Upload
matheusboni
View
221
Download
2
Tags:
Embed Size (px)
DESCRIPTION
Technical article about the use of a ref-edge spectral index for remote sensing of vegetation
Citation preview
Europ. J. Agronomy 46 (2013) 42 52
Contents lists available at SciVerse ScienceDirect
European Journal of Agronomy
jo u rn al hom epage: www.elsev ier .c
A red-e stiagroec
J. Delegid orDepartment of ncia, C
a r t i c l
Article history:Received 4 ApReceived in re30 November Accepted 9 De
Keywords:LAINDIRed-edgeCrops growth monitoringRemote sensingSentinel-2
cal pa red all knot, in tty of
campaigns SPARC 2003 and 2004 (Barrax, Spain) experimental LAI values over 9 different crop types havebeen collected while at the same time spaceborne imagery have been acquired using the hyperspectralCHRIS (Compact High Resolution Imaging Spectrometer) sensor onboard PROBA (Project for On-BoardAutonomy) satellite. This extensive dataset allowed us to evaluate the optimal band combination throughspectral indices based on normalized differences. The best linear correlation against the experimental LAIdataset was obtained by combining the 674 nm and 712 nm wavebands. These wavelengths correspond
1. Introdu
Leaf areaand modelling and for(Dorigo et one-sided adirectly rel2001). The has becomstudies (Doobservationfor dedicatet al., 2011the spatial the farmer
CorresponE-mail add
1161-0301/$ http://dx.doi.oto the maximal chlorophyll absorption and the red-edge position region, respectively, and are known tobe sensitive to the physiological status of the plant. Contrary to the NDVI (r2: 0.68), the red-edge NDIcorrelated strongly (r2: 0.82) with LAI without saturating at larger values. The index has been subse-quently validated against eld data from the 2009 SEN3EXP campaign (Barrax, Spain) that again spanneda wide variety of crop types. A linear relationship over the full LAI range was conrmed and the regres-sion equation was applied to a CHRIS/PROBA image acquired during the same campaign. A LAI map hasbeen derived with an RMSE accuracy of 0.6. It is concluded that the red-edge spectral index is a powerfulalternative for LAI estimation and may provide valuable information for precision agriculture, e.g. whenapplied to high spatial resolution imagery.
2012 Elsevier B.V. All rights reserved.
ction
index (LAI) is a key variable used by crop physiologistsers for estimating foliage cover, as well as monitor-ecasting crop growth, biomass production and yieldal., 2007; Casa et al., 2012). Green LAI is dened asrea of green leaves per unit ground area and is thusated to the growth status of the crop (Scurlock et al.,spatially explicit quantication of LAI over large arease an important aspect in agroecological and climaticrigo et al., 2007). At the same time, remotely senseds are increasingly being applied at a within-eld scaleed agronomical monitoring applications (Gianquinto; Sakamoto et al., 2012). For instance, knowledge ofdistribution of LAI and chlorophyll content can assisttowards a more precise distribution of fertilizers (e.g.
ding author. Tel.: +34 963544068; fax: +34 963543261.ress: [email protected] (J. Delegido).
nitrogen dressings) on the eld (Houles et al., 2007; Nguyen andLee, 2006). Because LAI is functionally linked to the canopy spec-tral reectance, its retrieval from optical remote sensing data hasprompted many studies using various techniques (Aparicio et al.,2000; Baret and Guyot, 1991; Haboudane et al., 2004). Essentially,these retrieval techniques can be classied into two groups (LeMaire et al., 2008; Zheng and Moskal, 2009): (i) empirical retrievalmethods, which typically consist of relating the biophysical param-eter of interest against spectral data through linear (e.g. vegetationindices) or nonlinear (e.g. machine learning approaches) algorith-mic techniques (Broge and Mortensen, 2002; Glenn et al., 2008;Myneni et al., 1995; Verrelst et al., 2012) and (ii) physically-basedretrieval methods, which refers to inversion of radiative transfermodels (RTMs) against remote sensing observations (e.g. Gobronet al., 2000; Goel, 1987; Houborg and Boegh, 2008; Jacquemoudet al., 1995). Both approaches have their strengths and weak-nesses, which led to the development of many hybrid forms. Forinstance, machine learning methods (e.g. neural networks) are typ-ically trained by synthetic spectra from RTMs (Hastie et al., 2009;Verger et al., 2008).
see front matter 2012 Elsevier B.V. All rights reserved.rg/10.1016/j.eja.2012.12.001dge spectral index for remote sensing eosystems
o , J. Verrelst, C.M. Meza, J.P. Rivera, L. Alonso, J. M Earth Physics and Thermodynamics, Image Processing Laboratory, Universidad de Vale
e i n f o
ril 2012vised form2012cember 2012
a b s t r a c t
Leaf area index (LAI) is a key biophysitwo-band vegetation indices based onence vegetation index (NDVI) are we(35). To bypass this saturation effecmation of green LAI over a wide varieom/ locate /e ja
mation of green LAI over
eno/Catedrtico Agustn Escardino 9, 46980 Paterna, Valencia, Spain
rameter for the monitoring of agroecosystems. Conventionalnd near-infrared relationships such as the normalized differ-wn to suffer from saturation at moderate-to-high LAI valueshis work a robust alternative has been proposed for the esti-crop types. By using data from European Space Agency (ESA)
J. Delegido et al. / Europ. J. Agronomy 46 (2013) 42 52 43
The advantage of vegetation indices is that they allow obtain-ing relevant information in a fast and easy way and the underlyingmechanisms are well-understood. Most widely known is the Nor-malized Difference Vegetation Index (NDVI) (Rouse et al., 1973).This succesexpresses tthe red chlothe NIR dueleaves mesovegetation,content (Baless, the reinstance NDtions of modHaboudane
At the sarapid techntroscopy orsolar radiabetween 50region (Ngous imaginfor the puret al., 2011Mauser, 200eral experimregion relaet al., 2004where a sha750 nm takphyll absordemonstratinuenced Lee et al., 20this region chlorophylllonger wav1994; Herrm
The promsors for a wthe design meters. UntHYPERION,erties at higof space mmonitoringHyspIRI mi(more thanspectral andEarth obsernarrowbandoperated byagroecosysthas been co740 nm. Thand aims toresolution oa high revis(ESA, 2010)edge indicegeneric settwhether nemations thaimage-wide
MeanwhResolution
for On-Board Autonomy) satellite (Barnsley et al., 2004) can serveas benchmark for the evaluation of new and existing hyperspectralspectral indices on their use for space-based vegetation moni-toring application. CHRIS/PROBA was designed as a technology
stratch in
o funlst et
CHRes im
acqelevantert et ondu
or ugetaidmbailn tesing iricaings wo-bwideuateappligain
thod
ariethe obon rel., 20formred-e
mahichcaus
veg1999uratWeis
typoadleet al.sion
at rel o002;-LAI v
Thisium
erg, 2h theconvof na
satuch a
indor thropoI. NDaramsful index from the early days of remote sensinghe normalized ratio between the reected energy inrophyll absorption region and the reected energy in
to scattering of light in the intercellular volume of thephyll, and provides an indicator of the greenness of the
which is in a way related to green LAI and chlorophyllret and Guyot, 1991; Myneni et al., 1995). Neverthe-lationship between NDVI and LAI is exponential, forVI approaches saturation asymptotically under condi-erate-to-high LAI values (e.g. >35) (Dorigo et al., 2007;
et al., 2004).me time, during the last two decades there has beenological progress in the development of imaging spec-
hyperspectral sensors that capture images of reectedtion in a large number of narrow bands (typically
and 250 bands) across the visible and near-infrareduyen and Lee, 2006; Schaepman et al., 2009). Vari-g spectrometers have been mounted onboard aircraftspose of precision farming applications (e.g. Delegidoa; Lee et al., 2004; Meggio et al., 2010; Oppelt and4). By analysing such imaging spectrometer data, sev-ents have demonstrated that an important spectral
ted to LAI is to be found in the red-edge region (Lee; Liu et al., 2004; Wu et al., 2010). This is the regionrp change in reectance between wavelengths 690 andes place, and characterizes the transition from chloro-ption to leaf scattering (Clevers et al., 2002). It has beened that the shape of the red-edge region is stronglyby LAI (Delegido et al., 2008; Herrmann et al., 2011;04) principally by the slope of the reectance curve in(Filella and Penuelas, 1994), while an increase in leaf
content causes a shift in the red-edge position towardselengths (Dash and Curran, 2004; Filella and Penuelas,ann et al., 2011; Moran et al., 2004).ise and potential of hyperspectral narrowband sen-
ide array of Earth resource applications has motivatedand also the launch of spaceborne imaging spectro-il now only experimental imaging spectrometers (e.g.
HICO, CHRIS) that detect vegetation biophysical prop-h spatial resolution from space exist, but these kinds
issions are being planned in near future for routinely land surfaces (e.g. the Germans Enmap mission, NASAsssion). Nevertheless, superspectral resolution sensors
10 and less than 50 bands, i.e. in-between multi- hyperspectral resolution) onboard of new generationvation spacecrafts have already incorporated red-edges. For instance, the forthcoming Sentinel-2 satellite
the European Space Agency (ESA), among others forems monitoring applications (Malenovsky et al., 2012),ngured with new narrowbands, centred at 705 nm ande rst Sentinel-2, is envisaged to be launched in 2013
deliver data taken over all land surfaces at a spatialf 10 m, 20 or 60 m (depending on the used bands) atiting time (each 5th day under cloud-free conditions). Despite the good performances of narrowband red-s in local eld experiments, its robustness in a moreing, is still an open issue. It remains to be investigatedw red-edge narrowbands can deliver more robust esti-n conventional indices such as NDVI when applied over
heterogeneous agroecosystems.ile, experimental missions such as ESAs Compact HighImaging Spectrometer (CHRIS) onboard PROBA (Project
demonits launtinue t(Verreis thatcapturimagesmost rter of iVerrelsbeen c2008),cic veand SkThenkayet beetions uan empThis brerful tover a to evalwhen index a
2. Me
A vwith tbased He et aNDVI, of the canopysors, wand beappliedet al., ing acc2003; etationand brFriedl conclumerelythe levet al., 2at mid1999).to-medBlomb
Witto the range bypasstors suwidelyoped fwere pand LAboth por and initially intended as a one year mission since 2001. But both the satellite and the CHRIS sensor con-ction well until now, making this sensor very successful
al., 2010). A constraint for operational use, however,IS does not deliver operational data streams but onlyages over requested sites. Nevertheless, by using suchuired over agricultural areas, it is possible to infer thent bands combination that are related to the parame-est (Darvishzadeh et al., 2008; Thenkabail et al., 2000;al., 2012). While these kinds of exercises have alreadycted in a theoretical setting using RTMs (Le Maire et al.,sing ground or airborne hyperspectral data for a spe-tion type such as pasture (Fava et al., 2009; Mutangaore, 2004) or a specic crop type (Casa et al., 2012;
et al., 2000), the evaluation of optimized indices has notsted over a multitude of crop types and growth condi-spaceborne data, which is essential when aiming to uselly-optimized index over large datasets of space images.us to the following objectives: (i) to infer the most pow-and spectral index from CHRIS data in estimating LAI
range of agricultural crops and growth conditions, (ii) this spectral index on its robustness for LAI estimatinged to an independent dataset, and (iii) to compare thisst other established vegetation indices sensitive to LAI.
s
y of spectral vegetation indices have been developedjective of passive estimating biophysical parametersmotely sensed spectral radiances (Bannari et al., 2007;06). One of the oldest and most widely used indices is theed from the normalized reectance values either sidedge, which discriminate between live green and otherterial. Because of having its origin in broadband sen-
forms still the majority of the Earth observing satellites,e of its simplicity, NDVI is one of the most extensivelyetation indices related to LAI (Glenn et al., 2008; Turner). Empirical approaches are predominant in deliver-e estimations at local to landscape scale (Cohen et al.,steiner and Khbauch, 2005). Studies on various veg-es, e.g. agroecosystems, grass and shrublands, coniferaf forests (Chen and Cihlar, 1996; Fassnacht et al., 1994;, 1994; Law and Waring, 1994) have led to the generalthat the NDVI has considerable sensitivities to LAI, butelatively low LAI values (Turner et al., 1999). Althoughf saturation is variable and species-dependent (Chen
Hoffmann and Blomberg, 2004), it is generally reachedalues around 35 (Thenkabail et al., 2000; Turner et al.,
means that NDVI may be a good predictor for only low- LAIs (Gonzlez-Sanpedro et al., 2008; Hoffmann and004; Yao et al., 2008).
advent of hyperspectral imagery, various alternativesentional NDVI have been proposed. For instance, a widerrowband vegetation indices have been developed toration effects or minimizing effects of confounding fac-s soil background. Table 1 shows some of the mostices used. Most of these indices were initially devel-e study of chlorophyll, except SR and OSAVI, whichsed to study LAI. TVI was proposed for both chlorophyllVI and others have subsequently been used to studyeters (Haboudane et al., 2008). In an attempt to further
44 J. Delegido et al. / Europ. J. Agronomy 46 (2013) 42 52
Table 1Vegetation indices used in this study, where R is reectance at wavelength (nm).
Index Formula Reference
NDVI (R800 R670)/(R800 + R670) Rouse et al. (1973)MCARITCARIMTCI TCIR-M TVI OSAVI PRISR SR705
optimize ththese indicTCARI/OSAVet al., 2010that the mto construclating all poto the NDV(NDIab):
NDIab =R
R
where Ra,bvisible andwas demona few select(Fava et al.,cally, some around 670informationsuccessful ((2007) founLAI and cotusing CHRIthat a stronis betweenOverall, whshowed thaon several plant matergeneral. In all two-banto optimizespectral datdeterminatinstead of awere undera large variregression ecompared aTable 1.
3. Experim
3.1. SPARC
The expmized LAI-sBarrax Caming the sumat Barrax, L
altitu 1nifo
y lan are tl raincted b
16 Jphys
in toer,
et ofmpleare
ESU anality ofanops anwas gs % (Feith threa, d me
paral we
16 atiaecuti6, HRIS
to 1ing
tion lar,
en 6 ir (Basor [(R700 R670) 0.2(R700 R550)] R700/R6703[(R700 R670) 0.2(R700 R550)R700/R670] [(R750 R710)/(R710 R680)] 1.2(R700 R550) 1.5(R670 R550)(R700/R670)1/2R750/R720 1 0.5[120(R750 R550) 200(R670 R550)] (R800 R670)/(R800 + R670 + 0.16) (R550 R531)/(R550 + R531)R800/R670R750/R705
e sensitivity of these indices, some authors started usinges together as a new index such as MCARI/OSAVI orI (Daughtry et al., 2000; Haboudane et al., 2008; Meggio
).A drawback of these established indices, however, isost sensitive bands are not necessarily the ones usedt the index. An alternative approach therefore is calcu-ssible two-band narrowband combinations accordingI formulation, being the Normalized Difference Index
b Rab + Ra
(1)
are the reectance values in the a and b bands in the near-infrared spectral range. Using this approach, itstrated that the best information is contained in onlyed bands or indices with the rest becoming redundant
2009; Ray et al., 2006; Thenkabail et al., 2000). Speci-authors have demonstrated that the band combination
and 800 nm, as used by NDVI, not always provides best about LAI, while other regions appeared to be moreThenkabail et al., 2000; Zhao et al., 2007). Zhao et al.d that the bands with best linear correlation betweenton eld data were 700710 and 750900. Similarly,S data over a shrubland, Stagakis et al. (2010) foundg linear correlation between NDIab and LAI when b
580 and 720 nm, and a between 710 and 1003 nm.ile showing superiority over NDVI, these studies alsot the accuracy of different optimized indices dependsfactors related to the biological characteristics of theial, and no single index could be considered superior inthis work the NDIab formulation was used to evaluated combinations in the range of 6001000 nm that leadd linear correlation with LAI using spaceborne hyper-a. Evaluation was done by calculating the coefcient ofion (r2). It should thereby being taken into account thatiming at high accuracies for a specic crop type, effortstaken to seek for an optimized index applicable overety of crops agroecosystem. Finally, the best-evaluatedquation was validated by an independent dataset andgainst the performance of the established indices of
700 m of 5 kmlarge, u65% drditionsannuacondu15 and
Bioparcelssunowlarge stary safrom bIn eachdigitalintensof the c(Wellestudy readinand 10area wsame ation an2004).
In images15 andhigh sp5 cons0, 3pass. C400 nmpromisresoluparticubetweat nadthe senental dataset
erimental data used for the development of an opti-ensitive index was obtained from the SPARC (Spectrapaigns) campaigns which were organized by ESA dur-mers of 2003 and 2004. The campaigns were conducteda Mancha region in Spain (coordinates 303N, 26W;
which is mrst geomeatmospheriGuanter etof calibratiotent and aeatmospheriparison of Cground-basis describedacquired dDaughtry et al. (2000)Haboudane et al. (2002)Dash and Curran (2004)Haboudane et al. (2008)Gitelson et al. (2005)Broge and Leblanc (2000)Rondeaux et al. (1996)Gamon et al. (1992)Jordan (1969)Gitelson and Merzlyak (1994)
de). The test area has a rectangular form and an extent0 km, and is characterized by a at morphology andrm land-use units. The region consists of approximatelyd and 35% irrigated agricultural parcels. The climate con-ypically Mediterranean with a hot and dry summer. Thefall average is about 400 mm. The 2003 campaign wasetween 12 and 14 July, and the 2004 campaign between
uly.ical parameters were measured on various agriculturaltal spanning 9 different crop types (garlic, alfalfa, onion,corn, potato, sugar beet, vineyard and wheat) and a
ground sampling points were identied (240 elemen- units (ESU) plots from crops and additional 60 samplessoils). ESU refers to a plot size of about 20 m 20 m., among other parameters, LAI was measured with ayzer (Licor LAI-2000), which works by comparing the
diffuse incident illumination measured at the bottomy with that arriving at the top (LI-COR technical report)
d Norman, 1991). Each LAI value used in the presentobtained as a statistical mean of 24 measures (8 data3 replications) with variable standard errors between 5rnndez et al., 2005). Fig. 1 shows an image of the studye 2004 campaign crops. The 2003 campaign was on thethough there were also some changes in crop cultiva-asured crop types (Delegido et al., 2008; Moreno et al.,
lel with the eld measurements, four CHRIS/PROBAre acquired during the days 12 to 14 July 2003 andJuly 2004. CHRIS on board PROBA satellite providesl resolution hyperspectral/multiangular data, acquiringve images from 5 different views (y-by zenith angles55) over a dedicated site in one single satellite over-
measures over the visible/near-infrared spectra from050 nm. It can operate in different modes, thereby com-between the number of spectral bands and the spatialto keep balanced the signal level and data volume. InCHRIS Mode 1 provides 62 bands (bandwidth rangesnm and 12 nm) and has a spatial resolution of 34 mrnsley et al., 2004). In the 2003 and 2004 campaigns,was congured at highest spectral resolution Mode 1,
ost favourable for vegetation studies. The images weretrically corrected (Alonso and Moreno, 2005) and thencally corrected according to the method proposed by
al. (2005). This method simultaneously derives a setn coefcients and an estimation of water vapour con-rosol optical thickness from the data themselves. Thec correction of the data was validated by direct com-HRIS-derived reectance retrievals with simultaneoused measurements acquired during the campaigns, as
in Guanter et al. (2005). From all the angular imagesuring the campaigns, only the ones corresponding to
J. Delegido et al. / Europ. J. Agronomy 46 (2013) 42 52 45
Fig. 1. Land umeasured. The
nadir view are minimiz
3.2. SEN3EX
In view odataset wament) campand formedEnvironmenThe SEN3EXit included neously acqand spaceboincluding Bwith differecampaign wSPARC, elemand other b
In this cover 14 difthe methodculating thcan be relawere condusugar beet,se map, in southeast of Iberian Peninsula, for selected crops during the SPARC 2004 ca grid reects the UTM-projection coordinates (Delegido et al., 2008).
were selected so that angular and atmospheric effectsed, and that highest spatial resolution is preserved.
P
f validating the best performing NDIab, an independents used coming from the SEN3EXP (Sentinel-3 Experi-aign. The SEN3EXP campaign was conducted in 2009
part of the European GMES (Global Monitoring fort and Security) Sentinel-3 programme (ESA, 2012).P campaign was set up in a similar way as SPARC;
collection of eld measurements, along with simulta-uired hyperspectral observations from various airbornerne sensors. Several sites across Europe where selectedarrax as representing a Mediterranean agroecosystemnt water regimes (rainfed and irrigated). The Barraxas carried out during 2024 June 2009. Similar as inentary sample units (ESUs) were dened wherein LAI
iophysical parameters were collected.ampaign, LAI was measured in 34 ESUs distributedferent agricultural elds (Delegido et al., 2011b) byology of hemispheric photographs, which allows cal-e gap fraction over the angular range of 180 andted to LAI (Weiss et al., 2004). LAI measurementscted over 9 different crop types, being sunowers,
almond trees, alfalfa, garlic, corn, vineyard, onion
and potatoity of the ESUs are mFig. 2.
Along wMode 1 imthis work close to themetrically same methcampaign.
4. Results
4.1. NDVI
The SPAimages, wable for remdifferent agcalculated quently plo(Fig. 3). Altto an r2 of 0already aroin Fig. 3, ledmpaign. The points marked with (+) indicate points where LAI was
, with LAI varying between 0 and 3 for the major-crops, and with an LAI around 7 for potatoes. Thearked by the labels on the agricultural parcels in
ith the eld measurements, several spaceborne CHRISages were acquired the 19 and 29 June 2009. Inwe used the 19 June CHRIS image, which is most
period of eld measurements. The images were geo-and atmospherically preprocessed according to theodology as described in the above-mentioned SPARC
RC eld dataset, along with the ensemble of CHRISs used to develop a simple spectral method applica-ote sensing estimation of LAI over a complete set ofroecosystems. As a reference, NDVI values were rstfrom the CHRIS reectance spectra and were subse-tted against the corresponding measured LAI valueshough a linear regression through the scatter plot led.687, note from this gure that NDVI starts saturatingund a LAI of 3. Alternatively, a power function, plotted
to a similar r2 of 0.681 and thus it did not improve the
46 J. Delegido et al. / Europ. J. Agronomy 46 (2013) 42 52
Fig. 2. Land us nts weof the referenc rticle.
relationshippotatoes, cothe use of Ntions.
0
1
2
3
4
5
6
0
Me
asure
d L
AI
Fig. 3. Measudataset).e map of the study area during SEN3EXP campaign. The area in which measuremees to colour in this gure legend, the reader is referred to the web version of this a. Given that crops can easily reach LAIs of above 4 (e.g.rn, sugar beet), this saturation is a major obstacle toDVI-related relationships for crop monitoring applica-
0.2 0.4 0.6 0.8 1
Alfalfa
Corn
Garli c
Onion
Potato
Sugarbeet
Sunflower
Vine
Wheat
Bar e Soil
Regression
NDVI
red LAI plotted against NDVI derived from CHRIS spectra (SPARC
4.2. Generi
To optimpossible twspectra in ta and b in twas subseqand statistip-value weshown in Fcient over aan optimizemaximum of a = 674 ahas a bandwof 6 nm. Win the loweation in greand chloropof a chemicSpecicallymaximum avariability iimum of chdirectly relPenuelas, 1LAI and expconventionbe viewed idid not leadabout 4% pore made is marked in red (Delegido et al., 2011b). (For interpretation)c NDIab
ize the retrieval of LAI by means of spectral indices, allo-band combinations have been calculated from CHRIShe form of generic NDIab (according to Eq. (1)) withhe region from 600 to 1000 nm. Each of these indicesuently correlated with green LAI using linear regressioncs such as the coefcient of determination (r2) and there calculated. One of the resulting correlation matrices,ig. 4a, enables us to inspect variation in the r2 coef-ll the two-band combinations. The gure is marked byd region shown in green with strong correlations and a
r2 of 0.717 was obtained by the two-band combinationnd b = 712 nm bands. In CHRIS, the 674 nm wavebandidth of 10 nm while the 712 nm band has bandwidth
ith these two spectral bands positioned in the red andr part of the red-edge, highest sensitivity towards vari-en LAI is obtained. Both bands show sensitivity to LAIhyll variations, and can be considered as the coupling
al absorption (chlorophyll) to a structural variable (LAI)., 674 nm is located in a relative chlorophyll absorptionnd 712 nm is located in the red-edge region, where thes driven by the transition from a maximum to a min-lorophyll absorption and the slope on the red-edge isated to structural effects, in particular LAI (Filella and994). This increases the sensitivity of the index to greenlains the obtained optimized result. In comparison, theal NDVI region with a = 674 nm and b = 803 nm can alson the same gure. Though it can be noted that this point
to most optimal correlations in the matrix; it performedorer compared to the optimized NDIab.
J. Delegido et al. / Europ. J. Agronomy 46 (2013) 42 52 47
Fig. 4. (a) Line binations of bands a and b (nm). (b) Coefcient r2 between measured LAIand calculated e outliers as identied in Fig. 5. Both gures were colour scaled between r2
of 0.6 and 0.85
It is alsoto optimizeall of theseances. Withmatrix the red-edge recombinatioand b = 712 mate most referred to
Given thshows the rsured LAI vathrough thetion:
LAI = 6.769
Although a points fell aThose 12 pwheat planrepresent gtation reseman NDIab inTherefore, tthe t, resu
LAI = 6.753
Given the vto be sufborne supeway.
orde theb). Ily aar determination coefcient r2 between measured LAI and NDIab for different com by NDIab for different combinations of bands a and b (nm) without considering th.
noteworthy that another region of combinations ledd results higher up at 674 and 927 nm (in green). In
combinations yielded the 674 nm band best perform- respect to b in the NDIab formulation, over the whole
In points(Fig. 4siderabbest correlations were obtained precisely within thegion, at 712 nm. Hence, given all possible two-bandns, an NDIab of a = 674, which is also used by NDVI,nm, which falls right in the red-edge, was found to esti-accurately green LAI. This optimized NDVI is hereafteras red-edge NDI and denoted as NDI674712.e above-identied best performing red-edge NDI, Fig. 5esulting relationship between its values and the mea-lues in a scatterplot. A linear relationship can be tted
data points according to the following regression equa-
NDI674712 r2 = 0.717 (2)
satisfactory relationship was obtained, some samplingway from the linear trend (marked with a circle in Fig. 5).oints have been identied as belonging to senescentts, i.e. with dry and yellowish leaves, and thus do notreen vegetation. Reectance spectra of senescent vege-ble closely to dry soil spectra and cannot be detected bydex that is only sensitive to variations in green leaves.hose data points have been reasonably removed fromlting in an improved correlation:
NDI674712 r2 = 0.824 (3)
ariety of crop types included, this relationship seemsciently robust for assessing green LAI from space-rspectral or hyperspectral imageries in a simple
emerged.
-1
0
1
2
3
4
5
6
-0.2
Me
asu
red
LA
I
Fig. 5. Measurbands at a: 67r to assess the inuence of these discarded data r2 matrix was recalculated without the outlierst can be noted that the correlations improved con-nd the same optimal band combinations clearly0 0.2 0.4 0.6 0.8
Alfalfa
Corn
Garli c
Onion
Potato
Sugarbeet
Sunflower
Vine
Wheat
Bare Soil
Regressi on
NDI 674-712
ed LAI plotted against NDIab derived from CHRIS spectra using wave-4 nm and b: 712 nm (SPARC dataset).
48 J. Delegido et al. / Europ. J. Agronomy 46 (2013) 42 52
4.3. Compa
In comphas been evSPARC eld
Table 2r2 coefcients
MCAR
r2 0.61p
J. Delegido et al. / Europ. J. Agronomy 46 (2013) 42 52 49
June 2
explain thedeveloped fvery similar700 nm, whtion is maxradiation isbands aroufor estimatiZhao et al., with LAI, ethe red-edgreduced seused to proin precision2010).
4.4. Validat
Validatioimportant sdataset wasin estimatinrst used. Twere subseues. The resto the follow
LAI = 0.91where NDIabands, as ovtral displaccorrelation step consistimage, whi(Fig. 7).
en thingulue-tg froly ves andthat The Fig. 7. Final LAI map over the study site as obtained from the 19
ir good performance. TCARI, TCI and TVI were originallyor the estimation of chlorophyll content and they are
to each other. TCI and TCARI use bands at 550, 670 andile TVI uses 550, 670 and 750 nm. Chlorophyll absorp-imized at 670 nm whereas a large portion of the solar
reected at 550 nm. These bands in combination withnd the red-edge (700, 750 nm) proved to be successfulng LAI (see also Lee et al., 2004; Thenkabail et al., 2000;
Givbe distdark-branginsparseowertypes yards. 2007). The remaining indices yielded poor correlationither because not using the 670 nm band or bands ine region or because of combining indices that causednsitivity, like MCARI/OSAVI, TCARI/OSAVI, originallyvide predictive relationships for chlorophyll estimation
agriculture (Haboudane et al., 2008; Meggio et al.,
ion
n of retrieval methods using independent datasets is antep in evaluating its actual performance. The SEN3EXP
used for validation of the red-edge NDI on its capabilityg LAI. Similar to the SPARC approach, a CHRIS image washe red-edge NDI was calculated. LAI eld measurementsquently linked to the corresponding red-edge NDI val-ults can be linearly tted by a regression line accordinging equation:
876 + 13.448NDI672.7710.5 r2 = 0.905 (4)
b has been calculated using the 672.70 and 710.50 nmer time CHRIS bands have suffered from a small spec-
ement. The regression equation yielded a strong linearwith eld LAI measurements. Subsequently the nals of applying this equation over the 29 June 2009 CHRISch leads to a LAI map over the Barrax agroecosystems
between 2 alfalfa, garldark-red pa7. Overall, forms adeqagroecosys
Furthermspatial resoations in LAprecision fa
The validthe root meLAI eld mmap (Fig. 8estimated LLAI vales ovin the eld.
5. Discussi
Since agonce or twtoring applavailabilityagroecosyset al., 200superspectr009 CHRIS image (SEN3EXP dataset).
e nal map in Fig. 7, different agricultural parcels canished based on spatial patterns of LAI estimations. Theo-blue colour tones represent low LAI distributions,m values close to zero to two and cover bare soils andgetated crop types (onion and some elds of garlic, sun-
corn which are in early growth stage) or row cropat pixel-level exhibit a low plant cover such as vine-light-blue-to-yellow colour tones represent LAI values
and 3 and cover the majority of major crop types suchic and some more developed corn and sunowers. Thercels represent potatoes elds with a high LAI aroundthe map shows that the proposed red-edge NDI per-uately for obtaining large areas LAI maps over an entiretem from space-based imagery.ore, Fig. 7 shows that the methodology applied to high
lution imagery allows identication of within-eld vari-I, which makes the approach potentially applicable torming.ity of the LAI map in Fig. 7 was assessed by calculatingan square error (RMSE). To do so, we have plotted theeasurements against the estimates obtained from the). It led to an RMSE deviation between measured andAI of 0.55. Note hereby the extremely high measureder potatoe crops observed both in the derived map as
on
ro-technical decisions are routinely made by the farmerice a week, a simple, robust and up-to-date moni-ication would be most welcome. Specically, frequent
of LAI maps will allow the farmer to better monitortems dynamics in time at the landscape level (Dorigo7). With the advent of high spatial resolution andal sensors, remote sensing techniques have become
50 J. Delegido et al. / Europ. J. Agronomy 46 (2013) 42 52
-2
0
2
4
6
8
-1
Me
asure
d L
AI
Fig. 8. Measur(SEN3EXP data
particularlysuch as LAI.2 mission wspectral banregion, anditoring appresolution. site about eto monitor cdone throuposed red-eof these kining applicatscale (Casa
The simpof LAI in a sspectral vegmodels thaRichter et aplest way tescape our nicantly sGlenn et albe stressedopment of of a more rexperiencedNDI provedwith greenvalues. It sships have involving anmechanismAlthough resituation anproved thawas essenteasily recal
The regression equation was validated against an independentdataset (SEN3EXP). Herewith, it appeared that the linear ttingbetween NDI and LAI deviated somewhat between model devel-
t (SPARC) and validation data. Apart from differences inonditc cor. It shl sathiftired ngthot imave cracteant cge slthe med fo
aims
nd windicensit0 anins tin morthwmorle bictiondents a s
furthetatr cam
clus
ile tstratped 0 1 2 3 4 5 6 7
AlfalfaCornFruitGarlicOnionPotat oSunf lowerVineBare Soil1:1 li ne
Calculated LAI
ed LAI versus estimated LAI derived from CHRIS spectra using Eq. (4)set.)
attractive for assessing crop biophysical parameters For instance ESAs forthcoming superspectral Sentinel-ill soon become available to users. Sentinel-2 has ad-settings that is optimized for measuring in red-edge
is dedicated to vegetation and agroecosystem mon-lications at a high spatial (up to 10 m) and temporalUnder cloud-free conditions Sentinel-2 revisit the sameach 5th day (ESA, 2010), which would allow the farmerrop growth on a weekly basis. This monitoring could begh image-wide LAI mapping by applying the here pro-dge NDI (NDI674712). It is foreseen that the availabilityds of LAI maps will open new agroecosystem monitor-ion in the coming years, ranging from eld up to globalet al., 2012).
opmenlocal cspherisensorimentasmall sconguwaveledoes nmay hits chasignicred-edwhen designwhich2010).
To elished good s550, 67it remabands be wolished multipand fratance ipossesaim toent vegof othe
6. Con
Whdemondevelole regression equation is intended for rapid predictiontraightforward way. In fact, regression models based onetation indices may be preferable to physically-based
t are complex to design and parameterize (Liang, 2007;l., 2009). Statistical approaches are amongst the sim-o predict biophysical parameters, however it does notattention that they provide relationships that are sig-pace, time and species dependent (Casa et al., 2012;., 2008; Verrelst et al., 2008). In this respect, it should
that the objective of this article was not the devel-a regression equation itself, but rather the deliveringobust index that could bypass much of the difculties
with broandband indices. The here proposed red-edge to be successful in establishing a linear relationship
LAI without being prone to saturation at higher LAIhould herewith be noted that the observed relation-been derived purely from experimental data, withouty modelling that makes assumptions about underlyings in establishing biochemical/structural relationships.gression coefcients can vary depending on the locald sensor characteristics, validation of the red-edge NDIt the intrinsic relationship between LAI and the indexially stable. Henceforth, regression coefcients can beibrated.
ground or atype or in owork a red-posed and large variethyperspect2003 and 2hyperspectwere colleccal states atwo-band ca generic vled to best 674 nm, whphyll absor712 nm, whstrongly reslope relatethis proposvalues. Anodetected inNDI) and 92in view of sedge NDI hions, this can also be due to imperfections in the atmo-rection or due to spectral degradation of the CHRISould be mentioned that CHRIS/PROBA, being an exper-ellite with initially a one year lifetime suffers from ang of wavelengths over time. Bands that were originallyat 674.42 nm and 712.17 nm were shifted to shorters to 672.70 nm and 710.50 nm. While this shift of 2 nmpact much the relatively stable red spectral region, itonsiderable impact in the red-edge region. Because ofristic steep slope, it means that a 2 nm shift can causehanges in the NDIab, i.e., being lower positioned in the
ope. Nevertheless, this anomaly is not foreseen to occurethod is applied to images originating from sensors
r operational use, such as forthcoming ESAs Sentinel-2 to deliver a consistent data ow for 12 years long (ESA,
ith, the red-edge NDI has been compared against estab-es. Only the 3-band indices TCARI, TCI and TVI showedivities to LAI. Both of these indices rely on bands atd around the red-edge region (700, 750 nm), althougho be evaluated whether these were the most optimizedre variable crop conditions. This suggests that it wouldhile to analyze further the use of generic and estab-e advanced (3- or 4-band) indices and relate them toophysical parameters such as LAI, chlorophyll contental vegetation cover. On the other hand, it is of impor-
ifying whether promising indices, e.g. the red-edge NDI,trong universality. Therefore, in a follow-up study weer evaluate the utility of promising indices over differ-ion types and atmospheric conditions, e.g. by using datapaigns (Delegido et al., 2011a).
ions
he utility of the red-edge spectral region has beened in various studies, the majority of these studies haveempirical relationships with green LAI on the basis ofirborne hyperspectral data, typically only for one cropne growth stage. In addition to existing indices, in thisedge normalized difference index (NDI) has been pro-validated using spaceborne hyperspectral data over ay of crops. Based on LAI eld measurements and CHRISral data simultaneously collected during the summer004 ESA SPARC campaigns, we have fully exploited theral information available in the CHRIS image. LAI datated over 10 different crop types in various phenologi-nd water regimes. The predictive power of all availableombinations have been analyzed according to NDIab,ariation of the NDVI formulation. The wavebands thatcorrelation with the LAI dataset were encountered atich is precisely situated in the region of maximal chloro-ption and also used by the conventional NDVI, and atich is situated in the red-edge region, a region that is
lated to the physiological status of the plant (red-edged to LAI). It led to an r2 of 0.82, and contrary to the NDVIed red-edge NDI did not lead to saturation at higher LAIther two-band region with high correlations has been
the waveband situated around 674 (same as red-edge7 nm, situated in the NIR. Such index can be of interestensors that have no red-edge bands available. The red-as been subsequently compared against other widely
J. Delegido et al. / Europ. J. Agronomy 46 (2013) 42 52 51
used established vegetation indices using the SPARC dataset. Thered-edge NDI outperformed most of the indices but also the 3-bandindices TCARI, TCI and TVI showed good sensitivities to LAI with anr2 on the order of 0.75.
Finally, validated bobservationcampaign. Itained, althregression lCHRIS imagbetween m
The metin LAI, whicsion farmin
Acknowled
This wo21432-C02-CompetitivIEF grant #2
References
Alonso, L., MgeometricCHRIS/Prohttp://eart
Aparicio, N., Vindices as Journal 92
Bannari, A., Khspectral chlaboratoryRemote Se
Baret, F., Guyoassessmen
Barnsley, M.J.,mission, aearth surfSensing 42
Broge, N.H., Mophyll densEnvironme
Broge, N.H., Leband and hand canop
Casa, R., Varellmodel witof model aEuropean
Chen, J.M., CihLandsat TM
Chen, J.M., PavD.R., King,coarse-resand groun
Clevers, J.G.P.WA.K., Scholband settin
Cohen, W.B., Mfor regressof Environ
Darvishzadeh,etation LAarchitectu
Dash, J., CurraJournal of
Daughtry, C.S.2000. Estireectance
Delegido, J., Fetent and LAdata. Inter
Delegido, J., Veedge bandSensors 11
Delegido, J., Vergara, C., Verrelst, J., Ganda, S., Moreno, J., 2011b. Remote estimationof crop chlorophyll content by means of high-spectral-resolution reectancetechniques. Agronomy Journal 103, 18341842.
Dorigo, W.A., Zurita-Milla, R., de Wit, A.J.W., Brazile, J., Singh, R., Schaepman, M.E.,2007. A review on reective remote sensing and data assimilation techniques
nhanrvatio2010.://ww12. G://wwt, K.S
cal ancultur
Colomticat
chara243.ez, G.,
variabSPARerlan, Penurophying 15.A., M
sland al JourJ.A., Ps diurt 41, 3nto, Gethodgen s, A., Mctancf Pho, A., V
of can03, ht.P., H
otely-esses:ors 8,
N., Pices op
Trans., 198
of bio212.z-Sanptionsote Se, L., Alo
Probsactioane, Dgratedent fo164
ane, Dtral v
canopote Seane, D
chlor Trans., Tibs
Miniuo, X.
hyper07.n, I., P
ssmening ofn, C.Msed on
197, R., Brse anote SeV., Guintergen roud, Stationthe predictive power of the red-edge NDI has beeny independent LAI eld measurements and satellites collected over the same site during the 2009 SEN3EXPt was observed that a high linear correlation was main-ough the numerical result depends on the slope of theine. The obtained regression equation was applied to ae for LAI mapping and validated with an RMSE deviationeasured and estimated LAI of 0.6.hodology allows identication of within-eld variationsh makes the approach potentially applicable to preci-g when applied to high spatial resolution imagery.
gement
rk has been made possible by the project AYA2010-01, funded by the Spanish Ministry of Economy andeness. J. Verrelst is supported by the EU Marie Currie52237.
oreno, J., 2005. Advances and limitations in a parametric correction of CHRIS/Proba data. In: Proceedings of the 3rdba Workshop, ESA/ESRIN, Frascati, Italy, Available on line at:h.esa.int/workshops/chris proba 05/papers/06 alonso.pdfillegas, D., Casadesus, J., Araus, J.L., Royo, C., 2000. Spectral vegetationnondestructive tools for determining durum wheat yield. Agronomy, 8391.urshid, K.S., Staenz, K., Schwarz, J.W., 2007. A comparison of hyper-lorophyll indices for wheat crop chlorophyll content estimation using
reectance measurements. IEEE Transactions on Geoscience andnsing 45, 30633074.t, G., 1991. Potentials and limits of vegetation indices for LAI and APARt. Remote Sensing of Environment 35, 161173.
Settle, J.J., Cutter, M.A., Lobb, D.R., Teston, F., 2004. The PROBA/CHRIS low-cost smallsat for hyperspectral multiangle observations of theace and atmosphere. IEEE Transactions on Geoscience and Remote, 15121520.rtensen, J.V., 2002. Deriving green crop area index and canopy chloro-ity of winter wheat from spectral reectance data. Remote Sensing ofnt 81, 4557.blanc, E., 2000. Comparing prediction power and stability of broad-yperspectral vegetation indices for estimation of green leaf area indexy chlorophyll density. Remote Sensing of Environment 76, 156172.a, H., Buis, S., Gurif, M., Solan, B., Baret, F., 2012. Forcing a wheat croph LAI data to access agronomic variables. Evaluation of the impactnd LAI uncertainties and comparison with an empirical approach.
Journal of Agronomy 37, 110.lar, J., 1996. Retrieving leaf area index of boreal conifer forests using
images. Remote Sensing of Environment 55, 153162.lic, G., Brown, L., Cihlar, J., Leblanc, S.G., White, H.P., Hall, R.J., Peddle,
D.J., Trofymow, J.A., 2002. Derivation and validation of Canada-wideolution leaf area index maps using high-resolution satellite imageryd measurements. Remote Sensing of Environment 80, 165184.., de Jong, S.M., Epema, G.F., van der Meer, F., Bakker, W.H., Skidmore,
te, K.H., 2002. Derivation of the red edge index using MERIS standardg. International Journal of Remote Sensing 23, 31693184.aiersperger, T.K., Gower, S.T., Turner, D.P., 2003. An improved strategyion of biophysical variables and Landsat ETM + data. Remote Sensingment 84, 561571.
R., Skidmore, A., Atzberger, C., Wieren, S., 2008. Estimation of veg-I from hyperspectral reectance data: effects of soil type and plantre. International Journal of Applied Earth Observation 10, 358373.n, P.J., 2004. The MERIS terrestrial chlorophyll index. InternationalRemote Sensing 25, 54035413.T., Walthall, C.K., Kim, M.S., Brown de Costoun, E., McMurtrey, J.E.,mating corn leaf chlorophyll concentration from leaf and canopy. Remote Sensing of Environment 74, 229239.rnndez, G., Ganda, S., Moreno, J., 2008. Retrieval of chlorophyll con-I of crops using hyperspectral techniques: application to Proba/CHRISnational Journal of Remote Sensing 29, 71077127.rrelst, J., Alonso, L., Moreno, J., 2011a. Evaluation of Sentinel-2 red-s for empirical estimation of green LAI and chlorophyll content., 70637081.
for eObse
ESA, http
ESA, 20http
FassnachoptiAgri
Fava, F.,Identure233
Fernndticaland Neth
Filella, I.chloSens
Friedl, Mgrastion
Gamon, trackmen
GianquiA mnitro
Gitelsonreenal o
GitelsontionL084
Glenn, EremprocSens
Gobron,indiIEEE
Goel, N.Stion3, 1
GonzlevariaRem
GuanterfromTran
HaboudIntecont81, 4
HaboudspeccropRem
HaboudcropIEEE
Hastie, TData
He, Y., Gable981
HerrmanasseSens
HoffmanL. Ba190,
HouborginveRem
Houles, for wnitro
Jacquemvegeced agroecosystem modeling. International Journal of Applied Earthn 9, 165193.
GMES Sentinel-2 Mission requirements document.w.esa.int/esaLP/SEMM4T4KXMF LPgmes 0.htmlMES (Global Monitoring for Environment and Security) info.w.gmes.info/., Gower, S.T., Norman, J.M., McMurtrie, R.E., 1994. A comparison ofd direct methods for estimating foliage surface area index in forests.al and Forest Meteorology 71, 183207.bo, R., Bocchi, S., Meroni, M., Sitzia, M., Fois, N., Zucca, C., 2009.
ion of hyperspectral vegetation indices for Mediterranean pas-cterization. International Journal of Applied Earth Observation 11,
Moreno, J., Ganda, S., Martnez, B., Vuolo, F., Morales, F., 2005. Statis-ility of eld measurements of biophysical parameters in SPARC-2003
C-2004 campaigns. In: Proc. SPARC Workshop, ESA, Enschede, Theds.elas, J., 1994. The red edge position and shape as indicators of plantll content biomass and hydric status. International Journal of Remote, 14591470.ichaelsen, J., Davis, F.W., Walker, H., Schimel, D.S., 1994. Estimatingbiomass and leaf area index using ground and satellite data. Interna-nal of Remote Sensing 15, 14011420.enuelas, J., Field, C.B., 1992. A narrow-waveband spectral index thatnal changes in photosynthetic efciency. Remote Sensing of Environ-544.., Orsini, F., Fecondini, M., Mezzetti, M., Sambo, P., Bona, S., 2011.ological approach for dening spectral indices for assessing tomatotatus and yield. European Journal Agronomy 35, 135143.erzlyak, M., 1994. Quantitative estimation of chlorophyll-a using
e spectra: experiments with autumn chestnut and maple leaves. Jour-tochemistry and Photobiology B: Biology 22, 247252.ina, A., Ciganda, V., Rundquist, D., Arkebauer, J., 2005. Remote estima-opy chlorophyll content in crops. Geophysical Research Letters 32,tp://dx.doi.org/10.1029/2005GL022688.uete, A.R., Nagler, P.L., Nelson, S.G., 2008. Relationship betweensensed vegetation indices, canopy attributes and plant physiological
what vegetation indices can and cannot tell us about the landscape. 21362160.nty, B., Verstraete, M.M., Widlowski, J.L., 2000. Advanced vegetationtimized for up-coming sensors: design performance and applications.actions on Geoscience and Remote Sensing 38, 24892505.7. Models of vegetation canopy reectance and their use in the estima-physical parameters from reectance data. Remote Sensing Reviews
edro, M.C., Le Toan, T., Moreno, J., Kergoat, L., Rubio, E., 2008. Seasonal of leaf area index of agricultural elds retrieved from Landsat data.nsing of Environment 112, 810824.nso, L., Moreno, J., 2005. A method for the surface reectance retrievala/CHRIS data over land: application to ESA SPARC campaigns. IEEEns on Geoscience and Remote Sensing 43, 29082917.., Miller, J.R., Tremblay, N., Zarco-Tejada, P.J., Dextraze, L., 2002.
narrow-band vegetation indices for prediction of crop chlorophyllr application to precision agriculture. Remote Sensing of Environment26.., Miller, J.R., Pattey, E., Zarco-Tejada, P.J., Strachan, I.S., 2004. Hyper-egetation indices and novel algorithms for predicting green LAI ofies: modeling and validation in the context of precision agriculture.nsing of Environment 90, 337352.., Tremblay, N., Miller, J.R., Vigneault, P., 2008. Remote estimation ofophyll content using spectral indices derived from hyperspectral data.actions on Geoscience and Remote Sensing 46, 423437.hirani, R., Friedman, J.H., 2009. The Elements of Statistical Learning:ng, Inference, and Prediction, second ed. Springer-Verlag, New York., Wilmshurst, J., 2006. Studying mixed grassland ecosystems I: suit-spectral vegetation indices. Canadian Journal of Remote Sensing 32,
imstein, A., Karnieli, A., Cohen, Y., Alchanatis, V., Bonl, D.J., 2011. LAIt of wheat and potato crops by VENS and Sentinel-2 bands. Remote
Environment 115, 21412151.., Blomberg, M., 2004. Estimation of leaf area index of Beta vulgaris
optical remote sensing data. Journal of Agronomy and Crop Science204.oegh, E., 2008. Mapping leaf chlorophyll and leaf area index usingd forward canopy reectance modeling and SPOT reectance data.nsing of Environment 112, 186202.erif, M., Mary, B., 2007. Elaboration of a nitrogen nutrition indicator
wheat based on leaf area index and chlorophyll content for makingecommendations. European Journal of Agronomy 27, 111.., Baret, F., Andrieu, B., Danson, F.M., Jaggard, K., 1995. Extraction of
biophysical parameters by inversion of the PROSPECT + SAIL model
52 J. Delegido et al. / Europ. J. Agronomy 46 (2013) 42 52
on sugar beet canopy reectance data application to TM data. Remote Sensingof Environment 52, 163172.
Jordan Carl, F., 1969. Derivation of leaf-area index from quality of light on the forestoor. Ecology 50, 663666.
Law, B.E., Waring, R.H., 1994. Remote sensing of leaf area index and radia-tion intercepted by understory vegetation. Journal of Applied Ecology 4,272279.
Lee, K.S., Cohen, W.B., Kennedy, R.E., Maiersperger, T.K., Gowerd, S.T., 2004. Hyper-spectral versus multispectral data for estimating leaf area index in four differentbiomes. Remote Sensing of Environment 91, 508520.
Le Maire, G., Franc ois, C., Soudani, K., Berveiller, D., Pontailler, S., Brda, N., Genet, H.,Davi, H., Dufrne, E., 2008. Calibration and validation of hyperspectral indicesfor the estimation of broadleaved forest leaf chlorophyll content leaf mass perarea, leaf area index and leaf canopy biomass. Remote Sensing of Environment112, 38463864.
Liang, S., 2007. Recent developments in estimating land surface biogeophysical vari-ables from optical remote sensing. Progress in Physical Geography 31, 501516.
Liu, J., Miller, J.R., Haboudane, D., Pattey, E., 2004. Exploring the relationship betweenred edge parameters and crop variables for precision agriculture. In: Geoscienceand Remote Sensing Symposium, 2004. IGARSS 04. Proceedings. IEEE Interna-tional.
Malenovsky, Z., Rott, H., Cihlar, J., Schaepman, M., Garcia-Santos, G., Fernandes, R.,Berger, M., 2012. Sentinels for Science: potential of Sentinel-1, -2, and -3 mis-sions for scientic observations of ocean, cryosphere, and land. Remote Sensingof Environment 120, 91101.
Meggio, F., Zarco-Tejada, P.J., Nnez, L.C., Sepulcre-Cant, G., Gonzlez, M.R., Martn,P., 2010. Grape quality assessment in vineyards affected by iron deciencychlorosis using narrow-band physiological remote sensing indices. RemoteSensing of Environment 114, 19681986.
Moran, M.S., Maas, S.J., Vanderbilt, V.C., Barnes, M., Miller, S.N., Clarke, T.R., 2004.Application of image-based remote sensing to irrigated agriculture. In: Ustin,S.L. (Ed.), Remote Sensing for Natural Resource Management and Environmen-tal Monito617676.
Moreno, J., AGarca, J.Coverview CHRIS/Prohttp://eart
Mutanga, O., Sestimationratory con
Myneni, R.B., Mof vegetatment 51, 1
Nguyen, H.T., LhyperspecJournal of
Oppelt, N., Maof wheat dRemote Se
Ray, D., Das, indices foferent irri5373538
Richter, K., Amental aretrieval o230247.
Rondeaux, G., Steven, M., Baret, F., 1996. Optimization of soil-adjusted vegetationindices. Remote Sensing of Environment 55, 951907.
Rouse, J.W., Haas, R.H., Schell, J.A., Deering, D.W., 1973. Monitoring vegetation sys-tems in the great plains with ERTS. In: Third ERTS Symposium, NASA SP-351,vol. 1, NASA, Washington, DC, pp. 309317.
Sakamoto, T., Gitelson, A., Nguy-Robertson, A., Arkebauer, T., Wardlow, B., Suyker,A., Verma, S., Shibayama, M., 2012. An alternative method using digital camerasfor continuous monitoring of crop status. Agricultural and Forest Meteorology154, 113126.
Schaepman, M.E., Ustin, S.L., Plaza, A.J., Painter, T.H., Verrelst, J., Liang, S., 2009. Earthsystem science related imaging spectroscopyan assessment. Remote Sensingof Environment 113, 123137.
Scurlock, J.M.O., Asner, G.P., Gower, S.T., 2001. Worldwide historical estimates andbibliography of leaf area index, 19322000. In: ORNL Technical MemorandumTM-2001/268, Oak Ridge National Laboratory, Oak Ridge, Tennessee, U.S.A.
Stagakis, S., Markos, N., Sykioti, O., Kyparissis, A., 2010. Monitoring canopy biophysi-cal and biochemical parameters in ecosystem scale using satellite hyperspectralimagery: an application on a Phlomis fruticosa Mediterranean ecosystem usingmultiangular CHRIS/PROBA observations. Remote Sensing of Environment 114,977994.
Thenkabail, P., Smith, R.B., Pauw, E., 2000. Hyperspectral vegetation indices and theirrelationships with agricultural crop characteristics. Remote Sensing of Environ-ment 71, 158182.
Turner, D.P., Cohen, W.B., Kennedy, R.E., Fassnacht, K.S., Briggs, J.M., 1999. Relation-ships between leaf area index and Landsat TM spectral vegetation indices acrossthree temperate zone sites. Remote Sensing of Environment 70, 5268.
Verger, A., Baret, F., Weiss, M., 2008. Performances of neural networks for derivingLAI estimates from existing CYCLOPES and MODIS products. Remote Sensing ofEnvironment 112, 27892803.
Verrelst, J., Schaepman, M.E., Koetz, B., Kneubuhler, M., 2008. Angular sensitivityanalysis of vegetation indices derived from CHRIS/PROBA data. Remote Sensingof Environment 112, 23412353.
, J., CarameIS/PRO402
, J., Mu012. Meval:
127., Bar
n situsampliner, Cdeumnal of J.M., Nitectuan, X
for chemoteiu, Q.,-plant
941, Huan
derivS Jour., Mosries, mring, Manual of Remote Sensing. John Wiley & sons, Hoboken, pp.
lonso, L., Fernndez, G., Fortea, J.C., Ganda, S., Guanter, L.,., Mart, J.M., 2004. The spectra Barrax campaing (SPARC):and rst results from Chris data. In: Proceedings of the 2ndba Workshop, ESA/ESRIN, Frascati, Italy, Available online ath.esa.int/workshops/chris proba 04/papers/10 MOREN.pdfkidmore, A.K., 2004. Hyperspectral band depth analysis for a better
of grass biomass (Cenchrus ciliaris) measured under controlled labo-ditions. International Journal of Applied Earth Observation 5, 8796.aggion, S., Iaquinta, J., Privette, J.L., et al., 1995. Optical remote sensing
ion: modeling, caveats, and algorithms. Remote Sensing of Environ-69188.ee, B.-W., 2006. Assessment of rice leaf growth and nitrogen status bytral canopy reectance and partial least square regression. EuropeanAgronomy 24, 349356.user, W., 2004. Hyperspectral monitoring of physiological parametersuring a vegetation period using AVIS data. International Journal ofnsing 25, 145159.G., Singh, J.P., Panigrahy, S., 2006. Evaluation of hyperspectralr LAI estimation and discrimination of potato crop under dif-gation treatments. International Journal of Remote Sensing 27,7.tzberger, C., Vuolo, F., Weihs, P., DUrso, G., 2009. Experi-
ssessment to the Sentinel-2 band setting for RTM-based LAIf sugar beet and maize. Canadian Journal of Remote Sensing 35,
Verrelstk pCHR4014
VerrelstJ., 2retri118,
Weiss, Mfor iand
Weisste(HorJour
Welles, arch
Wu, C., Hdataof R
Yao, Y., Lrow112,
Zhao, D.bandISPR
Zheng, Gtheolevers, J.G.P.W., Schaepman, M.E., 2010. Merging the Minnaert-ter with spectral unmixing to map forest heterogeneity withBA data. IEEE Transactions on Geoscience and Remote Sensing 48,
2.noz, J., Alonso, L., Delegido, J., Rivera, J.P., Camps-Valls, G., Moreno,achine learning regression algorithms for biophysical parameter
opportunities for Sentinel-2 and -3. Remote Sensing of Environment139.et, F., Smith, G.J., Jonckheere, I., Coppin, P., 2004. Review of methods
leaf area index (LAI) determination. Part II. Estimation of LAI, errorsing. Agricultural and Forest Meteorology 121, 3753..J., Khbauch, W., 2005. Regional yield forecasts of malting barley
vulgare L) by NOAA-AVHRR remote sensing data and ancillary data.Agronomy and Crop Science 191, 308320.orman, J.M., 1991. Instrument for indirect measurement of canopyre. Agronomy Journal 83, 818825.., Niu, Z., Dong, J., 2010. An evaluation of EO-1 hyperspectral Hyperionlorophyll content and leaf area index estimation. International Journal
Sensing 31, 10791086. Liu, Q., Li, X., 2008. LAI retrieval and uncertainty evaluations for typicaled crops at different growth stages. Remote Sensing of Environment06.g, L., Li, J., Qi, J., 2007. A comparative analysis of broadband and narrow-ed vegetation indices in predicting LAI and CCD of a cotton canopy.nal of Photogrammetry 62, 2533.kal, L.M., 2009. Retrieving leaf area index (LAI) using remote sensing:ethods and sensors. Sensors 9, 27192745.
A red-edge spectral index for remote sensing estimation of green LAI over agroecosystems1 Introduction2 Methods3 Experimental dataset3.1 SPARC3.2 SEN3EXP
4 Results4.1 NDVI4.2 Generic NDIab4.3 Comparison with other indices4.4 Validation
5 Discussion6 ConclusionsAcknowledgementReferences