3022. Sims, D. A,, & Gamon, J. A. (2003). Estrmat~onof vegetation water con- Xiao, X., Boles, S., Liu, J. Y., Zhuang, D. F., & Liu, M. L. (2002~). tent and photosynthetic tissue area from spectrnl reflectance: A compar- Characterizat~onof forest types In Northeastern China, using multi- ison of mdlces based on liquid water and chlorophyll absorption temporal SPOT-4 VEGL TATION sensor data. Rw~oteSensing of E~ni- features. Remote Seruing of ~,'nviimnment,84, 526-537. ronment, 82, 335-348. Stylinslti, C. D., Gamon. J. A., & Oechel, W. C. (2002). Seasonal patterns Xiao, X.. Braswell. B, Lhang, Q.. tEoles, S.. Frolkmg, S., & Moore, H. of reflectance indices, earoteno~dpigments and photosynthes~sof ever- (2003) Sensitivity of vegtatlon ind~cesto atmospheric aerogels: Con- green chaparral species. Oecologirr, 13 /, 366-374. tinental-scale observations In Northcrn Asla. Rsnote Senszng of'Eirvr- Sullivan, J. H. Bovard, B. D., & Mlddleton, E. M. (1997). Vanab~htyin ronment, 84, 385-392 leaf-level C02 and water fluxes In Prriur hunknancr and Prcen nrnriuncr Zarco-Tejuda. P. J., Rueda. C. A,, & Ustin, S. L. (2003). Water content in Saskatchcwan. Tkee Plysiologv, 17, 553 56 1. cstnnation In wgctatlon w~thMODlS reflectance data and model In- Tucker, C. J. (1979). Rcd and photograph~cmfrarcd hnear combinations for version methods. Remote Sensing of Environrnenl. 85. 109%124. monltormg vcgctatlon. Retnote Sensing of Environnlrnt, 8, 127 150.