senso s
A icle
MODIS Senso Capabili y o Bu ned A ea
Mapping—Assessmen o Pe o mance and
Imp o emen s P o ided by he La es S anda d
P oduc s in Bo eal Regions
JoséA. Mo eno-Ruiz 1, JoséR. Ga cía-Láza o 1, Manuel A belo 2,* and
Manuel Can ón-Ga bín1
1Depa amen o de In o má ica, Uni e sidad de Alme ía, 04120 Alme ía, Spain; [email p o ec ed] (J.A.M.-R.);
j ga [email p o ec ed] (J.R.G.-L.); [email p o ec ed] (M.C.-G.)
2Depa amen o de Física, Uni e sidad de La Laguna, 38200 San C is óbal de La Laguna, Spain
*Co espondence: [email p o ec ed]
Recei ed: 10 Augus 2020; Accep ed: 20 Sep embe 2020; Published: 22 Sep embe 2020


Abs ac :
This pape p esen s an accu acy assessmen o he main global scale Bu ned A ea (BA)
p oduc s, de i ed om daily images o he Mode a e-Resolu ion Imaging Spec o adiome e (MODIS)
Fi e_CCI 5.1 and MCD64A1 C6, as well as he p e ious e sions o bo h p oduc s (Fi e_CCI 4.1
and MCD45A1 C5). The exe cise was conduc ed on he bo eal egion o Alaska du ing he pe iod
2000–2017. All he BA polygons egis e ed by he Alaska Fi e Se ice we e used as e e ence da a.
Bo h new e sions doubled he annual BA es ima e compa ed o he p e ious e sions (66% o
Fi e_CCI 5.1 e sus 35% o 4.1, and 63% o MCD64A1 C6 e sus 28% o C5), educing he omission
e o (OE) by almos one hal (39% e sus 67% o Fi e_CCI and 48% e sus 74% o MCD) and
sligh ly inc easing he commission e o (CE) (7.5% e sus 7% o Fi e_CCI and 18% e sus 7%
o MCD). The Fi e_CCI 5.1 p oduc (CE =7.5%, OE =39%) p esen ed he bes esul s in e ms o
posi ional accu acy wi h espec o MCD64A1 C6 (CE =18%, OE =48%). These esul s sugges ha
Fi e_CCI 5.1 could be sui able o hose use s who employ BA s anda d p oduc s in geoin o ma ics
analysis echniques o wild i e managemen , especially in Bo eal egions. The Pa e o bounda y
analysis, pe o med on an annual basis, showed ha he e is s ill a po en ial heo e ical capaci y o
imp o e he MODIS senso -based BA algo i hms.
Keywo ds:
emo e sensing: bu ned a ea; wild i e; MODIS; MCD45A1; MCD64A1; i e_CCI;
pa e o bounda y
1. In oduc ion
Wild i es cause de o es a ion and habi a loss, and hey a e esponsible o eleasing a huge amoun
o ae osol pa icles and g eenhouse gases in o he a mosphe e. These emissions a y depending
on he Bu ned A ea (BA) ex ension and on he ype o biomass p esen in he egion whe e he i e
occu s. Fo example, Equa o ial Asia, which is esponsible o only 0.6% o he Global Bu ned A ea
(GBA), gene a es CO
2
and CH
4
emissions o 8% and 23%, espec i ely. Meanwhile, bo eal o es s,
esponsible o 2.5% o GBA, emi 9% o global CO
2
and 15% o CH
4
emissions [
1
]. Annually, be ween
5 and 15 million ha a e bu ned in bo eal o es s, mainly in Sibe ia, Canada and Alaska, and he
p ojec ions o di e en clima e models es ima e om dec eases o inc eases in BA, which, as sugges ed
by Ki zbe ge e al. [
2
], gene a es unce ain y in bo eal egions, whe e global wa ming may c ea e
con a y e ec s.
Senso s 2020,20, 5423; doi:10.3390/s20185423 www.mdpi.com/jou nal/senso s
Senso s 2020,20, 5423 2 o 23
An accu a e BA es ima ion is he e o e essen ial o p edic ing changes in he global clima e sys em,
inc eased g eenhouse gas concen a ions o he changing chemical composi ion o he a mosphe e due
o i e emissions. De ailed spa ial and empo al knowledge o BA is also essen ial in dynamic global
ege a ion models (DGVM) whe e, oge he wi h o he geospa ial da a, many ecological a iables can
be quan i ied and p ojec ed. We should also no o ge he impo ance o BA maps, in combina ion
wi h socioeconomic and me eo ological da a, in signaling which ac o s con ol he ecu ence o i e
and how long hey las a he egional o global le el [3].
Since he ea ly 1970s, senso s onboa d nume ous Ea h obse a ion missions, such as AVHRR
(Ad anced Ve y High Resolu ion Radiome e ), SPOT-VGT (Sa elli e Pou l’Obse a ion de la
Te e-Vege a ion), ATSR (Along T ack Scanning Radiome e ), MODIS (Mode a e Resolu ion Imaging
Spec o adiome e ), o Landsa , ha e made i possible o de i e BA p oduc s on a global and/o
egional scale. Among hese p oduc s, we can highligh GLOBSCAR [
4
], GBA2000 [
5
], GBS (Global
Bu ned Su aces) [
6
], GLOBCARBON [
7
], L3JRC [
8
], GEOLAND2 [
9
], Global Fi e Emission Da abase
(GFED) [
10
], BAECV (Bu ned A ea Essen ial Clima e Va iable) p oduc [
11
] and GIO-GL1 (Cope nicus
Global Land Se ice bu ned a ea p oduc ) based on he Tansey e al. algo i hm [
8
]. While some o he
abo e-men ioned p oduc s emain ope a ional, he main p oduc s cu en ly in use a e Fi e_CCI 5.1
de eloped by he ESA [
12
] and MCD64A1 C6 de eloped by he Uni e si y o Ma yland [
13
]. Bo h
BA de ec ion me hods a e based on e lec ance de i ed om sola e lec i e bands in combina ion
wi h he mal anomaly maps om ac i e i es (ho spo s) o MODIS [
14
]. The MODIS senso has been
ope a ional since 2000. I has 36 spec al bands om 0.45
µ
m o 14.385, a 12-bi adiome ic esolu ion,
and spa ial esolu ions o 250 m, 500 m, and 1 km [15,16].
Va ious in e na ional scien i ic p og ams ha deal wi h global i e assessmen de ined key
objec i es o spa ial and empo al accu acy, and a se o basic ea u es ha BA p oduc s mus ul ill.
The wo k by Mouillo e al. [
3
] p esen ed he summa ized ins uc ions o he In eg a ed Global
Obse ing S a egy (IGOS) [
17
,
18
], Global Te es ial Obse ing Sys em (GTOS), he G oup on Ea h
Obse a ions (GEO) Ca bon s a egy [
19
], he Global Clima e Obse ing Sys em (GCOS) and he NASA
Whi e Pape on Fi e Ea h Sys em Da a Reco ds (Fi e ESDR) [
20
]. On he one hand, long ime se ies
(g ea e han se e al decades) ha a e consis en and empo ally s able a e equi ed o unde s and
he in e ac ion be ween clima e, ege a ion, and i e. A spa ial esolu ion o be ween 250 and 500 m
would be desi able. Wi h ega d o he spa ial accu acy o he p oduc s, al hough some use s s a e ha
BA p oduc s a e accep able when omission and commission e o s a e balanced [
21
–
23
], mos se a
maximum o a ound 20% o bo h CE and OE [
23
–
25
]. Do he Fi e_CCI 5.1 and MCD64A1 C6 p oduc s
mee hose equi emen s? Gi en he echnical limi a ions o he MODIS senso , can BA mapping
esul s be imp o ed by modi ying he algo i hmic s a egy? The answe s o hese ques ions, as well as
quan i ying he accu acy o hese wo p oduc s, u ns ou o be e y aluable in o ma ion o p ope ly
managing wild i es and hei consequences using geoin o ma ics analysis echniques [26].
In his pape , we p esen a de ailed s udy o he empo al and spa ial accu acy o bo h da ase s,
ocusing on he bo eal egion o Alaska. The p e ious e sions o bo h p oduc s (Fi e_CCI 4.1 and
MCD45A1 C5) ha e also been included o analyze he impac o he changes made in he new e sions.
The Alaskan egion was selec ed o wo easons. Fi s , i is one o he ew egions in he wo ld ha has
an o icial da abase wi h de ailed eco ds o he a ea bu ned by all i es since 1940. These da a we e
used as a e e ence se o assess he accu acy o all he p oduc s. Second, he sca s le by bu ned
a eas in a bo eal egion such as Alaska pe sis o longe , hus acili a ing de ec ion and mo e accu a e
mapping [27,28].
The objec i es pu sued in his wo k a e as ollows:
•
To assess he spa io empo al accu acy o each o he annual ime se ies o he bu ned a ea p oduc s
e sus he e e ence da a (AFS) o he 2000–2017 pe iod.
•
Conce ning o empo al accu acy, o calcula e he pe cen ages o he annual bu ned a ea de ec ed
by each p oduc and o analyze he empo al co ela ion wi h he e e ence da a.
Senso s 2020,20, 5423 3 o 23
•
In ela ion o spa ial accu acy, o es ima e he main me ics de i ed om he con usion ma ix
(commission and omission e o s) and de e mine he Pa e o Bounda y (PB) o he na i e spa ial
esolu ions o each p oduc om he e e ence da a o sepa a e he e o s o each p oduc om he
in insic e o s associa ed wi h i s spa ial esolu ion.
•
To in e compa e he spa io empo al pe o mance o he la es e sions o he Fi e_CCI 5.1 and
MCD64A1 C6 p oduc s and o analyze any possible imp o emen s o e p e ious e sions
(i.e., Fi e_CCI 4.1 and MCD45A1 C5.1).
•
To quan i y he con ibu ion o bu ned a ea agmen a ion o he classi ied map e o s, linking he
a ea unde he annual Pa e o bounda y cu e wi h he o al annual e o s o each p oduc o i s
spa ial esolu ion.
2. Ma e ials and Me hods
2.1. S udy Region
The s udy a ea spans a la ge sec ion o Alaska, ex ending 10
◦
in la i ude (60
◦
N–70
◦
N) and 27.5
◦
in longi ude (Figu e 1). This a ea is domina ed by bo eal o es , a complex se o plan communi ies
modula ed mainly by i e, soil ype and d ainage. The bo eal o es o ms a mosaic o ha dwood–coni e
mixed s ands wi h closed canopy in well-d ained a eas, while in hose wi h pe ma os , open sp uce
s ands p edomina e. Bo eal o es makes up 90% o Alaska’s o es s, an a ea o app oxima ely
42 million ha [29].
Senso s 2020, 20, x FOR PEER REVIEW 3 o 24
• To in e compa e he spa io empo al pe o mance o he la es e sions o he Fi e_CCI 5.1 and
MCD64A1 C6 p oduc s and o analyze any possible imp o emen s o e p e ious e sions (i.e.,
Fi e_CCI 4.1 and MCD45A1 C5.1).
• To quan i y he con ibu ion o bu ned a ea agmen a ion o he classi ied map e o s, linking
he a ea unde he annual Pa e o bounda y cu e wi h he o al annual e o s o each p oduc o
i s spa ial esolu ion.
2. Ma e ials and Me hods
2.1. S udy Region
The s udy a ea spans a la ge sec ion o Alaska, ex ending 10° in la i ude (60° N–70° N) and 27.5°
in longi ude (Figu e 1). This a ea is domina ed by bo eal o es , a complex se o plan communi ies
modula ed mainly by i e, soil ype and d ainage. The bo eal o es o ms a mosaic o ha dwood–
coni e mixed s ands wi h closed canopy in well-d ained a eas, while in hose wi h pe ma os , open
sp uce s ands p edomina e. Bo eal o es makes up 90% o Alaska’s o es s, an a ea o app oxima ely
42 million ha [29].
Figu e 1. The s udy egion includes he en i e bo eal o es o Alaska (70° N–60° N, 168.5° W–141° W).
2.2. Re e ence Da a
Polygons delimi ing he a ea bu ned by wild i es in Alaska a e a ailable om he Alaska Fi e
Se ice (AFS, Fo Wainw igh , AK, USA). AFS has compiled a e y comple e and accu a e da abase
since 1940. In addi ion o he geog aphic coo dina es o he i e si e and pe ime e , AFS con ains
in o ma ion such as he name o he i e; s a and ex inc ion da es; es ima ed BA; cause (na u ally
(e.g., ligh ning), human negligence o maliciously) o municipali y o o igin. Fi e pe ime e s ha e
always been delinea ed using he bes a ailable da a sou ce, om adi ional hand-d awing on
opog aphic maps in he ea ly decades, o he in e p e a ion o ecen ine-scale sa elli e images wi h
spa ial esolu ions less han 30 m [30]. This in o ma ion was used as he e e ence da a ( he g ound
u h) o he accu acy assessmen o he MODIS-de i ed BA p oduc s.
Figu e 1.
The s udy egion includes he en i e bo eal o es o Alaska (70
◦
N–60
◦
N, 168.5
◦
W–141
◦
W).
2.2. Re e ence Da a
Polygons delimi ing he a ea bu ned by wild i es in Alaska a e a ailable om he Alaska Fi e
Se ice (AFS, Fo Wainw igh , AK, USA). AFS has compiled a e y comple e and accu a e da abase
since 1940. In addi ion o he geog aphic coo dina es o he i e si e and pe ime e , AFS con ains
in o ma ion such as he name o he i e; s a and ex inc ion da es; es ima ed BA; cause (na u ally
(e.g., ligh ning), human negligence o maliciously) o municipali y o o igin. Fi e pe ime e s ha e
always been delinea ed using he bes a ailable da a sou ce, om adi ional hand-d awing on
Senso s 2020,20, 5423 4 o 23
opog aphic maps in he ea ly decades, o he in e p e a ion o ecen ine-scale sa elli e images wi h
spa ial esolu ions less han 30 m [
30
]. This in o ma ion was used as he e e ence da a ( he g ound
u h) o he accu acy assessmen o he MODIS-de i ed BA p oduc s.
Fo he s udy pe iod, om 2000 o 2017, AFS eco ded 1868 i es [
31
]. The o al BA exceeded
11.6 million ha, wi h an annual a e age o 0.65 million ha, al hough s ong yea -on-yea luc ua ions
we e ound (see Figu e 2): in 12 o he 18 yea s analyzed, a o al BA o 0.5 million ha was no exceeded,
wi h 2001 and 2008 eco ding he lowes le els, 0.09 and 0.04 million ha, espec i ely. On he o he hand,
in 2004 and 2015, he o al BA exceeded 2 million ha, wi h alues o 2.71 and 2.08 million ha, espec i ely.
Senso s 2020, 20, x FOR PEER REVIEW 4 o 24
Fo he s udy pe iod, om 2000 o 2017, AFS eco ded 1868 i es [31]. The o al BA exceeded 11.6
million ha, wi h an annual a e age o 0.65 million ha, al hough s ong yea -on-yea luc ua ions we e
ound (see Figu e 2): in 12 o he 18 yea s analyzed, a o al BA o 0.5 million ha was no exceeded,
wi h 2001 and 2008 eco ding he lowes le els, 0.09 and 0.04 million ha, espec i ely. On he o he
hand, in 2004 and 2015, he o al BA exceeded 2 million ha, wi h alues o 2.71 and 2.08 million ha,
espec i ely.
Figu e 2. Tempo al dis ibu ion o he annual bu ned a ea in Alaska du ing he pe iod 2000–2017 and
dis ibu ion pe yea o he numbe o i es by size (in housands o hec a es), acco ding he Alaska
Fi e Se ice (AFS, Fo Wainw igh , AK, USA). Fi e ca ego ies we e conside ed o he sizes (BA
ex ension in ha) o he i es: e y small (<100 ha), small (≥100 ha and <1000 ha), medium (≥1000 ha
and <10,000 ha), la ge (≥10,000 ha and <100,000 ha) and e y la ge (≥100,000 ha).
Th oughou he pe iod conside ed, small and e y small i es accoun o an a e age o 60.50%
o he o al egis e ed i es, bu hey only ep esen ed 1.90% o he o al bu ned a ea (Figu e 3). In
con as , la ge and e y la ge i es accoun ed o 13.60% o he o al i es and 82.04% o he o al
bu ned a ea. Be ween hem, he 16 i es o mo e han 100,000 ha bu ned 20.07% o he o al bu ned
a ea in he s udy pe iod (Figu e 3).
Figu e 2.
Tempo al dis ibu ion o he annual bu ned a ea in Alaska du ing he pe iod 2000–2017
and dis ibu ion pe yea o he numbe o i es by size (in housands o hec a es), acco ding he
Alaska Fi e Se ice (AFS, Fo Wainw igh , AK, USA). Fi e ca ego ies we e conside ed o he sizes
(BA ex ension in ha) o he i es: e y small (<100 ha), small (
≥
100 ha and <1000 ha), medium (
≥
1000 ha
and <10,000 ha), la ge (≥10,000 ha and <100,000 ha) and e y la ge (≥100,000 ha).
Th oughou he pe iod conside ed, small and e y small i es accoun o an a e age o 60.50% o
he o al egis e ed i es, bu hey only ep esen ed 1.90% o he o al bu ned a ea (Figu e 3). In con as ,
la ge and e y la ge i es accoun ed o 13.60% o he o al i es and 82.04% o he o al bu ned a ea.
Be ween hem, he 16 i es o mo e han 100,000 ha bu ned 20.07% o he o al bu ned a ea in he s udy
pe iod (Figu e 3).
To cons uc he annual e e ence da a maps, all he pe ime e s o i es occu ing du ing he
MODIS e a (2000–2017) we e downloaded om AFS. The annual ec o laye s we e p ojec ed o he
Albe s Conical Equal A ea p ojec ion using he maximum a ea me hod o assign a bu ned/non-bu ned
class label [32]. The inal size o each pixel in he e e ence maps was 50 m ×50 m.
Senso s 2020,20, 5423 5 o 23
Senso s 2020, 20, x FOR PEER REVIEW 5 o 24
Figu e 3. Pe cen age o he o al numbe o eco ded i es by he Alaska Fi e Se ice (AFS) and o al
a ea bu ned acco ding o he i e ca ego ies conside ed o he sizes o he i es in Figu e 2 o he 18
yea s o s udy (2000–2017).
To cons uc he annual e e ence da a maps, all he pe ime e s o i es occu ing du ing he
MODIS e a (2000–2017) we e downloaded om AFS. The annual ec o laye s we e p ojec ed o he
Albe s Conical Equal A ea p ojec ion using he maximum a ea me hod o assign a bu ned/non-
bu ned class label [32]. The inal size o each pixel in he e e ence maps was 50 m × 50 m.
2.3. Bu ned A ea P oduc s
This s udy compa ed he main global bu ned a ea p oduc s using MODIS senso da a: Fi e_CCI
5.1 om he ESA p ojec o he same name, led by he Uni e si y o Alcalá de Hena es, and he o icial
NASA MODIS Di ec B oadcas Mon hly Bu ned A ea P oduc , de eloped by he Uni e si y o
Ma yland. Bo h o hei mos ecen e sions (Fi e_CCI . 5.1 and MCD64A1 C6), along wi h he
p e ious e sions (Fi e_CCI . 4.1 and MCD45A1 C5.1), we e conside ed o analyze possible
pe o mance imp o emen s. Table 1 shows hei main cha ac e is ics. The main conce ns o ESA and
NASA when upda ing hei p oduc s a e o ex end he ime se ies and imp o e he algo i hms o
ob ain he bes alida ion esul s. Besides, NASA as owne and de elope o MODIS senso s
pe iodically ep ocesses he en i e da a a chi e o inco po a e be e calib a ion and imp o ed
ups eam da a in o all MODIS p oduc s.
Table 1. Bu ned a ea p oduc s.
P oduc Time
Span Senso Me hod
Spa ial
Resolu ion
Algo i hm
Re e ence
Fi e_CCI 4.1 2005–
2011
MERIS + Te a
MODIS
Re lec ance +
ho spo s 300 m [33]
Fi e_CCI 5.1 2001–
oday Te a MODIS Re lec ance +
ho spo s 250 m [12]
MCD45A1
C5.1
2000–
2016
Te a/Aqua
MODIS Re lec ance 500 m [34–36]
MCD64A1
C6
2000–
oday
Te a/Aqua
MODIS
Re lec ance +
ho spo s 500 m [13]
To cons uc he annual BA maps o Alaska, he espec i e mon hly composi es o he ou
p oduc s we e downloaded. As wi h he e e ence maps, all hese maps we e e-p ojec ed o he
Albe s Conical Equal A ea, wi h a pixel size o 50 m × 50 m, and inally combined on an annual basis.
Figu e 4 shows he annual maps o each o he p oduc s gene a ed o he yea 2009.
Figu e 3.
Pe cen age o he o al numbe o eco ded i es by he Alaska Fi e Se ice (AFS) and o al
a ea bu ned acco ding o he i e ca ego ies conside ed o he sizes o he i es in Figu e 2 o he
18 yea s o s udy (2000–2017).
2.3. Bu ned A ea P oduc s
This s udy compa ed he main global bu ned a ea p oduc s using MODIS senso da a: Fi e_CCI
5.1 om he ESA p ojec o he same name, led by he Uni e si y o Alcal
á
de Hena es, and he
o icial NASA MODIS Di ec B oadcas Mon hly Bu ned A ea P oduc , de eloped by he Uni e si y
o Ma yland. Bo h o hei mos ecen e sions (Fi e_CCI . 5.1 and MCD64A1 C6), along wi h
he p e ious e sions (Fi e_CCI . 4.1 and MCD45A1 C5.1), we e conside ed o analyze possible
pe o mance imp o emen s. Table 1shows hei main cha ac e is ics. The main conce ns o ESA and
NASA when upda ing hei p oduc s a e o ex end he ime se ies and imp o e he algo i hms o ob ain
he bes alida ion esul s. Besides, NASA as owne and de elope o MODIS senso s pe iodically
ep ocesses he en i e da a a chi e o inco po a e be e calib a ion and imp o ed ups eam da a in o
all MODIS p oduc s.
Table 1. Bu ned a ea p oduc s.
P oduc Time Span Senso Me hod Spa ial
Resolu ion
Algo i hm
Re e ence
Fi e_CCI 4.1 2005–2011 MERIS +Te a MODIS Re lec ance +ho spo s 300 m [33]
Fi e_CCI 5.1 2001– oday Te a MODIS Re lec ance +ho spo s 250 m [12]
MCD45A1 C5.1
2000–2016 Te a/Aqua MODIS Re lec ance 500 m [34–36]
MCD64A1 C6 2000– oday Te a/Aqua MODIS Re lec ance +ho spo s 500 m [13]
To cons uc he annual BA maps o Alaska, he espec i e mon hly composi es o he ou
p oduc s we e downloaded. As wi h he e e ence maps, all hese maps we e e-p ojec ed o he Albe s
Conical Equal A ea, wi h a pixel size o 50 m
×
50 m, and inally combined on an annual basis. Figu e 4
shows he annual maps o each o he p oduc s gene a ed o he yea 2009.
Th ee o he ou p oduc s base hei algo i hmic s a egy on a hyb id app oach. They use he
MODIS ac i e i es p oduc (ho spo s) in combina ion wi h changes in daily su ace e lec ance o
iden i y bu ned pixels. In con as , he ou h p oduc (MCD45A1) only uses daily su ace e lec ance
image y. The ollowing is a b ie desc ip ion o he algo i hms used in each o he p oduc s.

Senso s 2020,20, 5423 6 o 23
Senso s 2020, 20, x FOR PEER REVIEW 6 o 24
Figu e 4. Annual bu ned a ea maps o he Alaskan egion in 2009. Red, bu ned pixels; whi e, s udy
egion; g een, poli ical bo de s and coas lines.
Th ee o he ou p oduc s base hei algo i hmic s a egy on a hyb id app oach. They use he
MODIS ac i e i es p oduc (ho spo s) in combina ion wi h changes in daily su ace e lec ance o
iden i y bu ned pixels. In con as , he ou h p oduc (MCD45A1) only uses daily su ace e lec ance
image y. The ollowing is a b ie desc ip ion o he algo i hms used in each o he p oduc s.
2.3.1. MCD45A1 Collec ion 5.1
This BA p oduc uses he MCD45 algo i hm o iden i y bu ned pixels a a 500 m spa ial
esolu ion. I de ec s changes in he ime se ies o daily bi-di ec ional e lec ance in bands 2 (0.841–
0.876 µm) and 5 (1.23–1.25 µm) o MODIS senso [37]. MCD45A1, which is cu en ly dep eca ed, was
he o icial p oduc un il he elease o MCD64A1 C6.
The algo i hm compa es he obse ed daily e lec ance alues o each pixel wi h he p edic ed
alues using a bi-di ec ional e lec ance model in a 16-day-minimum ime window ha selec s he
candida e da es o bu ning in he o wa d and backwa d di ec ions. I bo h da es ma ch, he pixel is
conside ed bu ned. Tha da e is hen used as a seed o iden i y, h ough a con ex ual i e a i e p ocess,
whe he neighbo ing pixels can be classi ied as bu ned o no . In a inal s ep, unselec ed candida e
pixels a e conside ed bu ned i hey ha e a leas h ee neighbo s bu ned, wi h hei bu n da e being
he a e age o hei neighbo s. The algo i hm inally excludes hose pixels al eady bu ned in p e ious
seasons and yea s [34–36].
Figu e 4.
Annual bu ned a ea maps o he Alaskan egion in 2009. Red, bu ned pixels; whi e, s udy
egion; g een, poli ical bo de s and coas lines.
2.3.1. MCD45A1 Collec ion 5.1
This BA p oduc uses he MCD45 algo i hm o iden i y bu ned pixels a a 500 m spa ial esolu ion.
I de ec s changes in he ime se ies o daily bi-di ec ional e lec ance in bands 2 (0.841–0.876
µ
m) and
5 (1.23–1.25
µ
m) o MODIS senso [
37
]. MCD45A1, which is cu en ly dep eca ed, was he o icial
p oduc un il he elease o MCD64A1 C6.
The algo i hm compa es he obse ed daily e lec ance alues o each pixel wi h he p edic ed
alues using a bi-di ec ional e lec ance model in a 16-day-minimum ime window ha selec s he
candida e da es o bu ning in he o wa d and backwa d di ec ions. I bo h da es ma ch, he pixel is
conside ed bu ned. Tha da e is hen used as a seed o iden i y, h ough a con ex ual i e a i e p ocess,
whe he neighbo ing pixels can be classi ied as bu ned o no . In a inal s ep, unselec ed candida e
pixels a e conside ed bu ned i hey ha e a leas h ee neighbo s bu ned, wi h hei bu n da e being
he a e age o hei neighbo s. The algo i hm inally excludes hose pixels al eady bu ned in p e ious
seasons and yea s [34–36].
2.3.2. MCD64A1 Collec ion 6
This p oduc applies he MCD64 algo i hm o iden i y bu ned pixels using 500 m daily su ace
e lec ance p oduc s o bands 5 (
ρ
5: 1.23–1.25
µ
m) and 7 (
ρ
7: 2.105–2.155
µ
m), along wi h daily ac i e
i e p oduc s (ho spo s), bo h de i ed om he imaging p oduc s o he MODIS senso on boa d he
Te a and Aqua sa elli es [38].
The algo i hm ini ially calcula es he maximum daily changes in he ime se ies o a bu n-sensi i e
ege a ion index VI =(
ρ
5
−ρ
7)/(
ρ
5+
ρ
7) o wo p e- and pos -da e empo al windows. F om hese
da es, i assigns he bu n/unbu n label o he pixel using he MODIS daily ac i e i e p oduc . I hen
ex ac s a se o aining samples and pe o ms an ini ial supe ised classi ica ion o all pixels based on
Senso s 2020,20, 5423 7 o 23
he no malized dis ance measu emen o each pixel o he nea es pixel in he aining se . The inal
classi ica ion is ob ained by using con ex ual in o ma ion (nea es neighbo s) [13].
2.3.3. Fi e_CCI 4.1
The Fi e_CCI 4.1 p oduc , which has now been discon inued, co e s only he pe iod om 2005
o 2011. I iden i ies bu ned pixels a a 300 m spa ial esolu ion using ime se ies o daily su ace
e lec ances om he MERIS senso on boa d he ENVISAT sa elli e, and he ac i e i es p oduc
de i ed om MODIS senso images. The algo i hm ini ially cons uc s mon hly composi es o su ace
e lec ances by selec ing he candida e pixels o be bu ned using he MODIS ho spo da es as c i e ia.
I calcula es cumula i e dis ibu ion unc ions o disc imina e he mos clea ly bu ned pixels by means
o nea -in a ed e lec ance h esholds. Then, i selec s seed pixels in a 5
×
5 pixel window cen e ed on
he ho spo , o g ow he bu ned egions by con ex ual analysis o he neighbo ing pixels. A inal il e
emo es isola ed pixels, bo h bu ned and non-bu ned [33].
2.3.4. Fi e_CCI 5.1
This is he la es e sion o he bu ned a ea p oduc om he Fi e_CCI p ojec . The algo i hm,
simila o i s p edecesso , is based on a wo-s age hyb id app oach. In he i s s age, he seed pixels
a e selec ed, guided by he daily ac i e i es ( he mal anomalies) de i ed om he MODIS senso s on
boa d he Te a and Aqua sa elli es. In he second s age, he g ow h and delimi a ion o he bu ned a ea
is pe o med using he daily su ace e lec ance p oduc s om MODIS senso bands 1 (0.62–0.67
µ
m)
and 2 (0.841–0.876 µm) a a 250 m spa ial esolu ion [12].
2.4. Accu acy Assessmen
To alida e he bu ned a ea p oduc s, an accu acy assessmen was pe o med agains he AFS
e e ence da a o hose pe iods when each o he p oduc s was a ailable (Table 1). The AFS pe ime e s
used in his s udy a e de i ed p ima ily om sa elli e images wi h spa ial esolu ions o less han 30 m,
including he Landsa and Sen inel-2 missions, he la e since 2016. Figu e 5shows a desc ip ion o
he wo k low o his assessmen exe cise ca ied ou on an annual basis. All annual bu ned a ea map
ime se ies we e delimi ed o he s udy egion wi h 50 m
×
50 m pixels in he same p ojec ion as he
e e ence da a. Each pixel he e o e ep esen s an a ea o 0.25 ha. Pixels we e labeled 0 (No Bu ned) o
1 (Bu ned). This simpli ied he calcula ion o he o al annual bu ned a ea (in hec a es) o each p oduc ,
which was ob ained by mul iplying he numbe o pixels labeled as 1 (Bu ned) by 0.25.
To assess he empo al accu acy, he pe cen age o annual bu ned a ea o each p oduc (P
Yea
(%))
was calcula ed wi h espec o he e e ence se (Equa ion (1)).
PYea (%)=Pn
i=1BAPYea (i)
Pn
i=1BARYea (i)×100 (1)
whe e BAP
Yea
(i) is he alue o he pixel i (0: No Bu ned; 1: Bu ned) o he indica ed yea o BA
p oduc ; BAR
Yea
(i) is he alue o he same pixel i o he e e ence da a; and n is he o al numbe o
pixels in each bu ned a ea map.
Fo each annual bu ned a ea pe cen age dis ibu ion, a cen ali y measu e ( o al alue o he
en i e pe iod), calcula ed as he weigh ed a e age o he annual pe cen ages o bu ned a ea (
PBAP
(%))
(Equa ion (2)), was ob ained. Y
0
and Y
E
a e he i s and las yea s o he BA p oduc , espec i ely.
Sca e plo s o he annual pe cen ages o each p oduc we e cons uc ed agains he e e ence da a o
analyze he linea dispe sion a ound he cen al alue and o iden i y he ex eme cases (yea s wi h
es ima e pe cen ages well below o abo e he mean).
PBAP(%)=PYE
Yea =Y0Pn
i=1BAP Yea (i)
PYE
Yea =Y0Pn
i=1BAR Yea (i)×100 (2)
Senso s 2020,20, 5423 8 o 23
Senso s 2020, 20, x FOR PEER REVIEW 8 o 24
Figu e 5. Flowcha ollowed o he accu acy assessmen o s anda d BA p oduc s.
To assess he empo al accu acy, he pe cen age o annual bu ned a ea o each p oduc (PYea (%))
was calcula ed wi h espec o he e e ence se (Equa ion (1)).
P(%)=∑()


∑()

 ×100 (1)
whe e BAPYea (i) is he alue o he pixel i (0: No Bu ned; 1: Bu ned) o he indica ed yea o BA
p oduc ; BARYea (i) is he alue o he same pixel i o he e e ence da a; and n is he o al numbe o
pixels in each bu ned a ea map.
Fo each annual bu ned a ea pe cen age dis ibu ion, a cen ali y measu e ( o al alue o he
en i e pe iod), calcula ed as he weigh ed a e age o he annual pe cen ages o bu ned a ea (P
(%))
(Equa ion (2)), was ob ained. Y0 and YE a e he i s and las yea s o he BA p oduc , espec i ely.
Sca e plo s o he annual pe cen ages o each p oduc we e cons uc ed agains he e e ence da a
o analyze he linea dispe sion a ound he cen al alue and o iden i y he ex eme cases (yea s wi h
es ima e pe cen ages well below o abo e he mean).
P
(%)=∑(∑ ()

 )


∑(∑ ()

 )

×100 (2)
Subsequen ly, a plo was cons uc ed o he annual bu ned a ea empo al dis ibu ion o each
p oduc and ha o he e e ence da a. Co ela ion analysis o each ime se ies was pe o med wi h
he e e ence da a, calcula ing he coe icien o de e mina ion R2 ( he squa e o Pea son’s co ela ion
coe icien ).
The spa ial accu acy assessmen o each annual BA map employed he con usion ma ix
me hod, which is commonly used o alida e hema ic maps [39]. Table 2 shows he con usion ma ix
o a pixel-le el hema ic classi ica ion wi h wo classes (Bu ned and Non-Bu ned). The independen
e e ence in o ma ion (AFS) is loca ed in he ma ix columns, and he bu ned a ea map da a o each
Figu e 5. Flowcha ollowed o he accu acy assessmen o s anda d BA p oduc s.
Subsequen ly, a plo was cons uc ed o he annual bu ned a ea empo al dis ibu ion o each
p oduc and ha o he e e ence da a. Co ela ion analysis o each ime se ies was pe o med
wi h he e e ence da a, calcula ing he coe icien o de e mina ion R
2
( he squa e o Pea son’s
co ela ion coe icien ).
The spa ial accu acy assessmen o each annual BA map employed he con usion ma ix me hod,
which is commonly used o alida e hema ic maps [
39
]. Table 2shows he con usion ma ix o
a pixel-le el hema ic classi ica ion wi h wo classes (Bu ned and Non-Bu ned). The independen
e e ence in o ma ion (AFS) is loca ed in he ma ix columns, and he bu ned a ea map da a o each
p oduc in he ows. The diagonal elemen s a e he co ec ly classi ied da a ( ue Bu ned and ue
Non-Bu ned). The o he cells indica e commission e o s (CE), i.e., pixels classi ied as Bu ned ha
a e no ac ually bu ned ( alse bu ned), o omission e o s (OE), pixels ha a e ac ually bu ned bu
ha ha e been classi ied as Non-Bu ned ( alse non-bu ned) [
40
]. O he commonly used me ics ha
can be de i ed om he con usion ma ix a e O e all Accu acy (OA) ( he pe cen age o co ec ly
classi ied pixels), Sensibili y (S) (o p oduce ’s accu acy) which calcula es he a e o ue bu ned pixels
( he p opo ion o bu ned pixels ha we e co ec ly iden i ied) and Speci ici y (Sp), o he a e o ue
Non-Bu ned pixels ( he p opo ion o p ope ly iden i ied Non-Bu ned pixels) (Table 2).
Senso s 2020,20, 5423 9 o 23
Table 2.
Con usion ma ix o a bina y classi ica ion o bu ned a ea and he main me ics de i ed om
i . OA, o e all accu acy; S, sensibili y o p oduce ’s accu acy; Sp, speci ici y; CE, commission e o ;
OE, omission e o .
Re e ence Da a
Bu ned Non-Bu ned To al
Classi ied Da a
Bu ned n11 n12 n1c
Non-Bu ned n21 n22 n2c
To al n1 n2 n
OA =n11+n22
nS=n11
n1 Sp =n22
n1
CE =n12
n1c OE =n21
n1
The OA and Sp a e me ics ha can c ea e a alse sense o co ec ness in classi ied maps, especially
in his s udy, due o he asymme y be ween he wo classes conside ed ( he Non-Bu ned class is he
majo i y and i s success a e would be e y high). On he o he hand, S is ela ed o he omission e o
(S =1
−
OE), so only commission and omission e o s o he Bu ned class we e conside ed. The wo
e o s a e no compa able since, al hough hey bo h ep esen pe cen ages o he pixels labeled as
Bu ned, in one case hey a e Bu ned conce ning he classi ied map (CE) and in he o he case o he
e e ence map (OE). The e o e, o he o al e o (TE) calcula ion, he weigh ed sum o bo h e o s was
conside ed, aking in o accoun he pe cen age (P) o BA iden i ied by he classi ied map, acco ding o
Equa ion (3).
To al E o (ha)=CE ×BAP +OE ×BAR =(CE ×P+OE)×BAR
To al E o (%)=To al E o (ha)
BAR =CE ×P+OE (3)
The annual dis ibu ion o OE and CE o each bu ned a ea p oduc was calcula ed, as well as
hei a e age alues ( he o als o all yea s conside ed). To iden i y ex eme alues, CE sca e plo s
we e cons uc ed agains he annual OE o each p oduc , and he annual de ia ions om he a e age
alues we e analyzed. Likewise, o analyze he possible ela ionship be ween he spa ial accu acy and
he annual bu ned a ea, sca e plo s o commission e o s and omission e o s we e cons uc ed o
each BA p oduc , agains he annual bu ned e e ence a ea.
To de e mine whe he agmen a ion o he bu ned a eas limi s he spa ial accu acy o he BA
p oduc s, PB we e cons uc ed a 250, 300 and 500 m spa ial esolu ions [
41
]. To ob ain each PB, he AFS
e e ence map was used, wi h a high spa ial esolu ion (50 m
×
50 m) pe pixel. Using a pixel spa ial
agg ega ion p ocess, he e e ence da a was esized o maps o 250, 300 and 500 m, whe e he alue o
each pixel con ained he pe cen age o bu ned a ea. I one o hese mixed pixels is classi ied as Bu ned,
in a s ic bina y classi ica ion, a commission e o equal o a 1-pixel alue is being made. Howe e , i i
is classi ied as Non-Bu ned, an omission e o equal o he pixel alue is p oduced. The inal decision
on how i is classi ied depends on a p pa ame e selec ed in he ange [0, 1], which se s he minimum
h eshold o assigning a mixed pixel o he Bu ned class. Then, o each deg aded map, he CE and
OE pai s, ob ained by a ying his pa ame e be ween 0 and 1 in equidis an s eps, we e calcula ed.
The se o pai s {(CE
i
, OE
i
)} esul ing om his p ocess is he PB (Figu e 6). The PB ep esen s he
lowes possible e o s, ob ained in a s ic bina y classi ica ion. These e o s a e a ibu able exclusi ely
o he agmen a ion o e e enced bu ned a eas ha occu in mixed pixels when he spa ial esolu ion
o he da a dec eases. All poin s on he PB ep esen ideal classi ica ions o a gi en spa ial esolu ion.
The PB u he demons a es ha minimizing he commission and omission e o s simul aneously
is con adic o y: a classi ica ion wi h minimum commission e o s would mean g ea e ailu e o
omission and, con e sely, minimizing he OE would inc ease he CE.
Senso s 2020,20, 5423 16 o 23
beha io we e included, o p o ide a isual compa ison o he PB. One can obse e ha in 2006 and
2008 he dis ance om he annual pai o e o s (EC, EO) o he PB is much la ge han in 2004 and 2015.
I can also be seen ha he PB a e mo e sepa a ed om he Ca esian axes in hose yea s. This indica es
a g ea e agmen a ion o he bu ned a eas when he spa ial esolu ion is deg aded, which con ibu es,
in pa , o he e o s o he co esponding annual map.
Senso s 2020, 20, x FOR PEER REVIEW 16 o 24
3.3. Pa e o Bounda ies
The Pa e o bounda ies we e cons uc ed a 250/300/500 m o all yea s in he s udy pe iod. Figu e
11 shows he Pa e o bounda ies a 250 and 500 m o 2004, 2006, 2008 and 2015, as well as he annual
CE and OE o he MCD64A1 and Fi e_CCI 5.1 p oduc s. Yea s 2006 and 2008 we e selec ed because
bo h ha e he bigges e o s in bo h p oduc s. In addi ion, wo yea s, 2004 and 2015, wi h good
beha io we e included, o p o ide a isual compa ison o he PB. One can obse e ha in 2006 and
2008 he dis ance om he annual pai o e o s (EC, EO) o he PB is much la ge han in 2004 and
2015. I can also be seen ha he PB a e mo e sepa a ed om he Ca esian axes in hose yea s. This
indica es a g ea e agmen a ion o he bu ned a eas when he spa ial esolu ion is deg aded, which
con ibu es, in pa , o he e o s o he co esponding annual map.
Figu e 11. Pa e o bounda ies a 250 and 500 m and annual CE and OE o he MCD64A1 and Fi e_CCI
5.1 p oduc s o 2004, 2006, 2008 and 2015.
To quan i y he impac o he PB on he annual maps o he di e en BA p oduc s, he annual
and o al AUPB alues we e calcula ed a spa ial esolu ions o 250/300/500 m (Table 5). On a e age,
he AUPB alue a 500 m iples he 250 m alue, in addi ion o ha ing g ea e annual a iabili y:
[0.0019–0.0146] e sus [0.0007–0.0048]. The yea 2008 shows he highes AUPB alues o all spa ial
esolu ions.
Figu e 11.
Pa e o bounda ies a 250 and 500 m and annual CE and OE o he MCD64A1 and Fi e_CCI
5.1 p oduc s o 2004, 2006, 2008 and 2015.
To quan i y he impac o he PB on he annual maps o he di e en BA p oduc s, he annual
and o al AUPB alues we e calcula ed a spa ial esolu ions o 250/300/500 m (Table 5). On a e age,
he AUPB alue a 500 m iples he 250 m alue, in addi ion o ha ing g ea e annual a iabili y:
[0.0019–0.0146] e sus [0.0007–0.0048]. The yea 2008 shows he highes AUPB alues o all
spa ial esolu ions.
Finally, Figu e 12 shows he annual TE sca e plo s o each p oduc e sus he annual PB
a ea co esponding o hei spa ial esolu ion. A linea eg ession model was cons uc ed o each
BA p oduc . The R
2
alues a e e y low o he olde e sions o he BA p oduc s (0.00 and 0.18,
espec i ely), bu inc ease ma kedly in he new e sions. Fi e_CCI 5.1 has a sligh ly highe alue han
MCD64A1 C6 (0.45 s. 0.41) and also a linea eg ession line slope ha is almos double (63.1 s. 32.8).

Senso s 2020,20, 5423 17 o 23
Table 5.
A eas (
×
10
−3
) enclosed by he annual Pa e o Bounda ies (AUPB) a di e en esolu ions and
he weigh ed a e age alue o all yea s.
Yea Spa ial Resolu ion
250 m 300 m 500 m
2000 4.885
2001 1.670 4.940
2002 0.824 2.527
2003 1.341 4026
2004 0.656 1.909
2005 1.066 1.333 3.191
2006 2.171 2.810 6.480
2007 2.125 2.767 6.724
2008 4.818 6.249 14.610
2009 0.931 1.149 2.714
2010 2.257 2.891 7.106
2011 2.810 3.680 9.056
2012 2.715 8.642
2013 1.534 4.723
2014 0.778 2.192
2015 1.470 4.356
2016 4.120 12.685
2017 2.733 8.253
All yea s 1.195 1.625 3.612
Senso s 2020, 20, x FOR PEER REVIEW 17 o 24
Table 5. A eas (× 10−3) enclosed by he annual Pa e o Bounda ies (AUPB) a di e en esolu ions and
he weigh ed a e age alue o all yea s.
Yea Spa ial Resolu ion
250 m 300 m 500 m
2000 4.885
2001 1.670 4.940
2002 0.824 2.527
2003 1.341 4026
2004 0.656 1.909
2005 1.066 1.333 3.191
2006 2.171 2.810 6.480
2007 2.125 2.767 6.724
2008 4.818 6.249 14.610
2009 0.931 1.149 2.714
2010 2.257 2.891 7.106
2011 2.810 3.680 9.056
2012 2.715 8.642
2013 1.534 4.723
2014 0.778 2.192
2015 1.470 4.356
2016 4.120 12.685
2017 2.733 8.253
All yea s 1.195 1.625 3.612
Finally, Figu e 12 shows he annual TE sca e plo s o each p oduc e sus he annual PB a ea
co esponding o hei spa ial esolu ion. A linea eg ession model was cons uc ed o each BA
p oduc . The R2 alues a e e y low o he olde e sions o he BA p oduc s (0.00 and 0.18,
espec i ely), bu inc ease ma kedly in he new e sions. Fi e_CCI 5.1 has a sligh ly highe alue han
MCD64A1 C6 (0.45 s. 0.41) and also a linea eg ession line slope ha is almos double (63.1 s. 32.8).
Figu e 12. Sca e plo s o he o al annual TE e o s (weigh ed sum o he annual commission and
omission e o s) e sus he a ea enclosed by he annual Pa e o bounda y. Da a o he yea s 2000 and
2001 we e no included, as he e we e incomple e da a om he MODIS senso .
Figu e 12.
Sca e plo s o he o al annual TE e o s (weigh ed sum o he annual commission and
omission e o s) e sus he a ea enclosed by he annual Pa e o bounda y. Da a o he yea s 2000 and
2001 we e no included, as he e we e incomple e da a om he MODIS senso .
4. Discussion
The e alua ion and alida ion o global BA p oduc s de i ed om sa elli e image y equi e a se
o eliable, independen and ep esen a i e e e ence i e pe ime e s ha co e as long a pe iod as
possible. I is necessa y o unde s and he unce ain y o hese p oduc s be o e inco po a ing hem as
inpu da a in o global ca bon, ege a ion o clima e models, as well as o he managemen o all he
Senso s 2020,20, 5423 18 o 23
wild i e phases. Nume ous p io s udies ha e e alua ed he beha io o he MODIS senso -de i ed
p oduc s analyzed in his s udy, bo h as pa o ESA’s Fi e Clima e Change Ini ia i e P ojec and in he
di e en e sions [
13
,
51
–
58
] o he MODIS Di ec B oadcas Mon hly Bu ned A ea P oduc . Howe e ,
many o hese wo ks cons uc e e ence i e pe ime e s using images om be e spa ial esolu ion
senso s such as Landsa TM/ETM o Sen inel-2, and which a e limi ed o sho ime pe iods, usually
one o se e al yea s, due o he di icul y in c ea ing la ge e e ence se s. This wo k p esen s an
independen in e compa ison exe cise on he accu acy o he wo main global BA p oduc s de i ed om
he MODIS senso co e ing an ex ensi e bo eal egion o e 18 yea s, con aining all he AFS- egis e ed
i e pe ime e s. To he bes o ou knowledge, he e ha e been no p e ious in e compa ison s udies o
hese ou p oduc s o e such a long pe iod.
In ela ion o he empo al accu acy (Figu e 8), and excep o he Fi e_CCI 4.1 p oduc , which has
he sho es ime se ies, he BA p oduc s analyzed con o m o he ime pa e n o he e e ence da a
wi h de e mina ion coe icien s abo e 0.97, and wi h esul s simila o hose ound by o he au ho s in
di e en ecosys ems. Thus, o example, Tu co e al. [
56
] epo ed high de e mina ion coe icien s o
0.96 and 0.97 in he mon hly BA es ima es o Fi e_CCI 5.1 and MCD64A1 C6 when compa ed o he
e e ence da ase om he Eu opean Fo es Fi e In o ma ion Sys em (EFFIS) ha includes bu ned a ea
da a o some Eu opean coun ies in he Medi e anean basin (Po ugal, Spain, Sou he n F ance and
G eece). Howe e , all he BA p oduc s analyzed unde es ima ed he annual bu ned a ea (Table 3) wi h
highe pe cen ages han hose ob ained by o he au ho s o egions o he han Alaska’s bo eal o es .
Tu co e al. [
56
] ob ained an unde es ima ion o only 14% o he Fi e_CCI 5.1 p oduc wi h he EFFIS
e e ence se . Campagnolo e al. [
57
] ound a 28% unde es ima ion o he MCD64A1 p oduc using a
e e ence se o mo e han 100 i e pe ime e s (1.24 Mkm
2
) de i ed om Landsa TM/ETM+images
dis ibu ed a ound he wo ld o 2008 and p e iously cons uc ed by Padilla e al. [54].
F om he esul s in e compa ison de e mined o he wo new e sions (Table 3), one can show
ha MCD64A1 p o ided be e BA es ima es han Fi e_CCI in some yea s, despi e ha ing a lowe
spa ial esolu ion (500 s. 250 m), e en in he yea s wi h he la ges BA, which a e he ones ha
con ibu ed mo e o he a e age alues (2004, 2005 and 2015); ne e heless, i pe o med ela i ely
wo se in o he yea s (2009, 2010 and 2011). Fo he MCD64A1 and Fi e_CCI 5.1 p oduc s, in he yea s
wi h he highes amoun o BA (o e 1 million ha), he annual es ima e pe cen ages ended o s abilize
a ound he a e age alue wi h lowe a iabili y (Figu e 7). One should bea in mind ha i is p ecisely
hese yea s ha con ibu e mos o he o e all a e age h oughou he s udy pe iod. Con e sely,
he yea s wi h he leas amoun o BA (below 0.5 million ha) had g ea e a iabili y; his was less
p onounced in he Fi e_CCI 5.1 p oduc han in MCD64A1, p obably due o i s be e spa ial esolu ion
(250 s. 500 m).
Wi h ega d o spa ial accu acy (Table 4), a global analysis showed ha he olde e sions, Fi e_CCI
4.1 and MCD45A1 C5.1, had he lowes commission e o s (5.9% and 6.6%, espec i ely) bu he
omission e o s we e e y high (67.5% and 73.7%, espec i ely). The new e sions, Fi e_CCI 5.1 and
MCD64A1 C6, had signi ican ly educed OE (up o 39.0% and 48.0% espec i ely), e en a he cos
o sligh ly wo se CE (7.5% and 17.8%, espec i ely). As he pe cen age o BA o all p oduc s was
below 100% ( hey unde es ima ed he o al BA), he weigh ed sum o TE ended o gi e a lowe weigh
o he CE han o he OE, a o ing he algo i hms ha educe he omission e o e en a he cos o
inc easing he commission e o s. This low imbalance be ween omission and commission e o s (along
wi h he sha p d op in OE) ansla es in o a be e BA es ima e. These esul s a e compa able o hose
ob ained by o he au ho s using o he e e ence se s [
51
,
58
]. In a ecen alida ion o Fi e_CCI 5.1
using 1200 global samples o e he 2003–2014 pe iod, Lizundia-Loiola e al. [
58
] ob ained alues o
CE =54.4% and OE =67.1%. In he S age 3 alida ion o MCD64A1 C6 using 558 pai s o Landsa
images om 2014 and 2015, Bosche i e al. [
51
] ob ained alues o CE =40.2% and OE =72.6% on a
global scale, which imp o ed signi ican ly o bo eal o es biomass (CE =23.9% and OE =27.0%).
The OE om his la e s udy, which was speci ic o he bo eal egion, di e s signi ican ly om ha
ob ained in he p esen wo k (48.0%) o he Alaska egion; we unde s and ha his may be due o he
Senso s 2020,20, 5423 19 o 23
small e e ence da ase used by Bosche i e al. [
51
] o he en i e bo eal egion. In con as , ou s udy
used all he bu ned a ea pe ime e s eco ded in he 2000–2017 pe iod.
The PB analysis (Figu e 11) allowed us o pa ially explain he annual a iabili y in he
commission/omission e o s o he BA p oduc s. The a e age AUPB alues (Table 5) e lec an
inc ease as he spa ial esolu ion wo sens, indica ing ha he e is a highe pe cen age o commission
and omission e o s a ibu able o he da a’s low spa ial esolu ion. Again, he e was high a iabili y
in he annual alues ela i e o he a e age alues, a ibu able o he agmen a ion o he bu ned
a eas. This high accu acy a iabili y a ibu able o landscape agmen a ion was also epo ed by
Rod igues e al. [
59
], in he accu acy assessmen o MCD64A1 C6 in he B azilian Ce ado s. Landsa
pe ime e s o he 2011–2019 pe iod. Rod igues e al. [
59
] ound ha , in he no he n Ce ado, which had
la ge a eas a ec ed by i e, he MCD64A1 C6 pe o mance was signi ican ly highe han in he sou he n
Ce ado a ea, whe e a mo e agmen ed landscape and smalle pa ches o i e p edomina ed.
The linea eg ession analysis be ween he TE and he AUPB ound an upwa d linea end o
he new e sions o he BA p oduc s; his pa ially jus i ies he high commission and omission e o s
encoun e ed in some yea s. Howe e , i is also appa en ha in some yea s (e.g., 2016), he high
le el o i e agmen a ion had li le in luence on he TE. Con e sely, in yea s wi h low le els o
agmen a ion (e.g., 2002 and 2005 o Fi e_CCI 5.1 and 2003 and 2014 o MCD64A1 C6), he TEs we e
signi ican . Fo hese yea s, he obse a ions we e limi ed by o he ac o s such as he poo beha io
o he de ec ion algo i hms, he low se e i y o bu ned a eas o ce ain en i onmen al condi ions; as
indica ed by Loboda e al. [
60
], hese ac o s migh explain such e o s. The mo e p ecise linea i o
he new e sions compa ed o he old ones, wi h R
2
alues close o 0.5 and a highe slope, e lec s
g ea e TE sensi i i y (CE and OE) o bu ned a ea agmen a ion.
O all he yea s analyzed, and excep o he yea s 2000 and 2001 (yea s wi h incomple e da a
om he MODIS senso ), he e a e wo yea s in which he la es e sions o MCD64A1 and Fi e_CCI
pe o med poo ly. The yea 2006 showed low de ec ion a es (29.6% and 45.83%, espec i ely), high
omission alues (83.9% and 60.9%) and high commission alues (29.6% and 14.7%). The bu ned a ea
eco ded (0.11 million ha) was below he annual a e age, as was he same pe cen age co esponding o
la ge i es (69.82%), which was also below he annual a e age, while he PB a eas a 250 (0.002257)
and 500 m (0.006480) we e almos double he a e age alues a hese esolu ions. Simila ly, in 2008,
he yea wi h he leas amoun o bu ned a ea (0.04 million ha) and he lowes pe cen age a ibu able
o la ge i es (36.65%), he e we e high commission and omission e o alues (57.3% and 71.2% o
MCD64A1 and 29.2% and 43.8% o Fi e_CCI) bu a la ge pe cen age o bu ned a ea de ec ed (67.49%
and 79.43%, espec i ely). Fo ha yea , he a ea enclosed by he PB a 250/500 m was he highes , so a
signi ican pa o he spa ial e o s was due o he da a’s low spa ial esolu ion.
5. Conclusions
An independen and de ailed s udy was ca ied ou o e alua e he spa io empo al accu acy
o he la es e sions (along wi h he p e ious e sions) o he wo main global scale BA p oduc s
de i ed om MODIS-senso sa elli e images, he e es ic ed o he Alaskan bo eal egion. As e e ence
da a, we used all he polygons o he BA eco ded by AFS o e he s udy pe iod. In addi ion,
a de ailed 50 m
×
50 m pixel analysis o he accu acy o each BA p oduc was pe o med. Fi e_CCI
5.1 and MCD64A1 C6 p esen ed signi ican imp o emen s o e hei p e ious e sions in e ms o
BA es ima ion. Imp o emen s we e achie ed by educing he imbalance be ween commission and
omission e o s and, especially, by g ea ly educing omission e o s e en a he expense o wo sening
commission e o s. Bo h p oduc s cu en ly p oduce simila BA es ima ion pe cen ages, al hough
he posi ional accu acy o Fi e_CCI is be e han ha o MCD64A1, which is in line wi h i s highe
spa ial esolu ion (250 s. 500 m). Fi e_CCI 5.1 would be he op ion chosen o use s who, h ough
geoin o ma ics analysis echniques, use BA p oduc s o o es i e managemen . Fo hose use s who
a e s udying he inc ease in g eenhouse gas concen a ions o he change in he chemical composi ion
Senso s 2020,20, 5423 20 o 23
o he a mosphe e due o i e emissions, any o he la es e sions o he p oduc s analyzed could be
sui able, bu always aking in o accoun he e o s o omission (o e 40%) in bo h cases.
The high a iabili y in annual esul s is no ewo hy, bo h in he pe cen ages o BA de ec ion and
in he omission and commission e o s, which pu s in o ques ion much accu acy assessmen wo k ha
uses limi ed spa io empo al e e ence da a. I would be ecommendable o use all possible e e ence
da a when a ailable.
The yea ly analysis o bu ned-a ea agmen a ion ac oss he co esponding Pa e o bounda ies has
allowed us o es ablish a quan i a i e measu e o he same (AUPB), which ela es o he o al e o s
(weigh ed sum o commission and omission e o s) ob ained o he la es e sions o he Fi e_CCI and
MCD64A1 p oduc s. The esul s om his wo k could be ex apola ed o o he bo eal egions ha do
no ha e such accu a e e e ence da ase s as a e a ailable in he no he n egions o No h Ame ica.
Au ho Con ibu ions:
Concep ualiza ion and me hodology, J.A.M.-R., J.R.G.-L. and M.A.; alida ion and o mal
analysis, J.A.M.-R., J.R.G.-L., M.A. and M.C.-G.; w i ing—o iginal d a p epa a ion, J.R.G.-L. and J.A.M.-R.;
w i ing— e iew and edi ing, M.A. and M.C.-G.; p ojec adminis a ion, J.A.M.-R. and M.A.; and unding
acquisi ion, M.A. and J.A.M.-R. All au ho s ha e ead and ag eed o he published e sion o he manusc ip .
Funding:
This esea ch was unded by he Minis e io de Ciencia, Inno aci
ó
n y Uni e sidades (MCIU), he Agencia
Es a al de In es igaci
ó
n (AEI) and he Fondo Eu opeo de Desa ollo Regional (FEDER) h ough he p ojec
RTI2018-099171-B-I00. The Uni e sidad de Alme
í
a and he FEDER-ANDALUCIA Ope a ing P og am pa ially
inanced his wo k h ough he UAL-TIC-A023-B1 b idge p ojec , in he 2018 call.
Acknowledgmen s:
We hank he ou anonymous pee e iewe s o hei aluable commen s and sugges ions.
The au ho s exp ess hei g a i ude o he Clima e Change Ini ia i e o he Eu opean Space Agency, he Fi e_CCI
p ojec , NASA, he Uni e si y o Ma yland and AFS, o he p ocessing and ee dis ibu ion o he bu ned a ea
p oduc da a used in his wo k.
Con lic s o In e es : The au ho s decla e no con lic o in e es .
Re e ences
1.
Van de We , G.R.; Rande son, J.T.; Giglio, L.; an Leeuwen, T.T.; Chen, Y.; Roge s, B.M.; Mu, M.; an
Ma le, M.J.E.; Mo on, D.C.; Colla z, G.J.; e al. Global i e emissions es ima es du ing 1997–2016. Ea h Sys .
Sci. Da a 2017,9, 697–720. [C ossRe ]
2.
Ki zbe ge , T.; Falk, D.A.; Wes e ling, A.L.; Swe nam, T.W. Di ec and indi ec clima e con ols p edic
he e ogeneous ea ly-mid 21s cen u y wild i e bu ned a ea ac oss wes e n and bo eal No h Ame ica.
PLoS ONE 2017,12, e0188486. [C ossRe ] [PubMed]
3.
Mouillo , F.; Schul z, M.G.; Yue, C.; Cadule, P.; Tansey, K.; Ciais, P.; Chu ieco, E. Ten yea s o global bu ned
a ea p oduc s om spacebo ne emo e sensing—A e iew: Analysis o use needs and ecommenda ions o
u u e de elopmen s. In . J. Appl. Ea h Obs. Geoin . 2014,26, 64–79. [C ossRe ]
4.
Simon, M.; Plumme , S.; Fie ens, F.; Hoelzemann, J.J.; A ino, O. Bu n a ea de ec ion a global scale using
ATSR-2: The GLOBSCAR p oduc s and hei quali ica ion. J. Geophys. Res. D A mos.
2004
,109, 1–16.
[C ossRe ]
5.
Tansey, K.; G
é
goi e, J.M.; S oppiana, D.; Sousa, A.; Sil a, J.; Pe ei a, J.M.C.; Bosche i, L.; Maggi, M.;
B i io, P.A.; F ase , R.; e al. Vege a ion bu ning in he yea 2000: Global bu ned a ea es ima es om SPOT
VEGETATION da a. J. Geophys. Res. D A mos. 2004,109, 1–22. [C ossRe ]
6.
Ca mona-Mo eno, C.; Belwa d, A.; Maling eau, J.P.; Ha ley, A.; Ga cia-Aleg e, M.; An ono skiy, M.;
Buchsh abe , V.; Pi o a o , V. Cha ac e izing in e annual a ia ions in global i e calenda using da a om
Ea h obse ing sa elli es. Glob. Chang. Biol. 2005,11, 1537–1555. [C ossRe ]
7.
Plumme , S.; A ino, O.; Simon, M.; S e en, W. Es ablishing a ea h obse a ion p oduc se ice o he
e es ial ca bon communi y: The globca bon ini ia i e. Mi ig. Adap . S a eg. Glob. Chang.
2006
,11, 97–111.
[C ossRe ]
8.
Tansey, K.; G
é
goi e, J.M.; De ou ny, P.; Leigh, R.; Pekel, J.F.; an Bogae , E.; Ba holom
é
, E. A new, global,
mul i-annual (2000–2007) bu n a ea p oduc a 1 km esolu ion. Geophys. Res. Le .
2008
,35, 01401.
[C ossRe ]
Senso s 2020,20, 5423 21 o 23
9.
Tansey, K.; B adley, A.; Sme s, B.; an Bes , C.; Lacaze, R. The Geoland2 BioPa bu ned a ea p oduc .
In P oceedings o he Eu opean Geosciences Union Gene al Assembly, Vienna, Aus ia, 22–27 Ap il 2012;
p. 4727.
10.
Giglio, L.; Rande son, J.T.; Van De We , G.R. Analysis o daily, mon hly, and annual bu ned a ea using
he ou h-gene a ion global i e emissions da abase (GFED4). J. Geophys. Res. Biogeosci.
2013
,118, 317–328.
[C ossRe ]
11.
Hawbake , T.J.; Vande hoo , M.K.; Beal, Y.J.; Takacs, J.D.; Schmid , G.L.; Falgou , J.T.; Williams, B.;
Fai aux, N.M.; Caldwell, M.K.; Pico e, J.J.; e al. Mapping bu ned a eas using dense ime-se ies o
Landsa da a. Remo e Sens. En i on. 1997,198, 504–522. [C ossRe ]
12.
Chu ieco, E.; Lizundia-Loiola, J.; Luc ecia Pe ina i, M.; Ramo, R.; Padilla, M.; Tansey, K.; Mouillo , F.;
Lau en , P.; S o m, T.; Heil, A.; e al. Gene a ion and analysis o a new global bu ned a ea p oduc based on
MODIS 250 m e lec ance bands and he mal anomalies. Ea h Sys . Sci. Da a
2018
,10, 2015–2031. [C ossRe ]
13.
Giglio, L.; Bosche i, L.; Roy, D.P.; Humbe , M.L.; Jus ice, C.O. The Collec ion 6 MODIS bu ned a ea mapping
algo i hm and p oduc . Remo e Sens. En i on. 2018,217, 72–85. [C ossRe ] [PubMed]
14.
Giglio, L.; Sch oede , W.; Jus ice, C.O. The collec ion 6 MODIS ac i e i e de ec ion algo i hm and i e
p oduc s. Remo e Sens. En i on. 2016,178, 31–41. [C ossRe ] [PubMed]
15.
Jus ice, C.O.; Ve mo e, E.; Townshend, J.R.G.; De ies, R.; Roy, D.P.; Hall, D.K.; Salomonson, V.V.; P i e e, J.L.;
Riggs, G.; S ahle , A.; e al. The mode a e esolu ion imaging spec o adiome e (MODIS): Land emo e
sensing o global change esea ch. IEEE T ans. Geosci. Remo e Sens. 1998,36, 1228–1249. [C ossRe ]
16.
Xiong, X.; Wenny, B.N.; Ba nes, W.D. O e iew o NASA Ea h Obse ing Sys ems Te a and Aqua mode a e
esolu ion imaging spec o adiome e ins umen calib a ion algo i hms and on-o bi pe o mance. J. Appl.
Remo e Sens. 2009,3, 032501. [C ossRe ]
17.
Ciais, P.; Moo e, B.; S e en, W.; Hood, M.; Quegan, S.; Cihla , J.; Raupach, M.; Tschi ley, J.; Inoue, G.;
Doney, S.; e al. In eg a ed Global Ca bon Obse a ion Theme: A S a egy o Realise a Coo dina ed Sys em
o In eg a ed Global Ca bon Cycle Obse a ions. A ailable online: h ps://www.globalca bonp ojec .o g/
global/pd /IGOS_Ca bonB ochu e.pd (accessed on 21 Sep embe 2020).
18.
Townshend, J.; La ham, J.; A ino, O. In eg a ed Global Obse a ions o he Land: An IGOS-P Theme.
A ailable online: h p://www. ao.o g/3/i0536e/i0536e00.h m (accessed on 21 Sep embe 2020).
19.
Ciais, P.; Dolman, H.; Da ga ille, R.; Ba ie, L.; Bombelli, A.; Bu le , J.; Canadell, P.; Mo iyama, T.; Bo ges, A.;
Bo e smann, H.; e al. GEO Ca bon S a egy. A ailable online: h ps://www.globalca bonp ojec .o g/global/
pd /GEO_CARBONSTRATEGY_20101020.pd (accessed on 21 Sep embe 2020).
20.
Jus ice, C.O.; Giglio, L.; Roy, D.P.; Csisza , I.; Bosche i, L.; Ko on zi, S.; Woos e , M.J. Whi e Pape on a NASA
Fi e ESDR. A ailable online: h ps://cce.nasa.go /m g2008_ab_p esen a ions/Fi e_Jus ice_whi epape .pd
(accessed on 9 Sep embe 2020).
21.
Smi aglia, D.; Filipponi, F.; Mand one, S.; To na o, A.; Ta amelli, A. Ag eemen index o bu ned a ea
mapping: In eg a ion o mul iple spec al indices using Sen inel-2 sa elli e images. Remo e Sens.
2020
,
12, 1862. [C ossRe ]
22.
Bas a ika, A.; Chu ieco, E.; Ma
í
n, M.P. Mapping bu ned a eas om landsa TM/ETM+da a wi h a
wo-phase algo i hm: Balancing omission and commission e o s. Remo e Sens. En i on.
2011
,115, 1003–1012.
[C ossRe ]
23.
Heil, A.; Yue, C.; Mouillo , F.; Kaise , J.W. ESA Clima e Change Ini ia i e—Fi e_cci D1.1 Use Requi emen
Documen (URD). A ailable online: h p://esa- i e-cci.o g/ iles/Fi e_cci_D1.1_URD_ 5.1.pd (accessed on
9 Sep embe 2020).
24.
Plumme , S.; Chu ieco, E.; Pe ina i, M.L.; O
ó
n, G.; S o m, T.; Klos e , S.; De ou ny, P.; Lama che, C.
Fi e_cci Use Requi emen s Documen & P oduc Speci ica ion Documen o AVHRR. A ailable online:
h ps://www.esa- i e-cci.o g/si es/de aul / iles/Fi e_cci_O2.D1_URD_PSD_AVHRR_ 1.1.pd (accessed on
9 Sep embe 2020).
25.
Clima e Modelling Use G oup Deli e able 1.1 Requi emen s Baseline Documen . A ailable online:
h p://ensembles-eu.me o ice.com/cmug/CMUG_PHASE_2_D1.1_Requi emen s_ 0.6.pd (accessed on
21 Sep embe 2020).
26.
Chu ieco, E.; Mouillo , F.; an de We , G.R.; San Miguel, J.; Tanasse, M.; Kou sias, N.; Ga c
í
a, M.; Yeb a, M.;
Padilla, M.; Gi as, I.; e al. His o ical backg ound and cu en de elopmen s o mapping bu ned a ea om
sa elli e Ea h obse a ion. Remo e Sens. En i on. 2019,225, 45–64. [C ossRe ]

Senso s 2020,20, 5423 22 o 23
27.
Hicke, J.A.; Asne , G.P.; Kasischke, E.S.; F ench, N.H.F.; Rande son, J.T.; Colla z, G.J.; S ocks, B.J.; Tucke , C.J.;
Los, S.O.; Field, C.B. Pos i e esponse o No h Ame ican bo eal o es ne p ima y p oduc i i y analyzed
wi h sa elli e obse a ions. Glob. Chang. Biol. 2003,9, 1145–1157. [C ossRe ]
28.
Mo eno-Ruiz, J.A.; Ga cia-Laza o, J.R.; Riano, D.; Ke au e , S.C. The syne gy o he 0.05
◦
(
∼
5 km) AVHRR
long- e m da a eco d (LTDR) and landsa TM a chi e o map la ge i es in he No h Ame ican bo eal egion
om 1984 o 1998. IEEE J. Sel. Top. Appl. Ea h Obs. Remo e Sens. 2014,7, 1157–1166. [C ossRe ]
29.
Wolken, J.M.; Hollingswo h, T.N.; Rupp, T.S.; Chapin, F.S.; T aino , S.F.; Ba e , T.M.; Sulli an, P.F.;
Mcgui e, A.D.; Euski chen, E.S.; Hennon, P.E.; e al. E idence and implica ions o ecen and p ojec ed
clima e change in Alaska’s o es ecosys ems. Ecosphe e 2011,2. [C ossRe ]
30.
AK Fi e His o y Pe ime e s. A ailable online: h ps://www.a cgis.com/home/i em.h ml?id=
d4b8d89 226 4c488e1e4ba054e49be9 (accessed on 23 Ap il 2019).
31.
Alaska Fi e Se ice (AFS) Alaska Wildland Fi e In o ma ion Map Se ies. A ailable online: h ps://blm-
egis.maps.a cgis.com/apps/MapSe ies/index.h ml?appid=32ec4 34 b234ce58d 6b1222a207e 1 (accessed on
9 Sep embe 2020).
32.
A none, E.; F ancipane, A.; Sca baci, A.; Puglisi, C.; No o, L.V. E ec o as e esolu ion and
polygon-con e sion algo i hm on landslide suscep ibili y mapping. En i on. Model. So w.
2016
,84,
467–481. [C ossRe ]
33.
Alonso-Canas, I.; Chu ieco, E. Global bu ned a ea mapping om ENVISAT-MERIS and MODIS ac i e i e
da a. Remo e Sens. En i on. 2015,163, 140–152. [C ossRe ]
34.
Roy, D.P.; Lewis, P.E.; Jus ice, C.O. Bu ned a ea mapping using mul i- empo al mode a e spa ial esolu ion
da a—A bi-di ec ional e lec ance model-based expec a ion app oach. Remo e Sens. En i on.
2002
,83,
263–286. [C ossRe ]
35.
Roy, D.P.; Jin, Y.; Lewis, P.E.; Jus ice, C.O. P o o yping a global algo i hm o sys ema ic i e-a ec ed a ea
mapping using MODIS ime se ies da a. Remo e Sens. En i on. 2005,97, 137–162. [C ossRe ]
36.
Roy, D.P.; Bosche i, L.; Jus ice, C.O.; Ju, J. The collec ion 5 MODIS bu ned a ea p oduc —Global e alua ion
by compa ison wi h he MODIS ac i e i e p oduc . Remo e Sens. En i on. 2008,112, 3690–3707. [C ossRe ]
37.
Bosche i, L.; Roy, D.; Ho mann, A.A.; Humbe , M. MODIS Collec ion 5.1 Bu ned A ea P oduc —MCD45.
A ailable online: h p://modis- i e.umd.edu/ iles/MODIS_Bu ned_A ea_Collec ion51_Use _Guide_3.1.0.pd
(accessed on 9 Sep embe 2020).
38.
Giglio, L.; Bosche i, L.; Roy, D.; Ho mann, A.A.; Humbe , M. Collec ion 6 MODIS Bu ned A ea P oduc
Use ’s Guide Ve sion 1.0. A ailable online: h ps://modis-land.gs c.nasa.go /pd /MODIS_C6_BA_Use _
Guide_1.0.pd (accessed on 9 Sep embe 2020).
39.
Congal on, R.G. A e iew o assessing he accu acy o classi ica ions o emo ely sensed da a. Remo e Sens.
En i on. 1991,37, 35–46. [C ossRe ]
40.
S ehman, S.V. Selec ing and in e p e ing measu es o hema ic classi ica ion accu acy. Remo e Sens. En i on.
1997,62, 77–89. [C ossRe ]
41.
Bosche i, L.; Flasse, S.P.; B i io, P.A. Analysis o he con lic be ween omission and commission in low spa ial
esolu ion dicho omic hema ic p oduc s: The Pa e o Bounda y. Remo e Sens. En i on.
2004
,91, 280–292.
[C ossRe ]
42.
B adley, A.P. The use o he a ea unde he ROC cu e in he e alua ion o machine lea ning algo i hms.
Pa e n Recogni . 1997,30, 1145–1159. [C ossRe ]
43.
Huang, J.; Ling, C.X. Using AUC and accu acy in e alua ing lea ning algo i hms. IEEE T ans. Knowl. Da a
Eng. 2005,17, 299–310. [C ossRe ]
44.
Gu, X.; Wu, Z.; Zhang, Y.; Yan, S.; Fu, J.; Du, L. P edic ion esea ch o he o es i e in Jiangxi p o ince in he
backg ound o clima e change. Sheng ai Xuebao 2020,40. [C ossRe ]
45.
Fe n
á
ndez-Manso, A.; Quin ano, C. A syne ge ic app oach o bu ned a ea mapping using maximum en opy
modeling ained wi h hype spec al da a and VIIRS ho spo s. Remo e Sens. 2020,12, 858. [C ossRe ]
46.
De Bem, P.P.; De Ca alho, O.A., J .; Ma ica di, E.A.T.; Guima
ã
es, R.F.; Gomes, R.A.T. P edic ing wild i e
ulne abili y using logis ic eg ession and a i icial neu al ne wo ks: A case s udy in B azil’s Fede al Dis ic .
In . J. Wildland Fi e 2019,28, 35–45. [C ossRe ]
47.
Mi sopoulos, I.; Mallinis, G. A da a-d i en app oach o assess la ge i e size gene a ion in G eece. Na . Haza ds
2017,88, 1591–1607. [C ossRe ]
Senso s 2020,20, 5423 23 o 23
48.
Go se ski, P.V.; Gessle , P.E.; Fol z, R.B.; Ellio , W.J. Spa ial p edic ion o landslide haza d using logis ic
eg ession and ROC analysis. T ans. GIS 2006,10, 395–415. [C ossRe ]
49.
Bo oughani, M.; Pou hashemi, S.; Hashemi, H.; Salehi, M.; Ami ahmadi, A.; Asadi, M.A.Z.; Be nd sson, R.
Applica ion o emo e sensing echniques and machine lea ning algo i hms in dus sou ce de ec ion and
dus sou ce suscep ibili y mapping. Ecol. In o m. 2020,56, 101059. [C ossRe ]
50.
Chang, Z.; Du, Z.; Zhang, F.; Huang, F.; Chen, J.; Li, W.; Guo, Z. Landslide suscep ibili y p edic ion based on
emo e sensing images and GIS: Compa isons o supe ised and unsupe ised machine lea ning models.
Remo e Sens. 2020,12, 502. [C ossRe ]
51.
Bosche i, L.; Roy, D.P.; Giglio, L.; Huang, H.; Zubko a, M.; Humbe , M.L. Global alida ion o he collec ion
6 MODIS bu ned a ea p oduc . Remo e Sens. En i on. 2019,235, 111490. [C ossRe ]
52.
Mo eno-Ruiz, J.A.; Ga c
í
a-L
á
za o, J.R.; A belo, M.; Riaño, D. A compa ison o bu ned a ea ime se ies in he
alaskan bo eal o es s om di e en emo e sensing p oduc s. Fo es s 2019,10, 363. [C ossRe ]
53.
Fo nacca, D.; Ren, G.; Xiao, W. Pe o mance o Th ee MODIS i e p oduc s (MCD45A1, MCD64A1,
MCD14ML), and ESA Fi e_CCI in a moun ainous a ea o No hwes Yunnan, China, cha ac e ized by
equen small i es. Remo e Sens. 2017,9, 1131. [C ossRe ]
54.
Padilla, M.; S ehman, S.V.; Ramo, R.; Co i, D.; Han son, S.; Oli a, P.; Alonso-Canas, I.; B adley, A.V.;
Tansey, K.; Mo a, B.; e al. Compa ing he accu acies o emo e sensing global bu ned a ea p oduc s using
s a i ied andom sampling and es ima ion. Remo e Sens. En i on. 2015,160, 114–121. [C ossRe ]
55.
Ga c
í
a-L
á
za o, J.R.; Mo eno-Ruiz, J.A.; Riaño, D.; A belo, M. Es ima ion o bu ned a ea in he No heas e n
Sibe ian bo eal o es om a Long-Te m Da a Reco d (LTDR) 1982-2015 ime se ies. Remo e Sens.
2018
,
10, 940. [C ossRe ]
56.
Tu co, M.; He e a, S.; Tou igny, E.; Chu ieco, E.; P o enzale, A. A compa ison o emo ely-sensed and
in en o y da ase s o bu ned a ea in Medi e anean Eu ope. In . J. Appl. Ea h Obs. Geoin .
2019
,82, 101887.
[C ossRe ]
57.
Campagnolo, M.L.; Oom, D.; Padilla, M.; Pe ei a, J.M.C. A pa ch-based algo i hm o global and daily bu ned
a ea mapping. Remo e Sens. En i on. 2019,232, 111288. [C ossRe ]
58.
Lizundia-Loiola, J.; O
ó
n, G.; Ramo, R.; Chu ieco, E. A spa io- empo al ac i e- i e clus e ing app oach o
global bu ned a ea mapping a 250 m om MODIS da a. Remo e Sens. En i on.
2020
,236, 111493. [C ossRe ]
59.
Rod igues, J.A.; Libona i, R.; Pe ei a, A.A.; Noguei a, J.M.P.; San os, F.L.M.; Pe es, L.F.; San a Rosa, A.;
Sch oede , W.; Pe ei a, J.M.C.; Giglio, L.; e al. How well do global bu ned a ea p oduc s ep esen i e
pa e ns in he B azilian Sa annas biome? An accu acy assessmen o he MCD64 collec ions. In . J. Appl.
Ea h Obs. Geoin . 2019,78, 318–331. [C ossRe ]
60.
Loboda, T.V.; Hoy, E.E.; Giglio, L.; Kasischke, E.S. Mapping bu ned a ea in Alaska using MODIS da a: A da a
limi a ions-d i en modi ica ion o he egional bu ned a ea algo i hm. In . J. Wildland Fi e
2011
,20, 487–496.
[C ossRe ]
©
2020 by he au ho s. Licensee MDPI, Basel, Swi ze land. This a icle is an open access
a icle dis ibu ed unde he e ms and condi ions o he C ea i e Commons A ibu ion
(CC BY) license (h p://c ea i ecommons.o g/licenses/by/4.0/).