Biogeosciences, 18, 621–635, 2021
h ps://doi.o g/10.5194/bg-18-621-2021
© Au ho (s) 2021. This wo k is dis ibu ed unde
he C ea i e Commons A ibu ion 4.0 License.
Re ie al and alida ion o o es backg ound e lec i i y om daily
Mode a e Resolu ion Imaging Spec o adiome e (MODIS)
bidi ec ional e lec ance dis ibu ion unc ion (BRDF) da a ac oss
Eu opean o es s
Jan Pisek1, Angela E b2, Lau i Ko honen3, Tobias Bie mann4, A naud Ca a a5, Edoa do C emonese6,
Ma hias Cun z7, Sil ano Fa es8, Giacomo Ge osa9, Thomas G ünwald10, Niklas Hase11, Michal Heliasz4,
And eas Ib om12, Alexande Knohl13, Johannes Koble 14, Ba K uij 15, Holge Lange16, Leena Leppänen17,
Jean-Ma c Limousin18, F ancisco Ramon Lopez Se ano19, Denis Lous au20, Pe Lukeš21, La s Lundin22,
Ricca do Ma zuoli9, Meelis Mölde 4, Leona do Mon agnani23,31, Johan Nei ynck24, Ma hias Peichl25,
Co inna Rebmann11, E a Rubio19, Ma ga ida San os-Reis26, C ys al Schaa 2, Ma ius Schmid 27,
Guillaume Simioni28, Kamel Soudani29, and Ca oline Vincke30
1Ta u Obse a o y, Uni e si y o Ta u, Tõ a e e, Ta umaa, Es onia
2School o he En i onmen , Uni e si y o Massachuse s Bos on, Bos on, Massachuse s, USA
3School o Fo es Sciences, Uni e si y o Eas e n Finland, Joensuu, Finland
4Lund Uni e si y, Lund, Sweden
5Fundación CEAM, Pa e na, Valencia, Spain
6ARPA Valle d’Aos a, Sain -Ch is ophe, I aly
7Uni e si é de Lo aine, Ag oPa isTech, INRAE, UMR Sil a, Nancy, F ance
8CNR – Na ional Resea ch Council, Rome, I aly
9Depa men o Ma hema ics and Physics, Uni e si à Ca olica del Sac o Cuo e, B escia, I aly
10Ins i u e o Hyd ology and Me eo ology, Depa men o Hyd o Sciences, Technische Uni e si ä D esden,
D esden, Ge many
11Helmhol z Cen e o En i onmen al Resea ch – UFZ, Leipzig, Ge many
12Depa men o En i onmen al Enginee ing, Technical Uni e si y o Denma k, Kongens Lyngby, Denma k
13Facul y o Fo es Sciences and Fo es Ecology, Uni e si y o Gö ingen, Gö ingen, Ge many
14Umwel bundesam GmbH, Vienna, Aus ia
15Depa men o En i onmen al Sciences, Wageningen Uni e si y & Resea ch, Wageningen, he Ne he lands
16No wegian Ins i u e o Bioeconomy Resea ch, Ås, No way
17Space and Ea h Obse a ion Cen e, Finnish Me eo ological Ins i u e, Sodankylä, Finland
18CEFE, Uni e si é Mon pellie , CNRS, EPHE, IRD, Uni e si é Paul-Valé y Mon pellie , Mon pellie , F ance
19IER-ETSIAM, Uni e sidad de Cas illa-La Mancha, Albace e, Spain
20INRAE, Bo deaux, F ance
21Global Change Resea ch Ins i u e, Academy o Sciences o he Czech Republic, B no, Czech Republic
22Depa men o Soil and En i onmen , Swedish Uni e si y o Ag icul u al Sciences, Uppsala, Sweden
23Facul y o Science and Technology, F ee Uni e si y o Bolzano, Bolzano, I aly
24INBO, Ge aa dsbe gen, Belgium
25Depa men o Fo es Ecology and Managemen , Swedish Uni e si y o Ag icul u al Sciences, Umeå, Sweden
26cE3c – Cen e o Ecology, E olu ion and En i onmen al Changes, Lisbon, Po ugal
27Fo schungszen um Jülich, Jülich, Ge many
28INRAE URFM, A ignon, F ance
29Uni e si é Pa is-Saclay, CNRS, Ag oPa isTech, Ecologie, Sys éma ique e E olu ion, O say, F ance
30Facul y o Bioscience Enginee ing, Ea h and Li e Ins i u e, Uni e si é Ca holique de Lou ain, Lou ain-la-Neu e, Belgium
Published by Cope nicus Publica ions on behal o he Eu opean Geosciences Union.
622 J. Pisek e al.: Re ie al and alida ion o o es backg ound e lec i i y
31Fo es Se ices, Au onomous P o ince o Bolzano, Bolzano, I aly
Co espondence: Jan Pisek ([email p o ec ed])
Recei ed: 30 Sep embe 2020 – Discussion s a ed: 4 No embe 2020
Re ised: 12 Janua y 2021 – Accep ed: 12 Janua y 2021 – Published: 27 Janua y 2021
Abs ac . In o ma ion abou o es backg ound e lec ance
is needed o accu a e biophysical pa ame e e ie al om
o es canopies (o e s o y) wi h emo e sensing. Sepa a ing
unde - and o e s o y signals would enable mo e accu a e
modeling o o es ca bon and ene gy luxes. We e ie ed
alues o he no malized di e ence ege a ion index (NDVI)
o he o es unde s o y wi h he mul i-angula Mode a e
Resolu ion Imaging Spec o adiome e (MODIS) bidi ec-
ional e lec ance dis ibu ion unc ion (BRDF)/albedo da a
(g idded 500m daily Collec ion 6 p oduc ), using a me hod
o iginally de eloped o bo eal o es s. The o es loo back-
g ound e lec ance es ima es om he MODIS da a we e
compa ed wi h in si u unde s o y e lec ance measu emen s
ca ied ou a an ex ensi e se o o es ecosys em expe i-
men al si es ac oss Eu ope. The e lec ance es ima es om
MODIS da a we e, hence, es ed ac oss di e se o es con-
di ions and phenological phases du ing he g owing season
o examine hei applicabili y o ecosys ems o he han bo-
eal o es s. He e we epo ha he me hod can deli e good
e ie als, especially o e di e en o es ypes wi h open
canopies (low oliage co e ). The pe o mance o he me hod
was ound o be limi ed o e o es s wi h closed canopies
(high oliage co e ), whe e he signal om unde s o y be-
comes oo a enua ed. The spa ial he e ogenei y o indi id-
ual ield si es and he limi a ions and documen ed quali y o
he MODIS BRDF p oduc a e shown o be impo an o he
co ec assessmen and alida ion o he e ie als ob ained
wi h emo e sensing.
1 In oduc ion
The e lec ance om he o es canopy backg ound/ o es
loo can o en con ound and e en domina e he adiome -
ic signal om he uppe o es canopy laye o he a mo-
sphe e. Fo es unde s o y is de ined he e as all he compo-
nen s ound unde he o es canopy, including unde s o y
ege a ion, lea li e , moss, lichen, ock, soil, snow, o a
mix u e he eo (Pisek and Chen, 2009). I unaccoun ed o ,
o es unde s o y can in oduce po en ial bias in he es ima-
ion o o e s o y biophysical pa ame e s (e.g., lea a ea in-
dex, LAI, and ac ion o abso bed pho osyn he ically ac i e
adia ion, APAR) and, subsequen ly, p oduc i i y es ima es
(e.g., he ne p ima y p oduc i i y – NPP) as he con ibu ion
o he unde s o y o he o al ene gy abso p ion capaci y o a
o es s and can be qui e signi ican (Cla k e al., 2001; Law
e al., 2001). The unde s o y ege a ion in o es ecosys ems
should be ea ed di e en ly om he o e s o y in ca bon cy-
cle modeling because o he di e en esidence imes o ca -
bon ixed h ough NPP in di e en ecosys em componen s
(Vogel and Gowe , 1998; Ren ch e al., 2003; Ma ques and
Oli ei a, 2004; Kim e al., 2016). Cu en ly, he unde s o y
is o en ea ed as an unknown quan i y in ca bon models
due o he di icul ies in measu ing i p ope ly and consis-
en ly ac oss la ge scales (Luyssae e al., 2007). The p e-
dic ions ega ding he spec al a ia ion in o es backg ound
ha e posed a pe sis en challenge (McDonald e al., 1998;
Gemmell, 2000) because o he high a iabili y in incoming
adiance below he o es canopy, challenges wi h he spec-
al cha ac e iza ion, and weak signal in some pa s o he
spec um o bo h o e s o y and unde s o y (Schaepman e
al., 2009), and he gene al a ying na u e o he unde s o y
(Mille e al., 1997).
Mul i-angle emo e sensing can cap u e signals o di e -
en o es laye s because he obse ed p opo ions o di -
e en o es laye s a y wi h he iewing angle, making i
possible o sepa a e o es o e s o y and unde s o y signal.
He e, we aim o consolida e p e ious e o s o acking un-
de s o y e lec ance and i s dynamics wi h mul i-angle Ea h
obse a ion da a (Canisius and Chen, 2007; Pisek and Chen,
2009; Pisek e al., 2010, 2012, 2015a, 2015b, 2016; Jiao e
al., 2014) by es ing he alidi y o his app oach, using Mod-
e a e Resolu ion Imaging Spec o adiome e bidi ec ional e-
lec ance dis ibu ion unc ion (MODIS BRDF)/albedo da a
(g idded 500m daily Collec ion; 6 MCD43 p oduc ), agains
in si u unde s o y e lec ance measu emen s o e an ex-
ended se o In eg a ed Ca bon Obse a ion Sys em (ICOS)
o es ecosys em si es. The alida ion p ocedu e was de-
ined o comply as much as possible wi h he bes p ac-
ices p oposed by he Commi ee on Ea h Obse a ion Sa el-
li es (CEOS) Wo king G oup on Calib a ion and Valida ion
(WGCV) Land P oduc Valida ion (LPV) subg oup (Ga -
igues e al., 2008; Ba e e al., 2006). I co esponds o
S age 1 alida ion, as de ined by he CEOS (Nigh ingale e
al., 2011; Weiss e al., 2014), whe e p oduc accu acy shall
be assessed o e a small ( ypically <30) se o loca ions and
ime pe iods by compa ison wi h in si u o o he sui able e -
e ence da a. Using he ex ended se o ICOS o es ecosys-
ems as alida ion si es, we asked he ollowing ques ions:
1. Can ICOS o es ecosys em si es se e as a sui able al-
ida ion da a se wi h espec o hei oo p in and he
pixel esolu ion o Ea h obse a ion (EO) p oduc s?
Biogeosciences, 18, 621–635, 2021 h ps://doi.o g/10.5194/bg-18-621-2021
J. Pisek e al.: Re ie al and alida ion o o es backg ound e lec i i y 623
Figu e 1. Dis ibu ion o s udy si es ac oss Eu ope; o u he de-
ails, e e o Table 1.
2. Can we e ie e eliable no malized di e ence ege a-
ion index (NDVI; Rouse e al., 1973; Tucke , 1979) dy-
namics o unde s o y wi h MODIS BRDF da a ac oss
di e se o es condi ions du ing he g owing season?
3. A e he e impo an di e ences be ween he o al (o e -
s o y and unde s o y) and unde s o y-only NDVI sig-
nals?
2 Ma e ials and me hods
2.1 S udy si es
The ICOS is a dis ibu ed pan-Eu opean esea ch in as-
uc u e p o iding in si u s anda dized, in eg a ed, long- e m
and high-p ecision obse a ions o lowe a mosphe e g een-
house gas (GHG) concen a ions and land–a mosphe e and
ocean–a mosphe e GHG in e ac ions (Gielen e al., 2017).
In his s udy, we ca ied ou he e alua ion o e he ne wo k
o 31 ICOS-a ilia ed o es ecosys em si es, complemen ed
wi h addi ional si es in Spain, Po ugal, Aus ia, and Finland.
Toge he , hese selec ed 40 s udy si es comp ise a la ge a i-
e y o o es o e - and unde s o y ypes, spanning a wide la -
i udinal g adien om almos 38◦N (Yes e, Spain) o 68◦N
(Ken ä o a, Finland). Si e loca ions a e shown in Fig. 1, and
ege a ion cha ac e is ics a e summa ized in Table 1.
2.2 Unde s o y spec a and o es canopy co e /closu e
in si u measu emen s
Following he e minology by Schaepman-S ub e
al. (2006), we e e o he e lec ance ac o s mea-
su ed by he ield spec ome e s as he sa elli e-de i ed
hemisphe ical–di ec ional e lec ance ac o s (HDRFs).
The gi en spec ome e ’s ield o iew is app oxima ed
as being angula (cone) and na owe han a whole hemi-
sphe e, wi h some aniso opy cap u ed which co esponds
o no mal emo e sensing iewing geome y. An o e iew
o he unde aken in si u campaigns a each si e and hei
cha ac e is ics a e gi en in Table 1.
The indi idual si es we e isi ed be ween Ap il 2016 and
Augus 2019, mos ly du ing he g owing season. Following
he p o ocol by Rau iainen e al. (2011), he unde s o y spec-
a we e measu ed wi h he Sun comple ely obscu ed by he
clouds o a a ound sunse (di use ligh condi ions), co e -
ing he isible/nea in a ed (NIR) egion, depending on he
spec ome e (see Table 1 o mo e de ails). A o al o h ee
unde s o y spec a we e measu ed e e y 2m along wo 50m
long ansec s laid a each si e, esul ing in 50 measu emen
poin s (150 indi idual measu emen s). T ansec s co e ed and
cha ac e ized condi ions wi hin he measu emen oo p in
o he gi en owe . I should be no ed ha he owe oo -
p in migh be di e en om he exac MODIS pixel oo -
p in (see Sec . 2.5 o he spa ial homogenei y assessmen
o MODIS pixels). The measu emen s conce ned condi ions
on he o es loo and low he baceous and sh ubby species o
ee seedlings and saplings, as he a ea sampled by each spec-
al measu emen was es ima ed o co espond o a ∼50cm
diame e ci cle on he g ound. The downwa d-poin ing spec-
o adiome e (no o e op ics we e used) was held by he op-
e a o ’s ou s e ched hand. A o al o h ee spec a abo e a
10in. (0.254m) Spec alon SRT-99–100 whi e panel we e
eco ded a he beginning, a e e e y ou unde s o y spec-
a measu emen poin s (e e y 8m), and a end o each an-
sec . A hemisphe ical, conical e lec ance ac o was ob-
ained wi h an uncalib a ed Spec alon e lec ance spec um
and he linea ly in e pola ed i adiance. Finally, b oadband
HDRFs o ed (620–670nm) and NIR (841–876nm) wa e-
leng hs we e compu ed wi h ela i e spec al esponse unc-
ions o he MODIS senso onboa d Te a. The unde s o y
NDVI alue o gi en si e was calcula ed om he ed and
NIR-band alues and a e aged o e he wo ansec s.
Es ima es o o e s o y oliage co e and c own co e we e
ob ained om digi al co e pho og aphs (DCPs). O e s o y
oliage co e was de ined as he pe cen age o g ound co -
e ed by he e ical p ojec ion o oliage and b anches and
c own co e as he pe cen age o g ound co e ed by he e -
ical p ojec ions o he ou e mos pe ime e s o he c owns
on he ho izon al plane (wi hou double-coun ing he o e -
lap; Gschwan ne e al., 2009). The DCPs we e aken om
below he canopy e e y 8m along ansec s a each si e. The
came a (Nikon CoolPix4500; 2272×1704 esolu ion) was
h ps://doi.o g/10.5194/bg-18-621-2021 Biogeosciences, 18, 621–635, 2021
624 J. Pisek e al.: Re ie al and alida ion o o es backg ound e lec i i y
Table 1. S udy si e cha ac e is ics and hei spa ial ep esen a i eness s a us. ICOS – In eg a ed Ca bon Obse ing Sys em si es; LTER – Long Te m Ecological Resea ch Ne wo k si es.
No e ha he sampling da es a e shown in he o ma yyyy/mm/dd.
Si e code Si e name La (◦) Long (◦) Sampling da e Spec ome e model Unde s o y ege a ion Rep esen a i eness
AT-Zbn Zöbelboden (LTER) 47.842 14.442 2017/11/18 ASD FieldSpec 4 Calamag os is a ia, B achypodium
syl a icum, Ho delymus eu opaeus,
and Senecio o a us
No ep esen a i e
BE-B a B asschaa (ICOS) 51.304 4.519 2019/01/12 Ocean Op ics; FLAME-S-VIS-NIR-ES Be ula spec., Que cus obu , and
So bus aucupa ia
Rep esen a i e
BE-Vie Vielsalm (ICOS) 50.3 5.983 2018/08/16 Ocean Op ics; FLAME-S-VIS-NIR-ES Spa se e n and moss co e Rep esen a i e a 0.5km
CH-Da Da os (ICOS) 46.817 9.85 2018/07/12 Ocean Op ics; FLAME-S-VIS-NIR-ES Dwa sh ubs, bluebe y, and mosses Rep esen a i e
CZ-BK1 Bílý Kˇ
íž (ICOS) 49.502 18.539 2016/04/17 ASD FieldSpec 4 Vaccinium my illus L. Rep esen a i e a 1.5km
CZ-Lnz Lanžho (ICOS) 48.682 16.948 2017/04/27 ASD FieldSpec 4 Allium u sinum and Asa um eu opeum Rep esen a i e
DE-Hai Hainich (ICOS) 51.079 10.453 2018/04/12 Ocean Op ics; FLAME-S-VIS-NIR-ES Anemone nemo osa and Allium u sinum Rep esen a i e
DE-HoH Hohes Holz (ICOS) 52.083 11.217 2018/04/11 Ocean Op ics; FLAME-S-VIS-NIR-ES Anemone nemo osa Rep esen a i e
DE-RuW Wüs ebach (ICOS) 50.505 6.331 2018/08/16 Ocean Op ics; FLAME-S-VIS-NIR-ES Spa se Deschampsia lexuosa,
Deschampsia cespi osa and Molinia
cae ulea
No ep esen a i e
DE-Tha Tha and (ICOS) 50.967 13.567 2018/04/12 Ocean Op ics; FLAME-S-VIS-NIR-ES Fagus syl a ica,Abies alba, and De-
schampsia lexuosa
Rep esen a i e
DK-So So oe (ICOS) 55.486 11.645 2018/09/26 Ocean Op ics; FLAME-S-VIS-NIR-ES Beech saplings and seedlings; P e id-
ium aquilinum
Rep esen a i e
ES-AP1 Almodó a del Pina 39.677 −1.848 2017/11/09 ASD FieldSpec HandHeld 2 Que cus ilex ssp. ballo a,Rosma i-
nus o icinalis, Thymus ulga is, La an-
dula la i olia, Que cus cocci e a, and
Genis a sco pius
Rep esen a i e
ES-CMu Cuenca del Majadas 40.252 −1.965 2017/11/12 ASD FieldSpec HandHeld 2 Junipe us communis,Junipe us
oxyced us, and C a aegus monogyna
Rep esen a i e a 0.5km
ES-CPa Co es de Pallas 39.224 −0.903 2017/11/08 Ocean Op ics; FLAME-S-VIS-NIR-ES Rosma inus o icinalis,Ulex
pa i lo us, and B achypodium e usum
Rep esen a i e >0.5km
ES-Ys Yes e 38.339 −2.351 2018/07/28 Ocean Op ics; FLAME-S-VIS-NIR-ES Rosma inus o icinalis L.,Thymus ul-
ga is L., and Cis us clusii Dunal
Rep esen a i e a 0.5km
FI-Hal Halssiaapa 67.368 26.654 2017/06/13 ASD FieldSpec P o Sedge ege a ion Rep esen a i e
FI-Hyy Hyy iälä (ICOS) 61.847 24.295 2018/06/28 Ocean Op ics; FLAME-S-VIS-NIR-ES Vaccinium spec. and No way sp uce
seedlings
Rep esen a i e a 0.5km
Biogeosciences, 18, 621–635, 2021 h ps://doi.o g/10.5194/bg-18-621-2021
J. Pisek e al.: Re ie al and alida ion o o es backg ound e lec i i y 625
Table 1. Con inued.
Si e Code Si e Name La (◦) Long (◦) Sampling da e Spec ome e model Unde s o y ege a ion Rep esen a i eness
FI-Ken Ken ä o a (ICOS) 67.987 24.243 2017/06/13 ASD FieldSpec P o Vaccinium my illus,Empe um
nig um, and Vaccinium i is-idaea and
he o es mosses Pleu ozium sch ebe i,
Hylocomium splendens, and Dic anum
polyse um
Rep esen a i e a 2km
FI-Kns Kale ansuo 60.647 24.356 2017/06/15 ASD FieldSpec P o Dwa sh ubs and mosses Rep esen a i e a 0.275km
FI-Le Le osuo (ICOS) 60.642 23.96 2017/06/15 ASD FieldSpec P o Dwa sh ubs, mosses, and he bs Rep esen a i e <1.0km
FI-Sod Sodankylä (ICOS) 67.362 26.638 2017/06/13 ASD FieldSpec P o Lingonbe y, Calluna ulga is, and
lichens
Sphe oid does no i <1.5km;
no ep esen a i e a >1.5km
FI-Va Vä iö (ICOS) 67.757 29.616 2017/06/14 ASD FieldSpec P o Mosses, lichens, and dwa sh ubs Rep esen a i e
FR-Bil Bilos – Salles (ICOS) 44.494 −0.956 2018/06/14 Ocean Op ics; FLAME-S-VIS-NIR-ES Molinia coe ulea Moench.,P e idium
aquilineum, and Ulex eu opaeus
Rep esen a i e <0.5km
FR-FBn Fon Blanche (ICOS) 43.241 5.679 2018/06/12 Ocean Op ics; FLAME-S-VIS-NIR-ES Que cus cocci e a,Philly ea la i olia,
and o he species
Rep esen a i e a 1.5km
FR-Fon Fon ainebleau–Ba beau (ICOS) 48.476 2.780 2018/06/16 Ocean Op ics; FLAME-S-VIS-NIR-ES Ca pinus be ulus Rep esen a i e a 0.5km
FR-Hes Hesse (ICOS) 48.674 7.066 2018/08/18 Ocean Op ics; FLAME-S-VIS-NIR-ES Fagus syl a ica seedlings and black-
be y
Sphe oid does no i
FR-MsS Mon ie s (ICOS) 48.537 5.312 2019/01/14 Ocean Op ics; FLAME-S-VIS-NIR-ES Spa se Sphagnum spec. ege a ion Rep esen a i e >1.0km
FR-Pue Puéchabon (ICOS) 43.741 3.596 2018/06/13 Ocean Op ics; FLAME-S-VIS-NIR-ES Buxus sempe i ens,Pis acia len iscus,
Philly ea la i olia,Sal ia osma inus,
and Ruscus aculea us
Rep esen a i e
IT-BF Bosco Fon ana (ICOS) 45.202 10.743 2018/07/10 Ocean Op ics; FLAME-S-VIS-NIR-ES Hede a helix,Co ylus spec., and Rus-
cus aculea us
Rep esen a i e a 1.5km
IT-Cp2 Cas elpo ziano 2 (ICOS) 41.704 12.357 2019/01/25 Ocean Op ics; FLAME-S-VIS-NIR-ES Phylli ea la i olia and Pis acia len is-
cus
Rep esen a i e a 1.5km
IT-Ren Renon (ICOS) 43.732 10.291 2018/07/11 Ocean Op ics; FLAME-S-VIS-NIR-ES Deschampsia lexuosa L.,
Vaccinium my illus L., and Rhododen-
d on e ugineum L.
Rep esen a i e a 0.5km
IT-SR2 San Rosso e (ICOS) 61.847 24.295 2018/06/28 Ocean Op ics; FLAME-S-VIS-NIR-ES Ligus um ulga e Rep esen a i e <1.5km
IT-T To gnon 45.833 7.567 2018/07/07 Ocean Op ics; FLAME-S-VIS-NIR-ES Junipe us communis,Rhododend on
e ugineum, and Fes uca a ia
No ep esen a i e
h ps://doi.o g/10.5194/bg-18-621-2021 Biogeosciences, 18, 621–635, 2021

626 J. Pisek e al.: Re ie al and alida ion o o es backg ound e lec i i y
Table 1. Con inued.
Si e Code Si e Name La (◦) Long (◦) Sampling da e Spec ome e model Unde s o y ege a ion Rep esen a i eness
NL-Loo Loobos (ICOS) 52.167 5.744 2018/08/13 Ocean Op ics; FLAME-S-VIS-NIR-ES P unus se o ina,Vaccinium my illus,
Deschampsia lexuosa, and mosses
Rep esen a i e a 0.5km
NO-Hu Hu dal (ICOS) 60.372 11.078 2018/09/27 Ocean Op ics; FLAME-S-VIS-NIR-ES Vaccinium spec. and No way sp uce
seedlings
Rep esen a i e
PT-Co Co uche (LTER) 39.138 -8.333 2016/10/08 Ocean Op ics; FLAME-S-VIS-NIR-ES Rumex ace osella,Tube a ia gu a a,
Tolpis ba ba a Plan ago co onopus,
Ag os is pou e ii,B iza maxima,
Vulpia b omoides, and Vulpia genicu-
la a
Sphe oid does no i
SE-H m Hyl emossa (ICOS) 56.098 13.419 2018/09/28 Ocean Op ics; FLAME-S-VIS-NIR-ES Con inuous moss co e Sphe oid does no i
SE-Knd Kindla (LTER) 59.754 14.908 2016/07/16 ASD FieldSpec P o E icaceous dwa sh ubs, mosses,
and lichens
Rep esen a i e
SE-No No unda (ICOS) 60.086 17.48 2018/10/22 ASD FieldSpec P o Bilbe y, lingonbe y, and moss Rep esen a i e <1.5km
SE-S b S a be ge (ICOS) 64.256 19.775 2019/08/23 ASD FieldSpec P o Bilbe y, lingonbe y, and moss Rep esen a i e
se o au oma ic exposu e, ape u e-p io i y mode, minimum
ape u e, and F2 lens (Mac a lane e al., 2007). The came a
was le eled a he heigh o 1.4m abo e he g ound, and he
lens was poin ed owa ds he zeni h. This se up p o ides a
iew zeni h angle om 0 o 15◦, which is compa able wi h
he i s ing o he LAI-2000 ins umen (Mac a lane e al.,
2007).
We used he algo i hm by Nobis and Hunzike (2005) o
h eshold he majo i y o he DCP images. Howe e , some
o he images we e isibly o e exposed, i.e., he 8bi digi al
numbe s (DNs) o he backg ound sky we e 255, and pa s o
any po ion o he sky we e black ( ypically a 240–250 DN).
Nex , a me hod based on ma hema ical image mo phology
(Ko honen and Heikkinen, 2009) was applied o es ima e he
oliage and c own co e ac ions. In his me hod, black and
whi e canopy images a e p ocessed wi h mo phological clos-
ing and opening ope a ions ha a e well known in digi al im-
age p ocessing (Gonzalez and Woods, 2002). As a esul , a
il e o la ge gaps was ob ained. When a uning pa ame-
e (called he s uc u ing elemen in image p ocessing) was
se so ha la ge gaps only occu ed be ween indi idual ee
c owns (Ko honen and Heikkinen, 2009), he p opo ions o
gaps inside and be ween indi idual c owns could be calcu-
la ed.
2.3 Backg ound signal e ie al me hod wi h EO da a
The o al e lec ance o a pixel (R) esul s om he weigh ed
linea combina ion o e lec ance alues by he o es canopy,
o es backg ound, and hei sunli and shaded componen s
(Li and S ahle , 1985; Chen e al., 2000; Bacou and B éon,
2005; Chopping e al., 2008; Roujean e al., 1992) as ollows:
R=kTRT+kGRG+kZT RZT +kZGRZG,(1)
which includes he e lec i i ies o he sunli c owns (RT),
sunli unde s o y (RG), shaded c owns (RZT ), and shaded
unde s o y (RZG). RGma ks he bidi ec ional e lec ance
ac o (BRF) o he a ge (unde s o y). The kja e he p o-
po ions o hese componen s a he chosen iewing angle o
in he ins an aneous ield o iew o he senso a he gi en i -
adia ion geome y. Following Canisius and Chen (2007), we
de i e he unde s o y e lec i i y (RG) wi h he assump ion
ha he e lec i i ies o he o e s o y and unde s o y a he
gi en illumina ion geome y di e li le be ween he chosen
iewing angles. While he componen s may no ully mee
he de ini ion o Lambe ian e lec o s (i.e., e lec ing elec-
omagne ic adia ion equally in all di ec ions), se e al p e-
ious s udies (e.g., Bacou and B éon, 2005; Dee ing e al.,
1999; Pel oniemi e al., 2005) ound o wa d-sca e ing e-
lec ance ac o s o a ious a ge s o he p incipal plane o
be ai ly cons an . The mos sui able iewing con igu a ion
o he e ie al has been iden i ied by Pisek e al. (2015a),
using a high angula esolu ion BRF da a se o Kuusk e
al. (2014) and accompanying in si u measu emen s o unde -
s o y e lec ance ac o s (Kuusk e al., 2013). The con igu-
Biogeosciences, 18, 621–635, 2021 h ps://doi.o g/10.5194/bg-18-621-2021
J. Pisek e al.: Re ie al and alida ion o o es backg ound e lec i i y 627
a ion consis s o he BRF a nadi (Rn=0◦), wi h he sola
zeni h angle (SZA) co esponding o he Sun’s posi ion a
10:00 local ime (LT) o gi en day and ano he zeni h angle
(Ra=40◦) wi h ela i e azimu h angle PHI=130◦. I can be
exp essed by Eqs. (2) and (3) as ollows:
Rn=kT nRT+kGnRG+kZT nRZT +kZGnRZG (2)
Ra=kT aRT+kGaRG+kZT aRZT +kZGaRZG.(3)
The p opo ions o he componen s (kj) we e ob ained using
he ou -scale model (Chen and Leblanc, 1997) wi h pa am-
e e s o gene alized deciduous and coni e ous ee s ands as
an inpu (see Table 2; Kuusk e al., 2013). The unde s o y
e lec ance a he desi ed wa eleng hs can be calcula ed by
combining and sol ing Eqs. (2) and (3) and he inse ion o
Rnand Raes ima es de i ed om app op ia e EO da a. The
indi idual componen s (sunli /shaded o e s o y and unde -
s o y) canno be esol ed wi h he MODIS spa ial esolu ion.
The e lec ances o shaded ee c owns (RZT ) and unde s o y
(RZG) a e ela ed o sunli ones ia Mas RZT =M×RTand
RZG =M×RG, whe e M=RZ/R o a e e ence a ge ,
which can be measu ed in he ield o p ede e mined wi h he
ou -scale model. He e, he same Mis assumed o o e s o y
ees and he unde s o y. Based on his ield wo k in Canadian
bo eal o es s, Whi e (1999) sugges ed ha angula ly con-
s an , wa eleng h-dependen M alues may be app op ia e,
a leas du ing he g owing season. The inpu s and pa ame-
e s om Table 2 may no be always p ecisely known while
e ie ing he unde s o y signal o e la ge a eas. Figu e 2
shows he ela ionships be ween he a ailable in si u da a o
ee heigh s o ee densi ies o e ou s udy si es wi h he
1km2 esolu ion es ima es om he global maps o Sima d e
al. (2011) and C ow he e al. (2015). The weak ela ionships
indica e he cu en unsui abili y o he si e-speci ic a iable
es ima es o in e es ( ee heigh and ee densi y) om cu -
en ly a ailable global maps a a gi en spa ial esolu ion o
ou pu pose. A he same ime, he calcula ed mean alues
o he ee heigh s o needlelea ed (17.5m) and b oadlea ed
ee s ands (22.7m) om Sima d e al. (2011) o e he s udy
si es we e easonably close o ou o iginal gene alized inpu
pa ame e alues in Table 2. Following Gemmell (2000), we
op ed o epo a ange o unde s o y NDVI (NDVIu) al-
ues ob ained wi h he combina ion o pa ame e alues om
Table 2 o each si e and da e. Speci ying he co ec con-
s ain s (window) o backg ound alone has been p e iously
ound o g ea ly educe he e o s in he es ima ion o o e -
s o y pa ame e s (Gemmell, 2000).
2.4 MODIS BRDF da a
The MCD43A1 V6 bidi ec ional e lec ance dis ibu ion
unc ion and albedo (BRDF/albedo) model pa ame e da a
se is a 500m g idded daily p oduc . MCD43A1 is gene -
a ed by in e ing mul i-da e, mul i-angula , cloud- ee, a mo-
sphe ically co ec ed, and su ace e lec ance obse a ions
acqui ed by MODIS ins umen s onboa d he Te a and Aqua
Figu e 2. (a) Rela ionship be ween a ailable in si u es ima es o
ee heigh (in me e s) wi h Sima d e al.’s (2011) es ima e. (b) Re-
la ionship be ween a ailable in si u es ima es o ee densi y ( ees
pe hec a e) wi h C ow he e al.’s (2015) es ima es. DBF – decidu-
ous b oadlea o es ; EBF – e e g een b oadlea o es ; DNF – de-
ciduous needlelea o es ; ENF – e e g een needlelea o es ; MF –
mixed o es .
sa elli es o e a 16d pe iod (Wang e al., 2018). The Julian
da e ep esen s he nin h day o he 16d e ie al pe iod, and
consequen ly, he obse a ions a e u he weigh ed o es i-
ma e he BRDF/albedo o ha pa icula day o in e es . The
MCD43A1 algo i hm uses all high-quali y obse a ions ha
adequa ely sample he iewing hemisphe e o i an app o-
p ia e semiempi ical BRDF model ( he RossThickLiSpa se-
Recip ocal model; Roujean e al., 1992; Luch e al., 2000)
o ha loca ion and da e o in e es . We compu ed he bidi-
ec ional e lec ance ac o (BRF) a he op o he canopy
wi h he iso opic pa ame e and wo ( olume ic and geo-
me ic) ke nel unc ions (Roujean e al., 1992) o MODIS
band 1 ( ed – 620–670nm) and band 2 (NIR – 841–876 nm).
We used he Ross and Li ke nels o econs uc he BRF al-
ues o equi ed geome ies (see Sec . 2.3) o each da e, and
hen we de i ed he unde s o y signal, using he o mulas de-
sc ibed in Sec . 2.3. The associa ed da a quali y (MCD43A2)
p oduc was employed o assess he e ec o he e ie al
quali y on he accu acy o he calcula ed unde s o y signal.
All MODIS da a ha e been accessed and p ocessed h ough
he Google Ea h Engine (Go elick e al., 2017).
2.5 Spa ial ep esen a i eness assessmen o he
alida ion si es
A me hod de eloped by Román e al. (2009), and e ined by
Wang e al. (2012, 2014, 2017) was adop ed o e alua e he
spa ial ep esen a i eness o in si u measu emen s o assess
he unce ain ies a ising om a di ec compa ison be ween
ield-measu ed o es unde s o y spec a and he co espond-
ing es ima es wi h MODIS BRDF da a. To cha ac e ize he
spa ial ep esen a i eness o a es si e o ep esen a sa el-
li e e ie al, his me hod uses h ee a iog am model pa-
ame e s ( he ange, sill, and nugge ), ob ained by he anal-
ysis o nea -nadi su ace e lec ances om cloud- ee 30m
Landsa /Ope a ional Land Image (OLI) da a (Román e al.,
2009) collec ed as close o he sampling da e as possible.
h ps://doi.o g/10.5194/bg-18-621-2021 Biogeosciences, 18, 621–635, 2021
628 J. Pisek e al.: Re ie al and alida ion o o es backg ound e lec i i y
Table 2. S and pa ame e s o he ou -scale model.
S and Deciduous Coni e ous
S and densi y ( eesha−1) 500, 1000, and 2000 500, 1000, and 2000
T ee heigh (m) 25 16
Leng h o li e c own (m) 9.2 4.2
Radius o c own p ojec ion (m) 1.87 1.5
Lea a ea index (m2m−2) 1, 2, and 3 1, 2, and 3
Figu e 3. Sho wa e BRF composi es cen e ed a ICOS si es o (a) No unda in Sweden and (c) Wüs ebach in Ge many. (b, d) Va iog am
es ima o s (poin s), sphe ical model esul s (do ed cu es), and sample a iances (solid s aigh lines) ob ained o e he si es wi h Ope a ional
Land Image (OLI) subse s and spa ial elemen s o 0.275, 0.5, 1.0, 1.5, and 2km as a unc ion o he dis ance be ween obse a ions. Va iog am
legend explana ions: a – a iog am ange; a – sample a iance; c – a iog am sill; c0 – nugge a iance.
Whe e alid image y was no a ailable wi hin a easonable
window o he sampling da e, image y om he co espond-
ing season o a di e en yea was used. As such, he analy-
sis was done o illus a e he ep esen a i eness o he owe
si e wi h espec o a pa icula poin in ime. Campagnolo
e al. (2016) showed ha he e ec i e spa ial esolu ion o
500m g idded MODIS BRDF p oduc a mid-la i udes is
a ound 833×618m because o he a ied oo p in s o he
sou ce mul i-angula su ace e lec ance obse a ions. We
analyzed each si e wi h i e di e en spa ial ex en s (0.275,
0.5, 1, 1.5, and 2km) o assess and illus a e he changes in
spa ial ep esen a i eness wi h di e en spa ial esolu ions.
3 Resul s and discussion
3.1 Spa ial ep esen a i eness
Table 1 p o ides he assessmen o spa ial he e ogenei y o
all si es included in his s udy, using OLI subse s acqui ed
a ound he ime o in si u measu emen s. The example e-
sul s o he ICOS si es o No unda (SE-No ) in Sweden and
Biogeosciences, 18, 621–635, 2021 h ps://doi.o g/10.5194/bg-18-621-2021
J. Pisek e al.: Re ie al and alida ion o o es backg ound e lec i i y 629
Wüs ebach (DE-RuW) in Ge many, using h ee OLI subse s,
a e shown in Fig. 3. The a iog am unc ions wi h ele an
model pa ame e s o he wo si es a e displayed in Fig. 3b
and d. The ange co esponds o he alue on he xaxis whe e
he model la ens ou . The e is no u he co ela ion o a
biophysical p ope y associa ed wi h ha poin beyond he
ange alue. The sill is he o dina e alue o he ange. A
smalle sill alue indica es a mo e homogenous su ace (less
a ia ion in su ace e lec ance). A su ace can be conside ed
spa ially ep esen a i e wi h espec o he MODIS oo p in
when he sill alue is <5.0e−4(Román e al., 2009; Wang e
al., 2017). The sill alues o all spa ial ex en s a e well below
he alue o 5.0e−4, up o 1km spa ial esolu ion in he case
o No unda (Fig. 3b), which indica es ha he ield measu e-
men s a e ep esen a i e and allow compa ison wi h MODIS
e ie als a a 500m spa ial esolu ion. While he Wüs ebach
si e can be conside ed spa ially homogeneous wi hin he im-
media e icini y o 275m a ound he owe , he sill alue ex-
ceeds he c i e ia o 5.0e−4 a >0.5km spa ial esolu ion.
Du ing la e summe /ea ly au umn o 2013, ees we e almos
comple ely emo ed in an a ea o 9ha wes o he owe in o -
de o p omo e he na u al egene a ion o a nea -na u al de-
ciduous o es om a sp uce monocul u e o es . The clea -
elling a ea can be seen in Fig. 3d. This ac ion esul ed in
an inc ease in he spa ial he e ogenei y o his ICOS si e.
In si u measu emen s collec ed wi hin he oo p in o he
Wüs ebach owe , hus, canno be deemed ully compa able
wi h he e ie als wi h MODIS a a 500m spa ial esolu ion.
O e all, mos o he si es we e ound ep esen a i e a he
spa ial esolu ion o MODIS BRDF g idded da a. The non-
ep esen a i e cases and he e ec on he unde s o y signal
e ie al and ag eemen wi h he co esponding in si u mea-
su emen s ca ied wi hin he measu emen oo p in o he
indi idual owe s a e u he discussed in Sec . 3.2 and 3.3.
Román e al. (2009) p o ide u he de ails on he assessmen
o spa ial ep esen a i eness, using a se o ou geos a is ical
a ibu es de i ed om semi a iog ams.
3.2 NDVI anges
The e is only a weak ela ionship be ween he o al (o e -
s o y and unde s o y) NDVI signal e ie ed wi h MODIS
BRDF da a and co esponding in si u unde s o y NDVI mea-
su emen s (R2=0.19; Fig. 4). To al NDVI alues alone do
no allow one o disen angle he co ec unde s o y signal.
In con as , ou e ie al me hod could ack he unde s o y
signal dynamics o e a b oad NDVI ange (Fig. 5). The p e-
dic ed unde s o y NDVI anges we e beyond he unce ain y
limi s o in si u unde s o y measu emen s (co esponding o
±1 s anda d de ia ion (SD) he e) in less han 15% o cases.
These si es wi h poo e ie als we e ca e ully in es iga ed
o iden i y he issues p ecluding good esul s. Below, we o-
cus on a discussion o esul s whe e he p edic ed and in si u
measu ed NDVI anges o he unde s o y laye did no ag ee.
Figu e 4. Rela ionship be ween o al (o e s o y and unde s o y)
NDVI alues compu ed om nadi NDVI alues, using MODIS
BRDF/albedo da a and in si u measu ed unde s o y NDVI alues
o e he s udy si es.
The unde s o y domina ed he o e all signal o open
sh ubland a he Co es de Pallas (ES-CPa) si e and he decid-
uous b oadlea o es si e a Mon ie s (FR-MsS) du ing he
lea -o pa o he season (Fig. 5). Bo h si es we e ound o
be spa ially ep esen a i e o compa ison wi h MODIS oo -
p in da a a he ime o he a ailable in si u measu emen s
(Table 1). The e a e only e y ew ees sca e ed ac oss he
Co es de Pallas si e, and g ound ege a ion is ully exposed.
Ex emely low ee densi y does no ma ch wi h any o he
o iginal gene alized inpu pa ame e alues in Table 2, and
he p edic ed unde s o y signal does no ma ch well wi h he
in si u measu emen s. In si u measu emen s a Mon ie s we e
ca ied ou du ing he lea -o pa o he season, which al-
lowed a ull exposu e o he unde s o y. Despi e his, he
p edic ed unde s o y NDVI ange om he MODIS da a did
no o e lap wi h he in si u measu emen s a Mon ie s a
all. Howe e , he MODIS BRDF alues o hese si es we e
ma ked wi h lowe da a quali y lags (QA >1), which co -
ec ly signals a dec ease in accu acy in he calcula ions o
he unde s o y e lec ance as well. O e all, ou esul s con-
i m ha , unde condi ions o e y low ee densi y/lea -o
condi ions, he unde s o y signal can be assumed o be iden-
ical o he o al scene NDVI.
The pe o mance o he me hod u ns ou o be limi ed
o e si es wi h a closed canopy, such as Bílý Kˇ
íž (CZ-
BK1), Hesse (FR-Hes), o Vielsalm (BE-Vie; Fig. 5). This
is because he shadowing e ec makes di use sca e ing he
dominan mechanism in such s ands, and he unde s o y ca -
ies only a negligible in luence on he op-o -canopy signal.
Bosco Fon ana (IT-BF ) is ano he b oadlea o es si e wi h
e y high oliage co e (FC=0.91), ye he p edic ed unde -
s o y NDVI ange en i ely o e laps wi h he collec ed in si u
alues. I should be no ed ha , in con as o o he si es wi h
h ps://doi.o g/10.5194/bg-18-621-2021 Biogeosciences, 18, 621–635, 2021