t 109 (2007) 416–428www.elsevier.com/locate/rse

Remote Sensing of Environmen

The effect of spatial resolution on the accuracy of leaf area index estimationfor a forest planted in the desert transition zone

Michael Sprintsin a, Arnon Karnieli a,⁎, Pedro Berliner a, Eyal Rotenberg b,Dan Yakir b, Shabtai Cohen c

a The Remote Sensing Laboratory, Jacob Blaustein Institute for Desert Research, Ben Gurion University of the Negev, Sede Boker Campus 84990, Israelb Weizmann Institute of Science, Rehovot 76100, Israel

c Institute of Soil, Water and Environmental Sciences, Volcani Center, Bet Dagan 50250, Israel

Received 5 January 2006; received in revised form 25 January 2007; accepted 27 January 2007

Abstract

An approach is presented for determining leaf area index (LAI) of a forest located at the desert fringe by using high spatial resolution imageryand by implementing values from a moderate spatial but high temporal resolution sensor. A 4-m spatial resolution multi-spectral IKONOS imagewas acquired under clear sky conditions on March 25, 2004. Normalized differences vegetation index (NDVI) and a linear mixture model wereapplied to calculate fractional vegetation cover (FVC). LAI was calculated using a non-linear relationship to FVC and then compared with groundtruth measurements made in ten 1000 m2 plots using the tracing radiation and architecture of canopies (TRAC) canopy analyzer under bright andclear sky conditions during March and April, 2004. Calculated LAI, corrected with a measured clumping index, was highly correlated withmeasured LAI (R2=0.79, pb0.01). This approach was used to produce a 4-m resolution LAI map of the forest. The procedure was then applied tothe MODIS 250-m resolution surface reflectance product, where MODIS LAI and VI products were used to calculate the extinction coefficient byinversion of the LAI-FVC relationship, and the extinction coefficient was then used to calculate LAI for moderate resolution. Histograms ofresulting LAI distributions and descriptive statistics at the different spatial resolutions are compared. LAI spatial distribution at lower resolutionwas similar to that obtained at higher resolution and remained close to being normally distributed.© 2007 Elsevier Inc. All rights reserved.

Keywords: LAI; Dry-lands forestry; Remote sensing; NDVI; IKONOS; MODIS

1. Introduction

Forest stand description includes factors related to eco-physiological processes responsible for forest growth. One ofthose factors is the stand leaf area index (LAI), defined byWatson(1947) as the total one-sided area of leaf tissue per unit groundsurface area (m2/m2). LAI is highly related to many processes(e.g. rain interception, evapotranspiration, photosynthesis, respi-ration, and leaf litterfall) and used as an input to various ecosystemmodels that produce detailed information about vegetation coverand condition (Jackson et al., 1983; Bonan, 1993). LAI can beestimated through different direct (contact) and/or optical indirect(non-contact) procedures (e.g. measurements of light transmis-

⁎ Corresponding author. Tel.: +972 8 6596855; fax: +972 8 6596805.E-mail address: karnieli@bgu.ac.il (A. Karnieli).

0034-4257/$ - see front matter © 2007 Elsevier Inc. All rights reserved.doi:10.1016/j.rse.2007.01.020

sion through canopies) that are well reported in the literature (e.g.Breda, 2003; Gower et al., 1999). Despite their extensive usage,these methods demand considerable amounts of labor and time,and yield information only for the close vicinity of the measuredpoint. Therefore satellite remote sensing systems, which provideextensive spatial information for different forest phenomena, havebeen proposed as a good solution for measurement of forestparameters and statistics at different scales (Stoms, 1994; Bergen& Dobson, 1999; Ceccato et al., 2001; Santoro et al., 2002; Sims& Gamon, 2003). Consequently, remote assessment of LAI is ofmajor interest (e.g. Andersen et al., 2002; Brown et al., 2000;Carlson & Ripley, 1997; Clevers, 1997; Colombo et al., 2003;Gong et al., 2003; Wang et al., 2004; Wang et al., 2005a,b).

Today LAI is supplied as an operational product of ModerateResolution Imaging Spectroradiometer (MODIS) onboard EOS(Earth Observation System) Terra and Aqua satellites. This

417M. Sprintsin et al. / Remote Sensing of Environment 109 (2007) 416–428

product was developed primary for global studies and conse-quently available only at 1-km resolution at 8-day intervals (Wanget al., 2004).However, that low spatial resolution data is of limitedutility for local scales especially in arid and semi-arid forestedareas that are mostly planted, non-timber, sparse, spread oversmall areas (no more than 3000–5000 ha) and usuallyaccompanied by agricultural activity of local farmers. This highlandscape heterogeneity can be completely dissolved in 1 km2

pixels and consequently, it could be concluded that such areas arenot well represented in readily available global data and requiremore detailed surface representation by high spatial resolutionsatellite images, such as IKONOS. This sensor provides arelatively new source of data for monitoring various environmentprocesses represented by “pure” small pixels and can be used as avalidation core for low andmoderate resolution imagery (that wasoriginally proposed for global studies), in which each pixel maybe made up of many land-cover types (Colombo et al., 2003) andrequires detailed validation for local implementation.

This validation problem is of interest and has been addressedin different studies. Most of it shows a successful application ofLAI retrieved from low spatial resolution imagery (e.g.AVHRR, MODIS, or SPOT VEGETATION) as compared toradiative transfer based (Tian et al., 2002a,b), field-sampleddata collected (Privette et al., 2002), or high spatial resolutionIKONOS (Morisette et al., 2003) or ETM+ (Cohen et al., 2003)data. However, though the use of high-resolution sensor data isattractive, this data is still very expensive. Thus for effective

Fig. 1. Landsat-TM of central Israel. Note the location of the Yatir forest on the desetones (semi-arid zone).

monitoring of terrestrial ecosystems for a long period of time itis more convenient to use low cost high temporal resolutiondata.

As an opposite of other 44 MODIS products, surfacereflectance data for red and near-infrared (NIR) bands is readilyavailable to the scientific community at higher spatial resolution(250-m per pixel). This provides more detailed surfacedescription and hence seems to be more promising to dry-landforestry studies. Consequently, our objective was to propose amethod for small scale LAI monitoring based on frequentlyavailable MODIS data. We utilized operational MODISproducts as sources of ancillary information for precisedescription of canopy architecture through the determinationof canopy extinction coefficient, and finally compare the resultsto LAI assessed using high-resolution IKONOS data.

2. Study area

The study was conducted in Yatir forest (31°21′ N and35°02′ E, 630 m AMSL; ∼3000 ha area) located in thetransition between arid and semi-arid climatic zones at the edgeof the Negev and Judean deserts (Fig. 1). The mean annualprecipitation (275–280 mm) usually occurs during ∼30 daysyear−1 between November and March and is characterized bylarge annual fluctuations and an uneven distribution of theevents within the rainy season. The average total annualpotential evapotranspiration is 1600 mm year−1, yielding a

rt fringe, visible as the sharp contrast between bright tones (arid zone) and dark

418 M. Sprintsin et al. / Remote Sensing of Environment 109 (2007) 416–428

long-term aridity index (potential evapotranspiration/precipita-tion; Budyko, 1974) of ∼5.7. Regions where aridity index isgreater than unity are broadly classified as dry since theevaporative demand cannot be met by precipitation (Arora,2002). Specifically, an aridity index between 5 and 12 is usuallyclassified as arid (Ponce et al., 2000).

The hottest month in the Yatir region is July and the coldest isJanuary. The average maximum and minimum temperatures are32.3 °C and 6.9 °C, respectively (Schiller & Cohen, 1998). Theforest was planted mostly during 1964–1969, is almost even-aged, and is close to being a monoculture dominated by Aleppopine (Pinus halepensis Mill.). The trees grow on shallowRendsina soil and lithosols (0.2–1 m deep) overlay chalks andlimestone. The groundwater table is deep (∼300m) and little andsparse understory vegetation develops during the rainy season anddisappears shortly thereafter (Grunzweig et al., 2003).

3. Theory

3.1. Fractional vegetation cover

Reflectance spectra derived from satellite-based sensorsusually constitute mixed signals of a number of endmemberssuch as vegetation, bare soil, and shadow (Richardson&Wiegand,1977; Carpenter et al., 1999; Graetz & Gentle, 1982; Pax-Lenneyet al., 2001; Strahler et al., 1986; Xiao&Moody, 2005). Assumingthat the spectral signature of a given pixel is the linear, proportion-weighted combination of the endmember (a pure surface materialor land-cover type that is assumed to have a unique spectralsignature) spectra (Xiao & Moody, 2005) Spectral MixtureAnalysis (SMA) has been used to estimate canopy proportionsfrom multi-spectral satellite data at sub-pixel level (Roberts et al.,1998; Wu & Murray, 2003). Endmember signatures can bedirectly selected from the image (image endmembers), orextracted from field or laboratory spectra of known materials(reference endmembers; Ichoku & Karnieli, 1996).

A simplified SMA model is the two-endmember model(Wittich & Hansing, 1995) that assumes that a given pixelconsists of only green vegetation and bare soil, and thus itsspectral vegetation index (VI) value is the linear combination ofcontributions from these two components. Such a modelconsists of a single equation, which improves computationalefficiency by simplifying the process of endmember selection.Consequently, surface reflectance measured by a satellite (ρ)can be taken as a weighted sum of canopy and backgroundreflectance (ρc and ρb respectively):

q ¼ qcdfc þ qbdð1−fcÞ ð1Þ

where fc is fractional vegetation cover, which is a summation ofcrown areas as seen from above.

Fractional vegetation cover is an important element inmodels that attempt to account for the exchanges of carbon,water, and energy at the land surface (Nemani & Running,1996; Ward & Robinson, 2000) since the change in vegetationpattern has a feedback influence on the local and regionalclimate by changing the patterns of evaporative water losses

from the surface to the atmosphere. fc is required for theparameterization of surface conditions and for modeling andland cover/land use change studies.

Rearranging (1) and representing the background and vegeta-tion reflectance by the minimum and maximum values of thenormalized difference vegetation index (NDVImin and NDVImax

respectively) yields (Tucker, 1979; Choudhury et al., 1988;Carlson & Arthur, 2000; Carlson & Ripley, 1997; Che & Price,1992; Gutman & Ignatov, 1998; Price, 1987; Rouse et al., 1974):

fc ¼ NDVI−NDVIminNDVImax−NDVImin

ð2Þ

A common approach to the retrieval of maximum andminimum NDVI values is through time series analysis.However, leaf optical properties can vary during the yearindependently of LAI, e.g. from airborne dust cover common indeserts (Derimian et al., 2006; Ganor, 1994). It follows thatparameters of Eq. (2) are site and time specific and thereforeshould be estimated locally, as discussed in Section 4.5.1.

3.2. Leaf area index

The fraction of incident light transmitted through a canopy(τ) is described by:

s ¼ expð−k⁎LAIÞ ð3Þwhere k is the extinction coefficient for the canopy and LAI is theleaf area index of randomly distributed leaves (effective leaf areaindex), which is for non-clumped canopies the same as the actualleaf area index (Chen et al., 1991). Note that for clumped canopiesactual leaf area index is larger than effective leaf area index.

Assuming that the tree crowns are opaque and the onlytransmission is the transmission through the gaps in the canopy,Eq. (3) can be rewritten as (Kucharik et al., 1999)

s ¼ 1−fc ð4Þ

where fc is the same as in Eq. (1).Consequently, for known values of fc and k, LAI could be

calculated as (Norman et al., 1995):

LAI ¼ −lnð1−fcÞk

ð5Þ

where k accounts for clumpiness.

3.3. Extinction coefficient

The extinction coefficient is a measure of attenuation ofradiation in the canopy. It is a function of wavelength, radiationtype, and direction, as well as stand structure and canopyarchitecture (Jarvis & Laverenz, 1983; Jones, 1992). Theextinction coefficient can be expressed as:

k ¼ Gðh; aÞXcosðhÞ ð6Þ

Fig. 2. A flowchart of LAI determination for moderate resolution data.

419M. Sprintsin et al. / Remote Sensing of Environment 109 (2007) 416–428

whereG(θ) is the foliage projection coefficient (or “shape factor”)characterizing the foliage angular distribution (α; Norman &Campbell, 1989), Ω is the clumping index and θ is the sensorviewing angle.

When no other data is available most applications assume thatleaves are randomly distributed in the horizontal plane andsymmetrically distributed around the azimuth (Coops et al.,2004), and thus the leaf angle distribution parameter (the ratio ofthe vertical to horizontal canopy elements) is unity andG(θ)=0.5.However, for more complex and clumped sparse canopies ofconifer stands this assumption does not hold, since within suchcanopies the leaves are typically not uniformly distributed andnon-randomness of the canopy makes the extinction coefficienthighly variable (usually lower) (e.g. Cohen et al., 1995). Thus therandom assumption is an idealization applicable only wheninformation on canopy structure is unavailable, while it ispreferable, especially for sparse and clumpy vegetation of aridregions, to determine k experimentally.

Different models based on probability statistics have beenused to estimate the extinction coefficients for homogeneous(e.g. Cowan, 1968; Miller, 1967; Monteith, 1969) and clumped(e.g. Nilson, 1971) canopies. Moreover, most available kestimates have been obtained for specific sky conditions andcanopy types (Cohen et al., 1997), which may make theminappropriate under other conditions than those prevalentduring their derivation. For this reason we used readily availableglobal data products determined by remote sensing (operation-ally produced MODIS LAI and NDVI 8-day compositeproducts) combined with easily available in-situ measurementsas sources of information in order to invert Eq. (5) and solve fork as follows:

kMODIS ¼ −lnð1−fc;MODISÞLAIMODIS

ð7Þ

where

fc;MODIS ¼ ðNDVIMODIS−NDVIminmeasÞ

ðNDVImaxMODIS−NDVI

minmeasÞ

ð8Þ

and the subscript MODIS stands for data obtained from theMODIS platform for a given date, the superscript max indicatesthe maximum value of a specific scene and NDVImeas

min stands forthe ratio computed from actual in-situmeasurements carried outwith field spectroradiometer using data that correspond to visualgaps within the IKONOS scene (Section 4.5.1).

We use krandom in conjunction with data obtained from thehigh-resolution platform (IKONOS) in order to obtain LAI athigh spatial resolution, as follows:

LAIIKONOS ¼ −lnð1−fc;IKONOSÞkrandom

ð9Þ

where

fc;IKONOS ¼ ðNDVIIKONOS−NDVIminmeasÞ

ðNDVImaxIKONOS−NDVI

minmeasÞ

ð10Þ

with the sub and superscripts min and meas having the same

meaning as above but obtained from the corresponding IKONOSimages (Fig. 2).

4. Materials and methods

4.1. Sampling design

Ten plots (Fig. 3) of ∼1000 m2 (approximately 32×32 m;8×8 IKONOS pixels; each, were chosen in order to collectground truth data. Each plot was divided into seven parallelEast–West oriented ∼32 m long and ∼4.6 m wide sub-plots.The total length of transects made along those sub-plots totaledapproximately 200 m in length. This length matches therecommended lengths for the use of the TRAC LAI_meter (i.e.100 to 300 m.; Leblanc et al., 2002).

The location of each plot perimeter was mapped in a fieldcampaign using a differential Global Positioning System (GPS)with ±2 m accuracy. The resulting polygons were adopted into aGIS vector layer.

4.2. Canopy structure: ground truth measurements

LAI measurements were carried out with the tracingradiation and architecture of canopies (TRAC) photo-sensordevice and software (Chen & Cihlar, 1995a,b; Chen et al.,1997). A transect was measured along the length of each of theseven sub-plots, and mean LAI was calculated for each plot.Transects were repeated three times a day for different solarzenith angles during March and April 2004.

TRAC design and operation are described in Chen andCihlar (1995a,b), Chen et al. (1997), Eriksson et al. (2005), andLeblanc et al. (2002, 2005). The instrument was designed toovercome the bias resulting from the clumpy nature ofdiscontinuous canopies (and particularly coniferous forests)that occurs at several scales, between plants within a stand, andbetween branches or shoots within plants, and has been found invarious studies to be a major factor of LAI underestimationwhen using non-contact indirect methods (e.g. Chen et al.,1997; Cohen et al., 1995; Kucharik et al., 1997; Lang, 1986,1987; Welles & Cohen, 1996).Clumping can be dealt with by

Fig. 3. IKONOS image of the Yatir forest. The rectangle indicates an approximately 1 km2 area surrounding the FluxNet site described by Grunzweig et al. (2003). Thewhite dots represent the location of the sampling plots.

420 M. Sprintsin et al. / Remote Sensing of Environment 109 (2007) 416–428

introducing a clumpiness factor, Ω (Eq. (6)), which representsthe ratio of effective leaf area index (LAIe), estimated withouttaking into account clumpiness, to true LAI (LAIt), i.e.:

LAIt ¼ LAIeX

ð11Þ

Clumping in the forest stand is related to stand density,which is usually expressed as the number of trees per unit areaand is sometimes taken as a coefficient (Daniel et al., 1979).Density always expresses a relationship between the number ofplants and their sizes (Zeide, 1995). Consequently, the clumpingcorrection should be inversely proportional to stand density.However, measuring density by remote sensing is not easy andrequires more rigorous analysis as even high-resolution space-born sensors have difficulty in detecting single trees in thestand. Therefore, a quantitative analysis of stand density wasout of the scope of this study and Ω was determinedindependently from the distribution of gap sizes measured byTRAC (Chen & Cihlar, 1995a,b). We were not able to perform adestructive sampling in order to calibrate TRAC. ATRAC datawas compared to LAI estimated from leaf litter. The latter wassystematically collected at three of the ten sample plots betweenOctober 2001 and October 2002 (see Section 5.2.2 for details)assuming that leaves did not lose dry weight during thesenescence thus leaf litter dry mass equals leaf dry mass(Grunzweig et al., 2003). These comparisons show negligible

(less then 1.5%) differences between LAI computed fromTRAC measurements (1.33 m2 m−2) and LAI estimated fromleaf litter collection (1.35 m2 m−2) and we consequently will usethe TRAC data in lieu of LAI ground truth.

4.3. Stand structure: direct measurements

In addition to LAI, standard tree biometrics includingequivalent diameter at breast height (DBH), crown diameter(CD), and tree height (H) were measured using a caliper, ameasuring tape and a clinometer, respectively, during a specialfield campaign in April–May 2004. The tree canopy wasassumed to be circular, thus CD was converted to crown area(CA), which was aggregated per plot. The canopy cover (CC) ofa plot was calculated as the ratio of the sum of crown area of alltrees within a plot to plot area. Stem density was defined as thenumber of trees per plot (trees/pl). Field measured parametersand a comparison of the training plots are presented in Table 1.

4.4. Image data retrieval

4.4.1. High spatial resolution dataA 4-m spatial resolution multi-spectral IKONOS image was

acquired onMarch 25, 2004. IKONOS is the first high-resolutioncommercial satellite that simultaneously captures 1-m panchro-matic (black and white) and 4-m multi-spectral (color) digitalimagery (blue, green, red, and near infrared). The panchromatic

Table 1Characteristics of the ground truth plots

PlotID

LAI(m2/m2)

Density(trees/pl)

DBH(cm)

CD(m)

CA(m2)

HEIGHT(m)

CC(%)

1 1.47 23 19 5 19 10 442 1.74 25 16 4 15 8 373 1.80 25 20 5 21 10 524 1.33 30 15 4 17 8 515 1.30 33 14 4 12 7 416 1.60 35 16 8 17 8 677 1.50 35 16 4 15 8 528 2.01 35 15 4 14 8 489 1.42 40 12 4 10 6 3910 2.39 40 15 4 13 8 50

421M. Sprintsin et al. / Remote Sensing of Environment 109 (2007) 416–428

and multi-spectral datasets are available separately or can bepurchased as combined 1-m (or coarser) multi-spectral images.Onboard sensors can point both along and across the satellitetrack, providing a revisit frequency of 1–3 days.

This image was radiometrically and atmospherically cor-rected following Space IMAGING® specifications (http://www.spaceimaging.com/products/ikonos/spectral.htm) and 6S radia-tive transfer model (Vermote et al., 1997) respectively, andregistered into the UTM projection by using Geographic ControlPoints.

4.4.2. Moderate and low spatial resolution dataMODIS provides high temporal resolution (every one to two

days), radiometric sensitivity in 36 spectral bands ranging inwavelength from the visible (VIS) to the short-wave infrared(SWIR), and moderate to low spatial resolution (from 250 to1000 m). The 44 standard MODIS products are available to thescientific community through the Earth Resources ObservationSystem (EROS) Data Active Archive Center (DAAC), usuallyat 1000-m spatial resolution. Since MODIS was originallyproposed for global studies (which explain its moderate/lowspatial resolution) it requires detailed validation for localimplementation.

Three standard MODIS products (Collection 4) were used:

• 1-km global data LAI product (MOD15A2) updated every8 days in order to eliminate the contamination from cloudcover (Knyazikhin et al., 1999). This product's algorithmuses vegetation maps built on the basis of six major biomes(cereal crops, shrubs, broad-leaf crops, savannas, broad-leafforest and needle-leaf forest) to constrain the vegetationstructural and optical parameter space (Fang & Liang, 2005).The main MODIS LAI algorithm is based on three-dimensional radiation transfer theory while LAI is retrievedby comparing the observed and modeled bidirectionalreflectance factor (BRF). The latter utilizes a soil reflectancemodel (Jacquemound et al., 1992) for each biome for varyingsun-view geometry and canopy/soil patterns. Moreover, abackup algorithm based on a straightforward NDVI-LAIrelationship is used when the main algorithm fails.

• 1-km global MODIS/Terra vegetation indices (VI) product(MOD13Q1). The VI product contains two indices, the NDVIand enhanced vegetation index (EVI) as well as VI quality

information. The product is derived from daily MODIS red,near infrared, and blue surface reflectance data and is providedevery 16 days as a plotted product in the Integerized Sinusoidalprojection. The VI algorithm operates on a per-pixel basis andrequires multiple observations (days) to generate a compositeVI using a methodology with three separate components:maximum value composite (MVC), constraint-view angle-maximum value composite (CV-MVC) and bidirectionalreflectance distribution function composite (BRDF-C).

• MODIS/Terra Surface Reflectance Daily L2G Global 250 mSIN Plot (MOD09GQK). This product is a two-band productcomputed from theMODISLevel 1B land red and near-infraredbands. The product is an estimate of the surface spectralreflectance for each band, as it would be measured at groundlevel if there were no atmospheric scattering or absorption.

The MODIS products were re-projected into a UTMprojection in order to be compatible with IKONOS data.

4.5. Image data processing

4.5.1. Vegetation indexNDVImeas

min was obtained from in-situ reflectance measure-ments carried above a surface composed primary of needle drylitter using a LICOR LI-1800 high spectral resolution fieldspectroradiometer in the range 400–1100 nm with spectralresolution of 2 nm. The LI-1800 was attached to a telescopewith a field of view (FOV) of 15°, which was positioned abovethe surface at a height of about 80 cm. Measurements wererepeated three times at each of five points randomly distributedover the forest and the average value was used in the analysis.Upwelling radiance from a white reference panel, measuredtwice at each sampling point (first and a last measurement) wasused for calculating reflectance of each surface-target scan(Gitelson et al., 2002; Pereira et al., 2004). The sample pointswere selected from the IKONOS image as visually identifiablegaps within a canopy cover and located using a GPS.

NDVIIKONOSmax was computed from IKONOS Red and near-

Infrared reflectance of an area of 2 by 2 pixels (8×8m each pixel?)with 100% coverage. The area was selected during the fieldreconnaissance, and subsequently located on the image usingGPS.

4.5.2. Image classificationWe performed a multi-spectral supervised classification of an

IKONOS image using a supervised classification procedure ofERDAS Imagine software. As a result, each pixel of the imagewas ascribed to one of seven end members (agricultural fields,asphalt or non-asphalt roads, covered pixel or gaps, and deep orshallow water for a water reservoir at the North-Eastern part ofthe forest) according to their spectral characteristics. Thesecharacteristics and sequence endmember determinations werebased on visual analysis of surface type. Canopy cover fortraining plots was calculated as a ratio of forested (Nf) and thetotal number of pixels that covers the specific plot (n):

CC ¼ Nf d100n

ð12Þ

Fig. 4. A comparison between TRAC and remote sensing measured LAI before (a)and after (b) the correction for clumpiness. CV stands for coefficient of variation.

422 M. Sprintsin et al. / Remote Sensing of Environment 109 (2007) 416–428

where n is a sum of all (forested and non-forested (Nnf)) pixelsin a sample plot (n=Nf+Nnf).

5. Results and discussion

5.1. LAI estimation for high spatial resolution (IKONOS)imagery

fc,IKONOS was calculated according to Eq. (10) usingvalues of 0.13±0.04 and 0.75±0.2 (±standard deviation) forNDVImeas

min and NDVIIKONOSmax respectively, determined by the

procedures described above. This resulted in a map of thespatial distribution of fc at a 4 m resolution. fc ranged fromapproximately 40 to 70% over the forested area within theentire image and averaged 49±7% for the ten training plots.These values were compatible with the average canopy covercalculated from measured canopy crown diameters of 48±9.Eq. (5) was solved for LAI assuming a random angulardistribution of leaf area. The resulting 4 m resolution LAIimage was then overlapped with the 10 ground-measuredplots' vector layer and the mean LAI value for each plot wasextracted using a Zonal Statistics procedure from the ERDASImagine software. The predicted variable, LAIIKONOS washighly correlated (R2 =0.79) to the ground truth LAI obtainedusing the TRAC. However, LAI was significantly (pb0.01)underestimated by remote sensing, the differences beingnoticeable at high LAI values (Fig. 4a). This presumablyresults from canopy clumping, which is, as mentionedpreviously, a major cause of LAI underestimation by non-contact methods.

The clumping index was determined from the distribution ofgap sizes measured by TRAC, which gave an average Ω valueof 0.84±0.05. This value can be interpreted (a) as the large-scale clumpiness related to tree distribution in the forest; and (b)as a compensation for the error that results from the fact that thetrue extinction coefficient is not known. It would be useful togain a better understanding of the relationship betweenclumpiness and density since stand density determination byremote sensing is likely to be easier than direct determination ofthe leaf angle distribution and the angle with which lighttransverses the canopy (required for calculating the extinctioncoefficient). Fig. 4b shows the remotely sensed LAI correctedfor clumping, and features a similar level of correlation as theprevious case (R2 =0.79) and a slope that is not significantlydifferent from 1 (Student's t-test for dependent variables). Thus,for high spatial resolution remote sensing (IKONOS), LAIcomputed from surface reflectance with Eq. (5) was signifi-cantly underestimated. However the introduction of the ground-based (TRAC) measured clumping factor resulted in values notsignificantly different from the ground truth (TRAC) measure-ments of LAI.

5.2. Validity of the MODIS LAI product

5.2.1. MODIS LAI data quality for our siteSince the empirical relationships between NDVI and LAI

were applied in the current study (described in details in

“Theory” section) it was important to be sure that such arelationship implemented by MODIS group backup algorithm(based on an empirical NDVI-LAI relationship) wasn't utilizedduring the LAI product release. This was done through theappraisal of the MODIS LAI quality control layer. Such a layer,computed for 1 km2 resolution, includes a status flag (SFC_QC)which can take one of five possible values: “0” and “1” refer toapplication of the main algorithm (RT) where “0” is the bestpossible result and “1” indicates that the RT method was usedwith saturation; “2” and “3” refer to high uncertainties in inputreflectance data or biome misclassifications (“2” for failure dueto geometrical problems and “3” for failure due to otherproblems) in which case the backup algorithm was used; and“4” indicates that both algorithms failed and therefore no data issupplied (Wang et al., 2004, 2005a,b). Analysis of the dataquality statistics for the study site shows that more than 24% ofthe LAI data was marked as high quality (24 pixels), more than66% was marked as sufficient results achieved by the RTalgorithm (65 pixels) and only 9% was misclassified (9 pixels).Consequently, it was concluded that the MODIS group backupalgorithm wasn't used to produce the LAI product, thus and theassumption of independence between LAI and NDVI can holdover the research site and period.

Table 2Comparison of MODIS-derived LAI and values acquired from several LAI meters

MODIS LP-80 TRAC SunLink LAI-2000

LAI (m2/m2) 1.7 1.69 1.33 1.56 1.49Absolute differences with MODIS LAI pixel value (m2/m2) 0.01 0.37 0.14 0.21Relative differences with MODIS LAI pixel value (%) 0.59 21.76 8.24 12.35Average 1.52Standard deviation 0.15

SunLink, SunScan and LP-80- are battery-operated linear sensors which measure PAR (Photosynthetically Active Radiation in the 400 to 700 nm waveband)transmittance by the canopy; the LAI-2000 calculates LAI from radiation measurements made with a “fish-eye” optical sensor (148° field-of-view) that determinescanopy light interception at 5 angles; TRAC described in details in sampling design section.

423M. Sprintsin et al. / Remote Sensing of Environment 109 (2007) 416–428

5.2.2. Inferring the validity of MODIS LAI product (compar-ison to ground-based measurements)

Due to its unique location the Yatir forest is a focal point forextensive dry-land research. Since 2000 the Weizmann Instituteof Science has maintained a flux tower in the forest as part of theCarboEuroFlux program (http://www.carboeurope.org/). In theframework of this project, indirect measurements of LAI wereconducted every 6 months from March 2001 until March 2004in three plots of ∼1000 m2 each, located within the flux towerfootprint area (1 km2). In March 2004 LAI was simultaneouslymeasured by several devices (Table 2). These measured LAIvalues were averaged per device and over the plots, andcompared with an extracted pixel value to evaluate the MODISLAI product as an independent source of information that wasnecessary in order to apply Eq. (7) in the following steps (seeFig. 2 for details).

The MODIS LAI value was within the range of the valuesobtained by the different devices, and not significantly differentfrom the mean (∼+12%), which led us to conclude that in ourcase the MODIS LAI product can be used without anycorrection factor.

5.3. LAI estimation for moderate spatial resolution imagery

The determination of LAI for moderate resolution, whichinvolved the use of moderate and low resolution MODISproducts, is described in a flowchart (Fig. 2) and in thefollowing sections. Changing scale of surface reflectance andthen applying the normal inversion for computing LAI wouldresult in errors due to the exponential nature of the relationshipbetween LAI and reflectance (e.g. Lang & Xiang, 1986).Therefore we chose not to downscale resolution of the LAIvalues directly (in the left branch of the flow chart). However,the extinction coefficient is linearly related to LAI and thereforewhen scale of the extinction coefficient is changed the meanLAI estimate should not change. Therefore, the 1 km resolutionLAI and NDVI products were used to determine the extinctioncoefficient using Eq. (7) (Section 5.3.1). The extinctioncoefficient distribution map was then downscaled to 250 mresolution using ERDAS Imagine software (ERDAS, 1999) andwas then used together with fractional vegetation cover (FVC),determined from an empirical relationship relating it to the250 m reflectance product to determine LAI with Eq. (5) atmoderate (250 m) spatial resolution (Section 5.3.3).

5.3.1. Fractional vegetation coverHistograms of fractional vegetation cover assessed for 250-

m resolution show trends identical to those observed at highresolution (see above). A comparison between trends presentedfor the whole study area (Fig. 5) shows a small shift at lowresolution to higher FVC values. This pattern was also appliedto the ten sampling plots and average FVC was 0.51±0.14 for250-m as compared to 0.49±0.07 for 4-m resolution.

Measured and calculated FVC were compared by calculatingthe relative error (RE) per training plot, as well as the averagerelative value for all ten plots by the following equation:

RE ¼ jsimulation−observationjobservation

ð13Þ

Relative errors were calculated for high resolution ascompared to ground measurements, as well as for moderateresolution compared to high resolution (Table 3). The averagerelative error was small both for CC vs. 4-m FVC (16%±12%)and for 4-m FVC vs. 250-m FVC (23%±21%). This accuracywas considered to be sufficient for further calculation. However,it should be noted that when the pixel resolution is coarser thana plot's scale poor aggregate values result, and it is better toextract values for sub-pixel heterogeneity (Kustas et al., 2004).This statement is true for our set of data since the averagerelative error calculated for CC vs. 250-m FVC was found to bemuch higher (31%±35%) (not presented in Table 3).

5.3.2. Determination of the extinction coefficientFollowing the procedure depicted by the left side of Fig. 2,

Eq. (7) was solved for the extinction coefficient (k) using 1-kmMODIS NDVI and LAI data. As was stated previously, ourintent was to present a method to assess LAI at a scale that couldbe compared to other available remote sensing products adaptedto environmental monitoring (e.g. ASTER, LandSat etc.).Consequently the spatial resolution of the image producedwas downscaled by changing the pixel size from 1 km (toocoarse for a 3000 ha forested area) to 250 m (close to high-resolution sensors).

Results showed relatively low variability over the forest withan average k of 0.45±0.11 (±STD), which is close to k for arandom leaf angle distribution (0.5). The small (not significant)deviation from random may result from the clumpy nature of theforest and indicates that the use of a clumping correction may be

Fig. 5. Histograms of fractional vegetation cover for 4-m and 250-m spatialresolution.

424 M. Sprintsin et al. / Remote Sensing of Environment 109 (2007) 416–428

important, as noted above (Sections 4.2 and 5.1). A similaraverage value of k, 0.63±0.05, was determined in previousground-based measurement campaigns referred to in Table 2(unpublished results).

The uneven distribution of k over the forest may result fromdifferent reasons such as variation in the soil depth, inhomoge-neous age and species distribution in some of the stands, aswell asmanagement practices, which are dictated by specific objectivesof the Israeli Forest Service (e.g. greening the landscape andrecreation), and are different in different parts of the forest.However, since the distribution of k-values over the forest was nothomogenous, ranging from 0.01 to 0.66, (Fig. 6) and differedfrom the normal as indicated by the Shapiro–WilkW test (Shapiro& Wilk, 1965) withW=0.9 and p=0.003, it was decided to treateach pixel separately during the subsequent LAI assessment.

5.3.3. Leaf area indexAfter assessment of the FVC (Section 5.3.1) and extinction

coefficient (Section 5.3.2) from the MODIS reflectance and LAI

Table 3Fractional vegetation cover (FVC) measured in-situ (canopy cover, CC) in thesampling plots and FVC assessed for high (4-m IKONOS) and moderate (250-mMODIS) spatial resolution imagery

Plot ID CC FVC 4-m⁎100

FVC 250-m⁎100

RE for CC vs.FVC 4-m

RE for FVC 250-mvs. 4-m

1 44 44 43 0.01 02 37 47 79 0.27 0.683 52 54 55 0.03 0.034 51 38 56 0.25 0.455 41 49 60 0.19 0.226 67 41 32 0.38 0.217 52 46 32 0.11 0.38 48 52 48 0.08 0.079 39 43 51 0.1 0.1910 50 61 55 0.22 0.09MEAN 48 48 51 0.16 0.23STD 9 7 14 0.12 0.21

RE stands for the relative error.

data we were able to calculate LAI for 250-m spatial resolutionand to draw a map of its spatial distribution. This map, as well asthe previously drawn 4-m spatial resolution LAI map ispresented in Fig. 7. Similar patterns of LAI distribution areshown in both maps, i.e. higher values over the mature northernand western parts of the forest (2–2.5) and lower values towardsthe south and south-east (1–2). This trend is compatible with theage distribution over the forest that becomes younger fromnorth-west (average tree age 36–37 years) to south-east(average trees age 20–23 years).

Unforested areas (“Bare soil” in the legend) in the high-resolution image were almost completely undetectable atcoarser resolution and influence the patchiness of the pictureby mixing with the surrounding vegetation and appearing asspots of low LAI (0.5–1).

As reported in previous studies (e.g. Tian et al., 2002a,b;Wang et al., 2004), a comparison of ground-based measure-ments from small areas to the moderate to coarse resolutionproducts requires development of detailed and appropriatestrategies in order to achieve a reasonable method forcomparison. In our case, the linear correlation between thetraining plot LAI and that of matching pixels, which resulted ina high correlation coefficient at high resolution (Fig. 2), wasvery low and non-significant at moderate resolution (R2 =0.16).This low correlation presumably resulted from the order ofmagnitude difference between pixel (∼62,500 m2) and trainingplot (1000 m2) size. The variation of LAI at different resolutionsand, consequently, the validity of moderate spatial resolutiondata were investigated, as an alternative, through analysis of thefrequency distribution of LAI values and its deviation from aNormal (Gaussian) distribution by calculating skewness andkurtosis (symmetry and “peakedness” measures of the distri-bution relative to a Normal distribution; Kustas et al., 2004).

Fig. 8 shows histograms of LAI distribution over the forestand statistics for both spatial resolutions and indicates similartrends regardless of the spatial resolution. The mean LAI was11% higher for the higher than for the coarser resolution, i.e.2.53 for 4-m and 2.25 for 250-m, and the coefficient of variation(CV) was high in both sets but lower for the coarser resolution,as expected (Kustas et al., 2004) since as spatial resolution

Fig. 6. The distribution of the extinction coefficient over the Yatir forest at 250-mspatial resolution. The broken vertical line indicates the forest average value.

Fig. 7. Leaf area index distribution over the Yatir forest for high (4-m) and moderate (250-m) spatial resolutions.

425M. Sprintsin et al. / Remote Sensing of Environment 109 (2007) 416–428

Fig. 8. Histograms for LAI distribution at different spatial resolutions. (a) 4-mIKONOS; (b) 250-m MODIS.

426 M. Sprintsin et al. / Remote Sensing of Environment 109 (2007) 416–428

decreases one can expect the increase in homogeneity of thearea covered by a larger pixel.

The results appear to indicate that a sufficient amount of thesurface information would be retrievable by moderate spatialresolution (250-m per pixel) imagery. However, we assume thata further decrease in resolution is not desirable since coarserpixel sizes will combine the attributes of different surfaceelements within such a large area (e.g. 1000 m2) into a singlereflectance value. So a significant amount of the surfaceinformation will be lost, and thus the possibility of detailedsurface characterization (Fig. 7). This supported by the slightdifferences observed for the statistical properties of both sets andin particular by the decrease in mean and CV with a decrease inthe spatial resolution (probably due to the contribution ofdifferent surface elements such as agricultural fields, open areasand coniferous stands to one “big” pixel reflectance) as well asby histograms of LAI distribution presented at Fig. 8. The lattershows that the distribution of LAI over the forest did not changedramatically with spatial resolution, although the high-resolu-tion distribution is symmetrical (skewness=0.00) and slightlycloser to a Normal one (kurtosis=−1.06) as compared to the

latter, (kurtosis=−1.2) and right tailed (skewness=0.08) coarserresolution distribution.

Although, spatial resolutions coarser then 250-m per pixelweren't considered in our study, results are similar to pre-viously reported in the literature. These showed that a dra-matic change (and consequently a significant errors) in thedistribution of the studied phenomena occurs primarily at thatpixel size (e.g. Atkinson, 1997; Kim et al., 2006; Kustas &Norman, 2000; Kustas et al., 2004). Thus we suppose that theslight, almost negligible, divergency between the results pre-sented in Fig. 8 may be considered as the beginning of thetrend of recession from the normality with further decrease inspatial resolution as in combination with decrease in mean andCV indicates deletion of the differences between various sur-face elements.

6. Conclusions

Remote sensing of vegetation cover and its related variablesand products is needed for understanding the impacts of landuse and climate change in arid and semi-arid regions. In thecurrent study we present a method to determine the LAI of aforest located at the desert fringe by analyzing results from ahigh spatial resolution sensor and applying the approach tomoderate spatial but high temporal resolution imagery.

LAI values assessed remotely using high spatial resolutionimagery were highly correlated with ground-based measure-ments made with the TRAC; however for successful LAIassessment the remote sensing data should be corrected with theclumping index, which has been shown to be the main source ofLAI underestimation by indirect methods. For a betterdescription of canopy patterns and as a compensation for non-random distribution of canopy elements, application tomoderate spatial resolution (250-m) was performed throughthe calculation of the canopy extinction coefficient on a regionalscale (pixel by pixel) using operationally produced MODIS LAIand VI products as ancillary sources of input data.

Histograms of resulting LAI distributions and descriptivestatistics at the different spatial resolutions indicate that theoverall distribution of LAI does not change dramatically withthe decrease in resolution and remains close to being normallydistributed. However, we suppose that one should be awarethat the effect of spatial resolution on the precision of theresults, as well as the implementation of global datasets (withmoderate to low spatial resolution) to local studies (especiallyto heterogeneous semi-arid regions) should be considered care-fully and site specific.

Although the results of this study suggest that LAImonitoring with available moderate (250-m) spatial resolutionimagery dataset supplied by EOS at a high temporal frequencycould be assessed with promising accuracy even for small scaleapplications, further investigation over a more heterogeneousareas (primarily in terms of species variability and age classes'distribution) could be useful in supplementing the expensivehigh spatial resolution data on the one hand and in com-plementing the currently available MODIS product on theother hand.

427M. Sprintsin et al. / Remote Sensing of Environment 109 (2007) 416–428

References

Andersen, J., Dybkjaer, G., Jensen, K. H., Refsgaard, J. C., & Rasmussen, K.(2002). Use of remotely sensed precipitation and leaf area index in adistributed hydrological model. Journal of Hydrology, 264(1–4), 34−50.

Arora, K. V. (2002). The use of the aridity index to assess climate change effecton annual runoff. Journal of Hydrology, 265, 164−177.

Atkinson, P. M. (1997). Selecting the spatial resolution of airborne MSSimagery for small-scale agricultural mapping. International Journal ofRemote Sensing, 18(9), 1903−1917.

Bergen, K. M., & Dobson, M. C. (1999). Integration of remotely sensed radarimagery in modeling and mapping of forest biomass and net primaryproduction. Ecological Modelling, 122(3), 257−274.

Bonan, G. B. (1993). Importance of leaf area index and forest type whenestimating photosynthesis in boreal forests. Remote Sensing of Environment,43, 303−314.

Breda, N. J. J. (2003). Ground-based measurements of leaf area index: A reviewof methods, instruments and current controversies. Journal of ExperimentalBotany, 54(392), 2403−2417.

Brown, L., Chen, J. M., Leblanc, S. G., & Cihlar, J. (2000). A shortwave infraredmodification to the simple ratio for LAI retrieval in boreal forests: An imageand model analysis. Remote Sensing of Environment, 71(1), 16−25.

Budyko, M. I. (1974). Climate and life.Orlando, FL: Academic Press 508 pp.Carlson, T. N., & Arthur, S. T. (2000). The impact of land use–land cover

changes due to urbanization on surface microclimate and hydrology: Asatellite perspective. Global and Planetary Change, 25, 49−65.

Carlson, T. N., & Ripley, D. A. (1997). On the relation between NDVI,fractional vegetation cover, and leaf area index. Remote Sensing ofEnvironment, 62(3), 241−252.

Carpenter, G. A., Gopal, S., Macomber, S., Martens, S., & Woodcock, C. E.(1999). A neural network method for mixture estimation for vegetationmapping. Remote Sensing of Environment, 70(2), 138−152.

Ceccato, P., Flasse, S., Tarantola, S., Jacquemound, S., & Gregorie, J. M. (2001).Detecting vegetation leaf water content using reflectance in optical domain.Remote Sensing of Environment, 77, 22−33.

Che, N., & Price, J. C. (1992). Survey of radiometric calibration results andmethods for visible and near-infrared channels of NOAA-7,-9 and-11AVHRRs. Remote Sensing of Environment, 41, 19−27.

Chen, J. M., Black, T. A., & Adams, R. S. (1991). Evaluation of hemi-sphericalphotography for determining plant area index and geometry of a forest stand.Agricultural and Forest Meteorology, 56, 129−143.

Chen, J. M., & Cihlar, J. (1995). Plant canopy gap-size analysis theory forimproving optical measurements of leaf-area index. Applied Optics, 34(27),6211−6222.

Chen, J. M., & Cihlar, J. (1995b). Quantifying the effect of canopy architectureon optical measurements of leaf area index using two gap size analysismethods. IEEETransactions onGeoscience and Remote Sensing, 33, 777−787.

Chen, J. M., Rich, P. M., Gower, T. S., Norman, J. M., & Pulmmer, S. (1997).Leaf area index on boreal forests: Theory, techniques and measurements.Journal of Geophysics Research, 102, 429−444.

Choudhury, T. N., Idso, S. B., & Reginato, R. J. (1988). Analysis of a resistance-energy balance method for estimating daily evaporation from wheat plotsusing one-time-of-day infrared temperature observations. Remote Sensing ofEnvironment, 19, 253−268.

Clevers, J. G. P. W. (1997). A simplified approach for yield prediction of sugarbeet based on optical remote sensing data. Remote Sensing of Environment,61(2), 221−228.

Cohen, W. B., Maiersperger, T. K., Yang, Z., Gower, S. T., Turner, D. P., Ritts,W. D., et al. (2003). Comparisons of land cover and LAI estimates derivedfrom ETM± and MODIS for four sites in North America: A qualityassessment of 2000/2001 provisional MODIS products. Remote Sensing ofEnvironment, 88, 233−255.

Cohen, S., Mosoni, P., & Meron, M. (1995). Canopy clumpiness and radiationpenetration in a young hedgerow apple orchard. Agricultural and Forestmeteorology, 76(3–4), 185−200.

Cohen, S., Rao, R. S., & Cohen, Y. (1997). Canopy transmittance inversionusing a line quantum probe for a row crop. Agricultural and ForestMeteorology, 86, 225−234.

Colombo, R., Bellingeri, D., Fasolini, D., & Marino, C. M. (2003). Retrieval ofleaf area index in different vegetation types using high-resolution satellitedata. Remote Sensing of Environment, 86(1), 120−131.

Coops, N. C., Smith, M. L., Jacobsen, K. L., Martin, M., & Ollinger, S. (2004).Estimation of plant area index using three techniques in a mature nativeeucalypt canopy. Austral Ecology, 29, 332−341.

Cowan, I. R. (1968). The interception and absorption of radiation in plant stands.Journal of Applied Ecology, 5, 367−379.

Daniel, T. W., Helms, J. A., & Baker, F. S. (1979). Principles of silviculture,Second edition U.S.A.: McGraw-Hill.

Derimian, Y., Karnieli, A., Kaufman, Y. J., Andreae, M. O., Andreae, T. W.,Dubovik, O., et al. (2006). Dust and pollution aerosols over the Negevdesert, Israel: Properties, transport, and radiative effect. Journal ofGeophysical Research, 111, D05205. doi:10.1029/2005JD006549

ERDAS (1999). Erdas field guide, 5th edn. Atlanta, GA: Erdas, Inc.Eriksson, H., Elkundh, L., Hall, K., & Lindroth, A. (2005). Estimating LAI in

deciduous forest stands. Agricultural and Forest Meteorology, 129, 27−37.Fang, H., & Liang, S. (2005). A hybrid inversion method for mapping leaf area

index from MODIS data: Experiments and application to broadleaf andneed-leaf canopies. Remote Sensing of Environment, 94, 405−425.

Ganor, E. (1994). The frequency of Saharan dust episodes over Tel-Aviv, Israel.Atmospheric Environment, 28(17), 2867−2871.

Gitelson, A. A., Kaufmanb, Y. J., Stark, R., & Rundquist, D. (2002). Novelalgorithms for remote estimation of vegetation fraction. Remote Sensing ofEnvironment, 80, 76−87.

Gong, P., Pu, R., Biging, G. S., & Larrieu, M. R. (2003). Estimation of forest leafarea index using vegetation indices derived from Hyperion hyperspectraldata. IEEE Transactions on Geoscience and Remote Sensing, 41(6),1355−1362.

Gower, S. T., Kucharil, C. J., & Norman, J. M. (1999). Direct and indirectestimation of leaf area index, fapar, and net primary production of terrestrialecosystems. Remote Sensing of Environment, 70, 29−51.

Graetz, R. D., & Gentle, M. R. (1982). The relationship between reflectance in theLandsat wavebands and the composition of an Australian semiarid shrubrangeland.Photogrammetric Engineering and Remote Sensing, 48, 1721−1730.

Grunzweig, J. M., Lin, T., Schwartz, A., & Yakir, D. (2003). Carbonsequestration in arid-land forest. Global Change Biology, 9(5), 791−799.

Gutman, G., & Ignatov, A. (1998). The derivation of the green vegetationfraction from NOAA/AVHRR data for use in numerical weather predictionmodels. International Journal of Remote Sensing, 19(8), 1533−1543.

Ichoku, C., & Karnieli, A. (1996). A review of mixture modeling techniques forthe sub-pixel land cover estimation. Remote Sensing Reviews, 33, 161−186.

Jackson, R. D., Hatfield, J. L., Reginato, R. J., Idso, S. B., & Pinter, P. J. (1983).Estimation of daily evapotranspiration from one time-of-day measurements.Agricultural Water Management, 7, 351−362.

Jacquemound, S., Baret, F., & Hanocq, J. F. (1992). Modeling spectral andbidirectional soil reflectance. Remote Sensing of Environment, 41(2–3),123−132.

Jarvis, P. G., & Laverenz, J. W. (1983). Productivity of temperate, deciduous andevergreen forests. In O. L. Lange, C. B. Omond, & H. Zeigler (Eds.), Phy-siological plant ecology, Vol. IV. (pp. 133−144) New York: Springer Verlag.

Jones, H. G. (1992). Plant and microclimate, 2nd end. Cambrige, UK:Cambrige University Press.

Kim, J., Guo, Q., Baldocchi, D. D., Leclerc, M. Y., Xu, L., & Schmid, H. P.(2006). Upscaling fluxes from tower to landscape: Overlaying fluxfootprints on high-resolution (IKONOS) images of vegetation cover. Agri-cultural and Forest Meteorology, 136, 132−146.

Knyazikhin, Y., Glassy, J., Privette, J. L., Tian, Y., Lotsch, A., Zhang, Y., et al.(1999). MODIS Leaf Area Index (LAI) and Fraction of PhotosyntheticallyActive Radiation absorbed by vegetation (FPAR) product (MOD 15)algorithm, theoretical basis document. Greenbelt, MD 20771, USA: NASAGoddard Space Flight Center.

Kucharik, C. J., Norman, J. M., & Gower, T. S. (1999). Characterization ofradiation regimes in nonrandom forest canopies: Theory, measurements, anda simplified modeling approach. Tree Physiology, 19, 695−706.

Kucharik, C. J., Norman, J. M., Murdock, L. M., & Gower, T. S. (1997).Characterizing canopy non-randomness with a Multiband Vegetation ImagerMVI. Journal of Geophysical Research, 102, 455−473.

428 M. Sprintsin et al. / Remote Sensing of Environment 109 (2007) 416–428

Kustas, W. P., & Norman, J. M. (2000). Evaluating the effects of subpixelheterogeneity on pixel average fluxes. Remote Sensing of Environment, 74,327−342.

Kustas, W. P, Li, F., Jackson, T. G., Prueger, J. H., MacPherson, J. I., & Wolde,M. (2004). Effects of remote sensing pixel resolution on modeled energyflux variability of croplands in Iowa. Remote Sensing of Environment, 92,535−547.

Lang, A. R. G. (1986). Leaf area and average leaf angle from transmittance ofdirect sunlight. Australian Journal of Botany, 34, 349−355.

Lang, A. R. G. (1987). Simplified estimate of leaf area index from transmittanceof the sun's beam. Agricultural and Forest Meteorology, 41, 179−186.

Lang, A. R. G., & Xiang, Y. (1986). Estimation of leaf area index fromtransmission of direct sunlight in discontinuous canopies. Agricultural andForest 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 fromdigital hemispherical photography in boreal forests. Agricultural and ForestMeteorology, 129, 187−207.

Miller, J. B. (1967). A formula for average foliage density. Australian Journal ofBotany, 15, 141−144.

Leblanc, S. G., Chen, J. M., & Kwong, M. (2002). Tracing radiation andarchitecture of canopies TRAC MANUAL Version 2.1, natural resourcesCanada.Canada Center for Remote Sensing, pp. 25.

Monteith, J. L. (1969). Light interception and radiative exchange in crop stands.In J. D. Eastin, F. A. Haskins, C. Y. Sullivan, & C. H. M. van Bavel (Eds.),Physiological aspects of crop yield (pp. 89−111). Madison, Wisconsin:American Society of Agronomy, Crop Science Society of America.

Morisette, J. T., Nickesonb, J. E., Davis, P., Wang, Y., Tian, Y., Woodcock, C. E.,et al. (2003). High spatial resolution satellite observations for validation ofMODIS land products: IKONOS observations acquired under the NASAScientific Data Purchase. Remote Sensing of Environment, 88, 100−110.

Nemani, R., & Running, S. W. (1996). Global vegetation cover changes from coarseresolution satellite data. Journal of Geophysical Research, 101, 7157−7162.

Nilson, T. A. (1971). A theoretical analysis of the frequency of gaps in plantstands. Agricultural Meteorology, 8, 25−38.

Norman, J. M., & Campbell, G. S. (1989). Crop canopy photosynthesis andconductance from leaf measurements. Workshop prepared for LI-COR, Inc.,Lincoln, NE.

Norman, J. M., Kustas, W. P., & Humes, K. S. (1995). Source approach forestimating soil and vegetation energy flaxes in observations of directionalradiometric surface temperature. Agricultural and Forest Meteorology, 77,263−293.

Pax-Lenney, M., Woodcock, C. E., Macomber, S. A., Gopal, S., & Song, C.(2001). Forest mapping with a generalized classifier and Landsat TM data.Remote Sensing of Environment, 77(3), 241−250.

Pereira, J. M. C., Mota, B., Privette, J. L., Caylor, K. K., Silva, J. M. N., Sá,A. C. L., et al. (2004). A simulation analysis of the detectability ofunderstory burns in miombo woodlands. Remote Sensing of Environment,93, 296−310.

Ponce, V. M., Pandey, R. P., & Ercan, S. (2000). Characterization of droughtacross the climate spectrum. Journal of Hydrological Engineering, ASCE, 5(2), 222−224.

Price, J. C. (1987). Calibration of satellite radiometers and the comparison ofvegetation indices. Remote Sensing of Environment, 21, 15−27.

Privette, J. L., Myneni, R. B., Knyazikhin, Y., Mukelabai, M., Roberts, G., Tian,Y., et al. (2002). Early spatial and temporal validation of MODIS LAIproduct in the Southern Africa Kalahari. Remote Sensing of Environment,83, 232−243.

Richardson, A. J., &Wiegand, C. L. (1977). Distinguishing vegetation from soilbackground information. Photogrammetric Engineering and RemoteSensing, 43, 1541−1552.

Roberts, D. A., Gardner, M., & Green, R. O. (1998). Mapping chaparral in theSanta Monica Mountains using multiple endmember spectral mixturemodels. Remote Sensing of Environment, 65, 267−279.

Rouse, J. W., Haas, R. H., Schell, J. A., Deering, D. W., & Harlan, J. C. (1974).Monitoring the vernal advancement of retrogradation of natural vegetation,NASA/GCFC, Type III. Final report, Greenbelt, MD, USA (pp. 1−371)..

Santoro, M., Askne, J., Smith, G., & Fransson, J. E. S. (2002). Stem volumeretrieval in boreal forests from ERS-1/2 interferometry. Remote Sensing ofEnvironment, 81(1), 19−35

Schiller, G., & Cohen, Y. (1998). Water balance of Pinus halepensis Mill.afforestation in arid region. Forest Ecology and Measurement, 105, 121−128.

Shapiro, S. S., & Wilk, M. B. (1965). An analysis of variance test for normality(complete samples). Biometrika, 52, 591−611.

Sims, D. A., & Gamon, J. A. (2003). Estimation of vegetation water content andphotosynthetic tissue area from spectral reflectance: A comparison of indicesbased on liquid water and chlorophyll absorption features. Remote Sensingof Environment, 84, 526−537.

Stoms, D. (1994). Scale dependence of species richness maps. ProfessionalGeographer, 46, 346−358.

Strahler, A. H., Woodcock, C. E., & Smith, J. A. (1986). On the nature of modelsin remote-sensing. Remote Sensing of Environment, 20, 121−139.

Tian, Y., Wang, Y., Zhang, Y., Knyazikhin, Y., Bogaert, J., & Myneni, R. B.(2002a). Radiative transfer based scaling of LAI retrievals from reflectancedata of different resolutions. Remote Sensing of Environment, 84, 143−159.

Tian, Y.,Woodcock,C. E.,Wang, Y., Privette, J. L., Shabanov,N.V., Zhou, L., et al.(2002b). Multiscale analysis and validation of the MODIS LAI product overMaun, Botswana: II. Sampling strategy. Remote Sensing of Environment, 83,431−441.

Tucker, C. J. (1979). Red and photographic infrared linear combinations formonitoring vegetation. Remote Sensing of Environment, 8, 127−150.

Vermote, E., Tanré, D., Deuzé, J. L., Herman,M., &Morcrette, J. J. (1997). Secondsimulation of the satellite signal in the solar spectrum, 6S: An overview. IEEETransactions on Geoscience and Remote Sensing, 35(3), 675−686.

Wang, Q. T., Adiku, S., Tenhunen, J., & Granier, A. (2005a). On the relationshipof NDVI with leaf area index in a deciduous forest site. Remote Sensing ofEnvironment, 85, 244−255.

Wang, Q. T., Tenhunen, J., Dinh, N. Q., Reichstein, M., Otieno, D., Granier, A.,et al. (2005b). Evaluation of seasonal variation of MODIS derived leaf areaindex at two European deciduous broadleaf forest sites. Remote Sensing ofEnvironment, 96, 475−484.

Wang, Y., Woodcock, C. E., Buermann, W., Stenberg, P., Voipio, P., Smolander,H., et al. (2004). Evaluation of the MODIS LAI algorithm at a coniferousforest site in Finland. Remote Sensing of Environment, 91, 114−127.

Ward, R. C., & Robinson, M. (2000). Principles of hydrology, 4th edition:McGraw Hill 450 pp.

Watson, D. J. (1947). Comparative physiological studies in the growth of fieldcrops, I. Variation in net assimilation rate and leaf area between species andvarieties, and within and between years. Annals of Botany, 11, 41−76.

Welles, J. M., & Cohen, S. (1996). Canopy structure measurement by gapfraction analysis using commercial instrumentation. Journal of Experimen-tal Botany, 47(302), 1335−1342.

Wittich, K. P., & Hansing, O. (1995). Area-averaged vegetative cover fractionestimated from satellite data. International Journal of Biometeorology, 38,209−215.

Wu, C., & Murray, A. T. (2003). Estimating impervious surface distribution byspectral mixture analysis. Remote Sensing of Environment, 84, 493−505.

Xiao, J., & Moody, A. (2005). A comparison of methods for estimatingfractional green vegetation cover within a desert-to-upland transition zone incentral New Mexico, USA. Remote Sensing of Environment, 98, 237−250.

Zeide, B. (1995). A relationship between size of trees and their numbers. ForestEcology and Management, 72, 265−272.