emo e sensing
A icle
A Li e Fuel Mois u e Con en P oduc om Landsa
TM Sa elli e Time Se ies o Implemen a ion in Fi e
Beha io Models
Ma iano Ga cía1,2 , Da id Riaño 3,4,* , Ma a Yeb a 5,6,7 , Ja ie Salas 1, Ad ián Ca dil 8,9 ,
San iago Monede o 8, Joaquín Rami ez 9, M. Pila Ma ín3, La a Vila 3, John Gaja do 10
and Susan Us in 4
1
En i onmen al Remo e Sensing Resea ch G oup, Depa men o Geology, Geog aphy and he En i onmen ,
Uni e sidad de Alcalá. Calle Colegios 2, 28801 Alcaláde Hena es, Spain; ma iano.ga [email p o ec ed] (M.G.);
ja ie [email p o ec ed] (J.S.)
2Complu um Tecnologías de la In o mación Geog á ica S.L. (COMPLUTIG), Colegios, 2,
28801 Alcaláde Hena es, Spain
3En i onmen al Remo e Sensing and Spec oscopy Labo a o y (SpecLab), Spanish Na ional Resea ch
Council (CSIC), 28037 Mad id, Spain; mpila [email p o ec ed] (M.P.M.); [email p o ec ed] (L.V.)
4Cen e o Spa ial Technologies and Remo e Sensing (CSTARS), John Mui Ins i u e o he En i onmen ,
Uni e si y o Cali o nia Da is, One Shields D i e, Da is, CA 95616, USA; [email p o ec ed]
5Fenne School o En i onmen & Socie y, Colleges o Science, The Aus alian Na ional Uni e si y, Ac on,
ACT 2601, Aus alia; [email p o ec ed]
6Bush i e and Na u al Haza ds Coope a i e Resea ch Cen e, 340 Albe S .,
Eas Melbou ne, Vic o ia 3002, Aus alia
7Resea ch School o Ae ospace, Mechanical and En i onmen al Enginee ing, College o Enginee ing and
Compu e Science, The Aus alian Na ional Uni e si y, Ac on, ACT 2601, Aus alia
8Technosyl a, 24009 León, Spain; [email p o ec ed] (A.C.); [email p o ec ed] (S.M.)
9Technosyl a, La Jolla, CA 92037-7231, USA; [email p o ec ed]
10
Ins i u o de Bosques y Sociedad, Facul ad de Ciencias Fo es ales Y Recu sos Na u ales, Uni e sidad Aus al
de Chile, Campus Isla Teja, Valdi ia 5090000, Chile; [email p o ec ed]
*Co espondence: [email p o ec ed]
Recei ed: 10 Ap il 2020; Accep ed: 24 May 2020; Published: 27 May 2020


Abs ac :
Li e Fuel Mois u e Con en (LFMC) con ibu es o i e dange and beha io , as i a ec s
i e igni ion and p opaga ion. This pape p esen s a wo laye ed Landsa LFMC p oduc based on
opog aphically co ec ed ela i e Spec al Indices (SI) o e a 2000–2011 ime se ies, which can be
in eg a ed in o i e beha io simula ion models. Nine chapa al sampling si es ac oss h ee Landsa -5
Thema ic Mappe (TM) scenes we e used o alida e he p oduc o e he Wes e n USA. The ela ions
be ween ield-measu ed LFMC and Landsa -de i ed SIs we e s ong o each indi idual si e bu
wo sened when pooled oge he . The Enhanced Vege a ion Index (EVI) p esen ed he s onges
co ela ions ( ) and he leas Roo Mean Squa e E o (RMSE), ollowed by he No malized Di e ence
In a ed Index (NDII), No malized Di e ence Vege a ion Index (NDVI) and Visible A mosphe ically
Resis an Index (VARI). The ela ions be ween LFMC and he SIs o all si es imp o ed a e using
hei ela i e alues and ela i e LFMC, inc easing om 0.44 up o 0.69 o ela i e EVI ( elEVI),
he bes p edic i e a iable. This elEVI se ed o es ima e he he baceous and woody LFMC based
on minimum and maximum seasonal LFMC alues. The unde s o y he baceous LFMC on he
woody pixels was ex apola ed om he su ounding pixels whe e he he baceous ege a ion is he
op laye . Running simula ions on he Wild i e Analys (WFA) i e beha io model demons a ed
ha his LFMC p oduc alone impac s signi ican ly he i e spa ial dis ibu ion in e ms o bu ned
p obabili y, wi h a e age bu ned a ea di e ences o e 21% a e 8 h bu ning since igni ion, compa ed
o commonly ca ied ou simula ions based on cons an alues o each uel model. The me hod
Remo e Sens. 2020,12, 1714; doi:10.3390/ s12111714 www.mdpi.com/jou nal/ emo esensing
Remo e Sens. 2020,12, 1714 2 o 15
could be applied o Landsa -7 and -8 and Sen inel-2A and -2B a e p ope senso in e -calib a ion
and opog aphic co ec ion.
Keywo ds:
li e uel mois u e con en ; Landsa -5 TM; i e beha io simula o ; i e dange ; i e
p opaga ion; da a no maliza ion
1. In oduc ion
Fi e dis u bances play a key ole in ege a ion succession, as well as in he ecosys em’s s uc u e
and unc ion [
1
]. Li e and dead biomass cons i u es he uel ha bu ns du ing a i e, and he uel
p ope ies desc ibe hei s a e o mois u e con en , as well as hei spa ial dis ibu ion and impac on
i e sp ead, in ensi y and se e i y [
2
]. Among hese p ope ies, i e igni ion and p opaga ion depend
on Li e Fuel Mois u e Con en (LFMC) [
3
–
6
]. Fuels wi h high LFMC ake longe o igni e as wa e
ac s as a hea sink, slowing down i e sp ead and in ensi y [
5
,
7
]. LFMC is de ined as he amoun o
wa e in he uel o e i s d y weigh imes 100. This amoun o wa e is calcula ed as he di e ence
be ween he esh weigh and he o en-d ied weigh a 60–100
◦
C o 24–48 h [
5
]. Mo e ecen ly,
Ma hews [
8
] sugges ed o d y samples a 105
◦
C o ensu e comple e wa e emo al om he samples.
The US Na ional Fi e Dange Ra ing Sys em (NFDRS) dis inguishes annual and pe ennial he baceous
LFMC depending on how he d ying o he li e uel occu s h oughou he yea [
9
]. In addi ion, he
NFDRS also conside s he woody LFMC, measu ing he mois u e o he oliage and o small wigs ha
a e <0.6 cm [9].
Clima e and plan adap a ion s a egies o d ough play a key ole on LFMC, wi h changes in
he wa e con en o lea es as well as in d y ma e [
10
]. LFMC emo e sensing es ima es ely on he
spec al changes due o he di ec impac o liquid wa e abso p ion ea u es and he indi ec impac o
pigmen and s uc u al changes associa ed wi h wa e con en a ia ion [
11
]. Two di e en app oaches
ha e been applied o moni o LFMC om emo e sensing da a: empi ical Spec al Indices (SI) [
12
–
14
]
and adia i e ans e models (RTM) [
15
–
17
]. RTM only ou pe o ms empi ical models i hey a e
app op ia ely pa ame e ized and cons ained, which equi es accu a e s uc u al in o ma ion [
15
].
Yeb a e al. [11] p o ide a comple e e iew on hese me hods and hei ope a ional implica ions.
Fi e managemen ools demand comp ehensi e spa ial and empo al LFMC co e age [
18
].
The e o e, ield sampling only se es o calib a e and alida e hese emo e sensing es ima es.
Ope a ional LFMC emo e sensing p oduc s bene i i e beha io models, as hey can imp o e i e
g ow h simula ions. Mos i e simula o s gene ally include a cons an LFMC alue o each uel
model, hus missing he spa ial LFMC a iabili y ac oss he landscape. The Modeling Dynamic
Fuels wi h an Index Sys em (MoD-FIS) om he LANDFIRE p og am (h ps://www.land i e.go /;
las accessed 3 Ap il 2020) goes u he , de ec ing he key seasonal changes in he baceous ege a ion
o adjus hei dynamic uel models. Howe e , ope a ional ools such as Wild i e Analys (WFA,
h p://wild i eanalys .com/[
19
]) demands ope a ional spa ially and empo ally explici LFMC p oduc s
o be e es ima e i e beha io . Based on BEHAVE su ace i e beha io model [
2
], WFA al eady diges s
cu en spa ial wea he da a a ailable in eal ime and allows he inclusion o p ede ined LFMC laye s
as an inpu . O he so wa e like FlamMap (h ps://www. i elab.o g/p ojec / lammap/; las accessed
3 Ap il 2020) o FARSITE (h ps://www. i elab.o g/p ojec / a si e/; las accessed 3 Ap il 2020) do no
include LFMC as a laye .
A common limi a ion in i e beha io models is also ha hey equi e he baceous and woody
LFMC, whe eas op ical emo e sensing is only sensi i e o he op laye . An al e na i e is o use
me eo ological phenological models like he G owing Season Index (GSI) [
20
]. The 2016 NFDRS
depends on his index o p edic he baceous and woody LFMC. Despi e his, he index equi es
ex apola ion om me eo ological loca ions o build a spa ially comp ehensi e LFMC map.
Remo e Sens. 2020,12, 1714 3 o 15
High empo al esolu ion senso s such as AVHRR [
13
,
21
] o MODIS [
16
,
22
] o VIIRS, allow he
cap u ing o daily changes in LFMC. Ne e heless, cloud co e age can educe hei e ec i e empo al
esolu ion. Besides, spa ial he e ogenei y o uels limi s hei applica ion, due o hei low spa ial
esolu ion (
≥
250 m). Medium spa ial esolu ion (20–30 m) senso s, such as Landsa -5 Thema ic Mappe
(TM), enable a be e spa ial cha ac e iza ion [
23
], bu only once e e y 16 days un il i s decommission.
The combina ion o Landsa -7 and 8, and Sen inel-2A and 2B ensu es an o e pass e e y 3 days a
he equa o and nea ly daily a mid-la i udes a 10–30 m spa ial esolu ion [
24
]. A e p ope senso
in e -calib a ion, hese da a open new oppo uni ies o quan i y LFMC o i e managemen applica ions.
Fo his end, LFMC signal needs o be disc imina ed om a mosphe ic and opog aphic e ec s, sun
and senso geome y, soil backg ound, species composi ion o o he plan cha ac e is ics [11].
To compensa e o some o hese ac o s, se e al au ho s p e e o ela e he LFMC dynamics o
Rela i e Spec al Indices ( elSI) [
25
–
28
]. The elSI no malize he SI o each pixel based on i s alues
wi hin a su icien ly la ge empo al se ies. The main goal o his s udy is o p opose an ope a ional
spa ially dynamic LFMC p oduc including an he baceous and woody laye eady o in eg a ion in
i e beha io models. The p oduc es s di e en Landsa -5 TM elSI no malized o e 10+yea long
ime se ies. Finally, i e simula ions wi h WFA demons a e how LFMC impac s i e beha io , bu ned
p obabili y (BP) and i e g ow h.
2. Me hods
2.1. S udy Si es and Landsa -5 TM Da a
This s udy selec ed nine chapa al si es o alida ion wi h a long Landsa -5 TM ime se ies
eco d and concu en ield LFMC sampling da a om he Na ional Fuel Mois u e Da abase (NFMD,
h p://www.w as.ne /index.php/na ional- uel-mois u e-da abase-mois u e-d ough -103; las accessed
3 Ap il 2020). The LFMC da a collec ion o he NFMC ollows s anda d p o ocols, as desc ibed in
Polle and B own [
29
]. B ie ly, LFMC is ob ained collec ing samples a di e en heigh s on he sh ubs
and om di e en indi iduals [29].
The si es we e in Cali o nia and O egon, wi hin h ee Landsa -5 TM scenes ac oss he Wes e n US
co e ing a wide ange o en i onmen al g adien s (Figu e 1; Table 1). Google Ea h (Google Inc., 2013)
isual inspec ion allowed he picking o sampling si es wi h no dis u bance o apid g ow h du ing he
sampling pe iod, and homogenous sh ub co e o a leas 1 km2as loca ion o NFMD si es can be o
by ens o hund eds o me e s. O he si e equi emen s we e o ha e a leas 20 Landsa -5 TM images
wi h <10% cloud co e o e a leas six yea s ha co e ed as much as possible o he phenological
cycle o he species sampled. This wo k selec ed only ield sampling da es wi hin
±
6 days om image
acquisi ion o educe he impac o LFMC empo al a ia ion be ween NFMD and Landsa -5 TM
acquisi ion. LFMC da a o he selec ed si es co e ed he whole i e season, om he beginning o
sp ing un il he end o he all, al hough in some cases sampling was ex ended yea - ound.
Table 1.
Desc ip ion o he Li e Fuel Mois u e Con en (LFMC) sampling si es and hei Landsa -5 TM
scene pa h and ow.
Si es Pa h Row La i ude
(N) Longi ude(W) Sampling Pe iod
(yyyy/mm/dd) Species #
Samples
Cla k Mo o way,
Malibu 41 36 34.0844 118.8625
2001/01/08 2011/06/22
Big-pod
buckb ush;
Chamise
65
Glendo a Rigde 41 36 34.1653 117.8650
2003/01/29 2011/10/28
Hoa ylea
ceano hus;
Chamise
55
Lau el Canyon, M
Olympus 41 36 34.1247 118.3689
2001/04/09 2011/10/28
Chamise 73
T ippe Ranch,
Topanga 41 36 34.0933 118.5978
2001/02/05 2011/10/28
Chamise 69
Peach Mo o way 41 36 34.3556 118.5347
2005/04/02 2011/10/28
Chamise 50
Place i a Canyon 41 36 34.3753 118.4389
2001/05/02 2011/10/28
Chamise 72
Remo e Sens. 2020,12, 1714 4 o 15
Table 1. Con .
Si es Pa h Row La i ude
(N) Longi ude(W) Sampling Pe iod
(yyyy/mm/dd) Species #
Samples
Kinsman 42 34 37.1981 119.4197
2001/09/20 2011/08/23
Whi elea
Manzani a;
Big-pod
buckb ush
22
Keeney 42 29 43.9133 117.1783
2000/07/17 2011/08/30
Wyoming Big
sageb ush 39
Shi ail 42 29 44.53 117.4186
2000/07/24 2011/09/16
Wyoming Big
sageb ush 41
Remo e Sens. 2020, 12, x FOR PEER REVIEW 4 o 15
Place i a
Canyon 41 36 34.3753 118.4389 2001/05/02 2011/10/28 Chamise 72
Kinsman 42 34 37.1981 119.4197 2001/09/20 2011/08/23
Whi elea
Manzani a;
Big-pod
buckb ush
22
Keeney 42 29 43.9133 117.1783 2000/07/17 2011/08/30
Wyoming Big
sageb ush 39
Shi ail 42 29 44.53 117.4186 2000/07/24 2011/09/16
Wyoming Big
sageb ush 41
Figu e 1. Landsa scenes co esponding o he s udy si es selec ed.
The Landsa Ecosys em Dis u bance Adap i e P ocessing Sys em (LEDAPS) p o ided
o ho ec i ied Landsa -5 TM su ace e lec ance da a a 30 m spa ial esolu ion h ough he Uni ed
S a es Geological Su ey (USGS) Ea h Explo e web si e (h p://ea hexplo e .usgs.go /; las
accessed 3 Ap il, 2020). Thei adiome ic calib a ion in ol ed he ans o ma ion o he digi al
numbe s o a -senso adiance, adjus men o op o a mosphe e e lec ance and a mosphe ic
co ec ion using he 6S adia i e ans e model [30]. Landsa -5 TM scenes we e clipped o he la ges
possible window con aining da a o he whole ime se ies. In addi ion, he sa elli e o e pass a ies
o each image acquisi ion. The e o e, a mask ga e a -9999 alue o all da es o he pixels ou o he
o e pass in a leas one image acquisi ion in he ime se ies o ensu e he same common pixels o all
images in each scene. Fu he p ocessing in ol ed a opog aphic co ec ion o educe he di e ences
in he ime se ies due o sun illumina ion condi ions and e ain. The Sun-Canopy-Senso Co ec ion
wi h he C pa ame e (SCS+C) no malized e lec ance (𝐿) o his opog aphic ac o [31] (Equa ion
1):
𝐿, =𝐿cos𝛼cos𝜃+𝐶
cos𝑖+𝐶 (1)
whe e 𝐿 is he e lec ance o each Landsa -5 TM band (b); 𝛼 is he e ain slope; 𝜃 is he sola
zeni h angle; 𝑖 is he incidence angle, which is he angle be ween he no mal o he g ound and he
sola zeni h; and 𝐶 is he quo ien be ween he slope and in e cep o he linea eg ession equa ion
be ween 𝐿 and cos𝑖.
This s udy also es ed Ci co [32], C-Teille [33] and smoo hed C-Teille [34] opog aphic
co ec ions, bu SCS+C wo ked bes ( esul s no shown), imp o ing he ela ionship be ween he SI
and he ield LFMC when compa ing be o e and a e co ec ing he opog aphic e ec . The SCS+C
equi ed a Digi al Ele a ion Model (DEM) o pe o m he co ec ion. The Na ional Ele a ion Da ase
Figu e 1. Landsa scenes co esponding o he s udy si es selec ed.
The Landsa Ecosys em Dis u bance Adap i e P ocessing Sys em (LEDAPS) p o ided
o ho ec i ied Landsa -5 TM su ace e lec ance da a a 30 m spa ial esolu ion h ough he Uni ed
S a es Geological Su ey (USGS) Ea h Explo e web si e (h p://ea hexplo e .usgs.go /; las accessed
3 Ap il 2020). Thei adiome ic calib a ion in ol ed he ans o ma ion o he digi al numbe s o
a -senso adiance, adjus men o op o a mosphe e e lec ance and a mosphe ic co ec ion using he
6S adia i e ans e model [
30
]. Landsa -5 TM scenes we e clipped o he la ges possible window
con aining da a o he whole ime se ies. In addi ion, he sa elli e o e pass a ies o each image
acquisi ion. The e o e, a mask ga e a -9999 alue o all da es o he pixels ou o he o e pass in
a leas one image acquisi ion in he ime se ies o ensu e he same common pixels o all images in
each scene. Fu he p ocessing in ol ed a opog aphic co ec ion o educe he di e ences in he ime
se ies due o sun illumina ion condi ions and e ain. The Sun-Canopy-Senso Co ec ion wi h he C
pa ame e (SCS+C) no malized e lec ance (Ln) o his opog aphic ac o [31] (Equa ion (1)):
Ln,b=Lb
cos αcos θ+Cb
cos i+Cb
(1)
whe e
Lb
is he e lec ance o each Landsa -5 TM band (b);
α
is he e ain slope;
θ
is he sola zeni h
angle;
i
is he incidence angle, which is he angle be ween he no mal o he g ound and he sola
zeni h; and
Cb
is he quo ien be ween he slope and in e cep o he linea eg ession equa ion be ween
Lband cos i.
This s udy also es ed Ci co [
32
], C-Teille [
33
] and smoo hed C-Teille [
34
] opog aphic co ec ions,
bu SCS+C wo ked bes ( esul s no shown), imp o ing he ela ionship be ween he SI and he ield
LFMC when compa ing be o e and a e co ec ing he opog aphic e ec . The SCS+C equi ed
a Digi al Ele a ion Model (DEM) o pe o m he co ec ion. The Na ional Ele a ion Da ase deli e ed
Remo e Sens. 2020,12, 1714 5 o 15
he DEM in g id loa o ma a app oxima ely 10 m spa ial esolu ion (h ps://www.usgs.go /co e-
science-sys ems/ngp/ nm-deli e y/; las accessed 3 Ap il 2020). DEM mosaicking, ep ojec ion o UTM
11 N WGS84 and nea es neighbo esampling o 30 m we e necessa y o ma ch each Landsa -5 TM
scene. Fu he mo e, SCS+C applied a di e en
Cb
pa ame e depending on he ege a ion s uc u e:
he baceous, sh ub and o es . The 30 m Na ional Land Co e om F y e al. [
35
] p o ided he base
map o eclassi y 71–74 and 81 classes as he baceous; 51–52 as sh ubs; 41–43 as o es ; and he es as
non-na u al ege a ion. Finally, a linea eg ession equa ion be ween
Lb
and
cos i
o all pixels in each
o he ou eclassi ied classes calcula ed Cb o each class.
2.2. Spec al Indices
LFMC p edic ion om Landsa -5 TM es ed ou SI p e iously used o e ie e LFMC (Table 2).
The NDVI ela es o LFMC only indi ec ly h ough changes in lea pigmen s. I has success ully
es ima ed LFMC, especially o g asslands [
13
,
23
]. The NDII p edic ed LFMC o e Medi e anean
en i onmen s [
14
,
23
,
36
]. NDII di ec ly ela es o LFMC h ough spec al changes occu ing in he
sho wa e in a ed (SWIR) egion (band 5 in Landsa -5 TM), because o a iabili y in he ege a ion
wa e con en . The EVI es ima ed sh ub LFMC success ully o e chapa al ege a ion wi h AVIRIS
da a [
14
,
15
,
37
]. EVI was o iginally designed o he MODIS senso based on addi ional spec al bands
han NDVI. Fu he mo e, i p o ides be e sensi i i y o high biomass while minimizing soil and
a mosphe e in luences. The VARI es ima ed LFMC o e chapa al [
15
,
22
]. VARI is indi ec ly ela ed o
LFMC h ough changes in lea pigmen s.
Table 2. Spec al Indices (SI) selec ed o e ie e LFMC.
SI Equa ion
No malized Di e ence Vege a ion Index (NDVI) [38]ρNIR−ρR
ρNIR+ρR(2)
No malized Di e ence In a ed Index (NDII) [39]ρNIR−ρSWIR
ρNIR+ρSWIR (3)
Enhanced Vege a ion Index (EVI) [40]GρNIR−ρR
ρNIR+C1∗ρR−C2∗ρB+L(4)
Visible A mosphe ically Resis an Index (VARI) [41]ρG−ρR
ρG+ρR−ρB(5)
ρB
,
ρG
,
ρR
,
ρNIR
and
ρSWIR
=blue, g een, ed, nea in a ed and sho wa e in a ed e lec ance, espec i ely;
Gis a gain ac o ; C
1
and C
2
a e he coe icien s o he ae osol esis ance e m, and Lis a soil-adjus men ac o .
These pa ame e s ha e a alue o 2.5, 6, 7.5 and 1, espec i ely.
Compu ing he elSI compensa ed di e ences among pixels in ac ional co e , species composi ion,
soil backg ound and o ien a ion among o he ac o s in o de o p edic LFMC [
25
–
28
]. elSI is calcula ed
as he di e ence be ween he SI a a speci ic ime and he minimum SI (SI
min
) in he empo al se ies o
each pixel, di ided by he di e ence be ween he maximum SI (SI
max
) in he empo al se ies and SI
min
.
Newnham e al. [
27
] highligh ed he impo ance o selec ing an app op ia e ime in e al and c i e ia
o ob ain he SI
min
and SI
max
. Acco ding o hei indings, he ime in e al should be long enough o
enable cap u ing he ull ange o spec al a ia ion, bu sho enough o a oid cap u ing a ia ion
caused by land co e changes. The sho es pe iod o he alida ion si es in Table 1was six yea s,
whe eas he longes was ele en. The ime ame co e ed he en i e phenological ege a ion cycle in all
cases. Fu he mo e, he LEDAPS Landsa -5 TM p oduc con ained a mask ha elimina ed any cloud
o cloud shadow pixel when sea ching o SI
min
and SI
max
. Finally, a simila no maliza ion p ocess
om SI o elSI con e ed he ield LFMC da a o ela i e LFMC ( elLFMC). As a esul , minimum and
maximum alues by si e compensa ed o di e ences in sampling me hods and species composi ion
ac oss si es.
2.3. Landsa TM LFMC P oduc
This esea ch applied an empi ical me hod o es ima e LFMC o elLFMC h ough linea
in e pola ion om a SI o a elSI. These linea eg ession models a e he adi ional app oach o ela e he

Remo e Sens. 2020,12, 1714 6 o 15
spec al in o ma ion de i ed om emo ely sensed da a and ield measu ed LFMC [
13
,
14
,
22
]. Since he
elEVI bes p edic ed LFMC (see esul s sec ion), i was he base o he Landsa -5 TM he baceous and
woody LFMC p oduc . Pixels classi ied as sh ubs o o es included he baceous and woody LFMC
laye s whe eas he baceous pixels included only he he baceous LFMC laye . To o e come he limi a ion
o op ical emo e sensing no measu ing he unde s o y laye , a sp ing me apho ex apola ed he
unde s o y he baceous alues o he woody pixels om he su ounding he baceous pixels using
he “inpain _nans” ool (h ps://www.ma hwo ks.com/ma labcen al/ ileexchange/4551-inpain -nans,
las accessed 3 Ap il 2020). This me hod adjus s a pa ial di e en ial equa ion o ex apola e 2-dimesional
da a wi h sp ings ha connec pixels o hei neighbo s in all di ec ions. The pe o mance o he
me hod was e alua ed by compa ing a andom ex ac ion o 3000 he baceous elEVI pixel alues om
16 Ap il 2009 o he ex apola ed esul s a hese loca ions.
In o ma ion on uel models based on he Sco and Bu gan’s [
42
] i e beha io uel models
classi ica ion e sion “LF 2014” wi h me ada a “20161031” a 30 m esolu ion came om he LANDFIRE
p og am. Fuel models classes we e g ouped in o he baceous, sh ubs o o es s acco ding o hei
main i e ca ie . The elEVI was unde s ood as elLFMC and con e ed o absolu e LFMC om
he minimum and maximum ield measu ed LFMC sampled in he Jaspe Ridge Biological P ese e,
CA(USA) using 30.0% o 197.2 % o he baceous, 36.4% o 222.1 % o sh ubs and 53.4% o 164.7% o
o es s [
15
]. O he scaling would be possible, bu Jaspe Ridge is p e e ed since ield campaigns in
Sp ing, Summe and Fall we e ca ied ou co e ing he phenological ege a ion cycle, no only o
sh ubs, bu also o he baceous ege a ion and o es . The nine chapa al si es om he NFMD had
a minimum alue o 43.5% and a maximum o 231.0% o sh ubs. Hence, using Jaspe Ridge da a
would in oduce a small bias.
2.4. Fi e Beha io Modeling wi h he LFMC P oduc
A 1000 by 1000 pixels window wi hin he Landsa -5 TM Pa h 41 and Row 36 was used o es
he di e ences in i e beha io due o LFMC. This si e loca ed No hwes o he ci y o Los Angeles
includes pa o Los Pad es Na ional Fo es (Uppe Le Co ne : 119.115W 35.024N; Lowe Righ
Co ne : 118.781W 34.759N). WFA so wa e was used o ca y ou he i e beha io simula ions [
19
].
WFA p o ides eal- ime analysis o wild i e beha io and sp ead o di ec ly suppo mul i-agency
wild i e inciden managemen [
43
]. The semi-empi ical i e sp ead model in WFA uses he Ro he mel
equa ions o model su ace and c own i e beha io [
44
]. The model es ima ed how i es sp ead unde
di e en LFMC scena ios, keeping equal all o he inpu s. This is no a common app oach in i e isk
analysis, which usually uses s ochas ic inpu s. Howe e , he goal he e was o assess he di e ence in
beha io only due o LFMC.
The simula ions calcula ed i eline in ensi y, Flame Leng h (FL) and Ra e o Sp ead (ROS) a
he 30 m pixel le el esolu ion, conside ing he maximum po en ial i e beha io in each pixel [
45
].
In addi ion, WFA p edic ed he independen sp ead o 111,559 i es wi h a du a ion o eigh hou s wi h
a 90 by 90 m igni ion poin e e y h ee pixels in all di ec ions. The hou ly bu ned a ea quan i ied he
impac o each i e simula ion on he landscape. A e unning all he i e simula ions, WFA calcula ed
he ou pu BP ha ep esen s he amoun o imes he i es eached each pixel.
Modeling i e beha io and sp ead equi ed se e al spa ial- empo al inpu s: he DEM al eady
used o he opog aphic co ec ion, Sco and Bu gan’s uel models used o gene a e he LFMC p oduc
desc ibed abo e, a cons an mode a e wind speed a 20 ee o 11 km/h a a 45
º
di ec ion om No heas
o Sou hwes as well as cons an Dead Fuel Mois u e Con en (DFMC) alues o 5%, 7% and 9% o 1,
10 and 100 h uels, espec i ely.
Simula ions conside ed ou LFMC scena ios: a cons an o e all median LFMC alue o 99% o
he baceous, 129% o sh ubs and 109% o o es om Jaspe Ridge ield da a; a cons an o e all en h
pe cen ile LFMC alue o 30% o he baceous, 55% o sh ubs and 65% o o es also om Jaspe Ridge
ield da a; and he LFMC p oduc gene a ed wi h he elEVI om Landsa -5 TM da a on 16 Ap il 2009
and 25 Oc obe 2009. The median and en h pe cen ile we e calcula ed conside ing he minimum and
Remo e Sens. 2020,12, 1714 7 o 15
maximum alues o all samples collec ed in Jaspe Ridge o each ege a ion ype [
15
]. The Landsa -5
TM da es we e selec ed since hey would ep esen wo dis inc LFMC scena ios, in sp ing and end
o he summe /beginning o au umn. No e also ha acco ding o Ro he mel’s su ace i e beha io
model, he baceous LFMC is cu ed a 30% and woody LFMC en e s do mancy a 60% [42,46].
3. Resul s
The EVI p esen ed he highes and he lowes RMSE wi h he ield measu ed LFMC o i e ou
o he nine alida ion si es (Table 3). When all si es we e conside ed oge he , a signi ican d op in
and an inc ease in RMSE was obse ed, wi h he highes alue as low as 0.44% and he lowes RMSE
as high as 33.35 % o EVI. The ela ionship be ween LFMC and SI o all si es imp o ed a e using
elLFMC and elSI, inc easing up o 0.69% and dec easing he RMSE o 19% o elEVI, which was
again he bes p edic i e a iable.
Table 3.
and Roo Mean Squa e E o (RMSE) be ween ield measu ed LFMC/ elLFMC and SI/ elSI.
All a e s a is ically signi ican (P- alue <0.001).
RMSE (%)
Si e Depen.
Va .
Indep.
Va . NDVI NDII EVI VARI NDVI NDII EVI VARI
Cla kMo o way, Malibu LFMC SI 0.85 0.77 0.89 0.65 15.39 18.29 13.07 22.05
Glendo a Ridge, Glendo a LFMC SI 0.69 0.65 0.80 0.33 19.38 20.30 16.03 25.17
Lau el Canyon LFMC SI 0.81 0.85 0.87 0.48 15.53 13.95 13.11 23.46
T ippe Ranch LFMC SI 0.84 0.72 0.77 0.73 26.33 33.80 31.39 33.57
Peach Mo o way LFMC SI 0.87 0.89 0.93 0.79 11.67 10.44 8.72 14.50
Place i a Canyon LFMC SI 0.80 0.84 0.86 0.52 20.24 18.32 17.30 28.91
Kinsman LFMC SI 0.66 0.82 0.82 0.61 17.60 13.28 13.40 18.54
Keeney LFMC SI 0.79 0.64 0.74 0.36 22.02 27.61 24.11 33.58
Shi ail LFMC SI 0.73 0.67 0.69 0.69 24.48 26.53 26.12 26.23
All si es LFMC SI 0.22 0.32 0.44 0.35 36.26 35.16 33.35 34.85
All si es elLFMC SI 0.46 0.52 0.62 0.50 0.24 0.23 0.21 0.23
All si es LFMC elSI 0.49 0.51 0.57 0.49 32.47 32.00 30.51 32.31
All si es elLFMC elSI 0.61 0.66 0.69 0.55 0.21 0.20 0.19 0.22
The elEVI was hus used o p edic he he baceous and woody LFMC om Landsa -5 TM da a
on 16 h o Ap il and on 25 h o Oc obe (Figu e 2), o analyze i e beha io h ough WFA simula ions.
The ex apola ion algo i hm o es ima e he unde s o y he baceous LFMC o he woody pixels was
es ed o e a andom se o o e s o y he baceous pixels. The compa ison be ween hei ac ual
he baceous elEVI alues and he ex apola ed ones yielded =0.94 and RMSE =7.43%.
The a e age LFMC alue based on elEVI o 16 h o Ap il 2009 was close o he cons an median
LFMC map (Figu e 2), whe eas he one o 25 h o Oc obe 2009 was close o he cons an en h
pe cen ile LFMC map. These cons an LFMC alues could be used as a baseline o compa ison. I is
e iden ha his app oach canno cap u e he spa ial a iabili y o LFMC ac oss he landscape as he
elEVI maps do. Fo example, he baceous LFMC o 16 h o Ap il on he No he n pa o Figu e 2was
gene ally below he median LFMC map, bu he Sou he n pa was abo e i . Ins ead, he baceous LFMC
o 25 h o Oc obe was gene ally like he en h pe cen ile LFMC map, excep o he Sou hwes e n pa
whe e i was highe . In he case o he woody o 16 h o Ap il, he Eas e n pa was highe han he
median, bu he Wes e n pa was lowe . In con as , woody LFMC o 25 h o Oc obe was gene ally
like he en h pe cen ile LFMC map, bu wi h andomly dis ibu ed pa ches ha ing highe LFMC
(Figu e 2).
Remo e Sens. 2020,12, 1714 8 o 15
Remo e Sens. 2020, 12, x FOR PEER REVIEW 8 o 15
Figu e 2. He baceous and woody LFMC p oduc s om a cons an median and en h pe cen ile LFMC,
and om Landsa -5 TM elEVI da a on Ap il 16 h and on Oc obe 25 h o 2009 and es ima ed i e
beha io o each scena io in e ms o ROS and FL.
The LFMC p oduc s om Figu e 2 caused di e ences in he i e simula ions pe o med on WFA
(Figu es 2, 3 and 4). All inpu a iables in he simula ions we e he same excep he he baceous and
woody LFMC laye s. Despi e his, he di e ences in i e beha io (bo h ROS and FL; Figu e 2),
a e age bu ned a ea pe i e (Figu e 3) and BP (Figu e 4) we e signi ican and inc eased wi h he ime
since igni ion. Resul s based on 16 h o Ap il and 25 h o Oc obe LFMC ell in be ween he minimum
alue o he en h pe cen ile and he maximum o he median LFMC (Figu e 3). We obse ed spa ial
di e ences in e ms o ROS and FL among he LFMC p oduc s (Figu e 1). Fo ins ance, he ROS o
16 h o Ap il was highe in he no he n pa o he s udy a ea han in he sou he n pa , whe eas his
pa e n was he in e se o o he scena ios such as 25 h o Oc obe , sugges ing he impo ance o
conside ing he spa ial a iabili y o LFMC in ope a ional en i onmen s. These esul s a e consis en
wi h he BP ou pu s (Figu e 4), gi en ha he a eas wi h highe BP alues also had highe ROS. BP
maps de i ed using he en h pe cen ile and elEVI o he 25 h o Oc obe , showed simila spa ial
dis ibu ion (Figu e 4), al hough highe BP alues whe e ob ained o he o me which esul ed in
an a e age di e ence in bu ned a ea g ea e han 30 ha 8 h a e igni ion (21% and 36% highe han
he bu ned a ea o he elEVI 25 h o Oc obe and 16 h o Ap il, espec i ely). Di e ences in he
spa ial dis ibu ion in BP maps we e mo e e iden when compa ing he elEVI maps o 16 h o Ap il
o he cons an median LFMC alue.
Figu e 2.
He baceous and woody LFMC p oduc s om a cons an median and en h pe cen ile LFMC,
and om Landsa -5 TM elEVI da a on Ap il 16 h and on Oc obe 25 h o 2009 and es ima ed i e
beha io o each scena io in e ms o ROS and FL.
The LFMC p oduc s om Figu e 2caused di e ences in he i e simula ions pe o med on WFA
(Figu es 2–4). All inpu a iables in he simula ions we e he same excep he he baceous and woody
LFMC laye s. Despi e his, he di e ences in i e beha io (bo h ROS and FL; Figu e 2), a e age bu ned
a ea pe i e (Figu e 3) and BP (Figu e 4) we e signi ican and inc eased wi h he ime since igni ion.
Resul s based on 16 h o Ap il and 25 h o Oc obe LFMC ell in be ween he minimum alue o he
en h pe cen ile and he maximum o he median LFMC (Figu e 3). We obse ed spa ial di e ences in
e ms o ROS and FL among he LFMC p oduc s (Figu e 1). Fo ins ance, he ROS o 16 h o Ap il was
highe in he no he n pa o he s udy a ea han in he sou he n pa , whe eas his pa e n was he
in e se o o he scena ios such as 25 h o Oc obe , sugges ing he impo ance o conside ing he spa ial
a iabili y o LFMC in ope a ional en i onmen s. These esul s a e consis en wi h he BP ou pu s
(Figu e 4), gi en ha he a eas wi h highe BP alues also had highe ROS. BP maps de i ed using
he en h pe cen ile and elEVI o he 25 h o Oc obe , showed simila spa ial dis ibu ion (Figu e 4),
al hough highe BP alues whe e ob ained o he o me which esul ed in an a e age di e ence in
bu ned a ea g ea e han 30 ha 8 h a e igni ion (21% and 36% highe han he bu ned a ea o he
elEVI 25 h o Oc obe and 16 h o Ap il, espec i ely). Di e ences in he spa ial dis ibu ion in BP
maps we e mo e e iden when compa ing he elEVI maps o 16 h o Ap il o he cons an median
LFMC alue.
Remo e Sens. 2020,12, 1714 9 o 15
Remo e Sens. 2020, 12, x FOR PEER REVIEW 9 o 15
Figu e 3. A e age bu ned a ea a e ime since igni ion o i e simula ions and boxplo s o he bu ned
a ea 8-hou s a e igni ion in Wild i e Analys (WFA) wi h he di e en LFMC scena ios om Figu e
2.
Figu e 4. Bu ned p obabili y (BP) maps o he i e simula ions in WFA wi h he di e en LFMC
scena ios om Figu e 2.
4. Discussion
Many o he s udies ha e es ima ed LFMC om op ical emo e sensing [11], including Landsa -
5 TM da a [23]. The no el y o he Landsa -5 TM elEVI LFMC p oduc gene a ed in his s udy is ha
i p o ides an he baceous and woody laye ha can be in eg a ed in o i e beha io models like WFA.
This in eg a ion p oduces spa ial di e ences in i e beha io (ROS, FL and i eline in ensi y) and
subsequen ly, BP and a e age bu ned a ea, compa ed o using a cons an LFMC alue (Figu e 4). The
eason o choose Landsa -5 TM da a was i s long ime se ies, coinciding wi h ield da a in he NFMD
o alida ion. Howe e , he applica ion o his app oach o Landsa -7 and -8 and Sen inel-2A and -
2B should be possible in o de o gene a e sys ema ic p oduc s a 10–30 m spa ial esolu ion. Such
p oduc s will imp o e hose om MODIS-like senso s wi h >250 m spa ial esolu ion in e ms o
ope a ional adop ion [18]. Gi en i s highe empo al esolu ion, MODIS like senso s could also be
used o gap illing [47]. Howe e , such a p oduc would equi e p ope senso in e -calib a ion o
Figu e 3.
A e age bu ned a ea a e ime since igni ion o i e simula ions and boxplo s o he bu ned
a ea 8-hou s a e igni ion in Wild i e Analys (WFA) wi h he di e en LFMC scena ios om Figu e 2.
Remo e Sens. 2020, 12, x FOR PEER REVIEW 9 o 15
Figu e 3. A e age bu ned a ea a e ime since igni ion o i e simula ions and boxplo s o he bu ned
a ea 8-hou s a e igni ion in Wild i e Analys (WFA) wi h he di e en LFMC scena ios om Figu e
2.
Figu e 4. Bu ned p obabili y (BP) maps o he i e simula ions in WFA wi h he di e en LFMC
scena ios om Figu e 2.
4. Discussion
Many o he s udies ha e es ima ed LFMC om op ical emo e sensing [11], including Landsa -
5 TM da a [23]. The no el y o he Landsa -5 TM elEVI LFMC p oduc gene a ed in his s udy is ha
i p o ides an he baceous and woody laye ha can be in eg a ed in o i e beha io models like WFA.
This in eg a ion p oduces spa ial di e ences in i e beha io (ROS, FL and i eline in ensi y) and
subsequen ly, BP and a e age bu ned a ea, compa ed o using a cons an LFMC alue (Figu e 4). The
eason o choose Landsa -5 TM da a was i s long ime se ies, coinciding wi h ield da a in he NFMD
o alida ion. Howe e , he applica ion o his app oach o Landsa -7 and -8 and Sen inel-2A and -
2B should be possible in o de o gene a e sys ema ic p oduc s a 10–30 m spa ial esolu ion. Such
p oduc s will imp o e hose om MODIS-like senso s wi h >250 m spa ial esolu ion in e ms o
ope a ional adop ion [18]. Gi en i s highe empo al esolu ion, MODIS like senso s could also be
used o gap illing [47]. Howe e , such a p oduc would equi e p ope senso in e -calib a ion o
Figu e 4.
Bu ned p obabili y (BP) maps o he i e simula ions in WFA wi h he di e en LFMC
scena ios om Figu e 2.
4. Discussion
Many o he s udies ha e es ima ed LFMC om op ical emo e sensing [
11
], including Landsa -5
TM da a [
23
]. The no el y o he Landsa -5 TM elEVI LFMC p oduc gene a ed in his s udy is
ha i p o ides an he baceous and woody laye ha can be in eg a ed in o i e beha io models like
WFA. This in eg a ion p oduces spa ial di e ences in i e beha io (ROS, FL and i eline in ensi y)
and subsequen ly, BP and a e age bu ned a ea, compa ed o using a cons an LFMC alue (Figu e 4).
The eason o choose Landsa -5 TM da a was i s long ime se ies, coinciding wi h ield da a in he
NFMD o alida ion. Howe e , he applica ion o his app oach o Landsa -7 and -8 and Sen inel-2A
and -2B should be possible in o de o gene a e sys ema ic p oduc s a 10–30 m spa ial esolu ion.
Such p oduc s will imp o e hose om MODIS-like senso s wi h >250 m spa ial esolu ion in e ms o
ope a ional adop ion [
18
]. Gi en i s highe empo al esolu ion, MODIS like senso s could also be
used o gap illing [
47
]. Howe e , such a p oduc would equi e p ope senso in e -calib a ion o
de e mine he spec al band adjus men ac o s. This should no only be done o e pseudo-in a ian