emo e sensing
A icle
Remo ely Sensed Va iables o Ecosys em Func ioning
Suppo Robus P edic ions o Abundance Pa e ns
o Ra e Species
Sal ado A enas-Cas o 1,2,* , Ad ián Regos 1,3 , João F. Gonçal es 1,
Domingo Alca az-Segu a 4,5 and João Hon ado 1,6
1CIBIO-InBIO—Cen o de In es igação em Biodi e sidade e Recu sos Gené icos, Labo a ó io Associado,
Uni e sidade do Po o, Campus Ag á io Vai ão, 4485-661 Vila do Conde, Po ugal
2
CICGE—Cen o de In es igaç
ã
o em Ci
ê
ncias Geo-Espaciais, Faculdade de Ci
ê
ncias, Uni e sidade do Po o,
Obse a ó io As onómico “P o . Manuel de Ba os”, Alameda do Mon e da Vi gem,
4430-146 Vila No a de Gaia, Po ugal
3Depa amen o de Zooloxía, Xené ica e An opoloxía Física, Uni e sidade de San iago de Compos ela,
15782 San iago de Compos ela, Spain
4Depa men o Bo any and In e -Uni e si y Ins i u e o Ea h Sys em Resea ch, Uni e si y o G anada,
18071 G anada, Spain
5
Andalusian Cen e o he Assessmen and Moni o ing o Global Change (CAESCG), Uni e si y o Alme
í
a,
04151 Alme ía, Spain
6Faculdade de Ciências, Uni e sidade do Po o, 4169-007 Po o, Po ugal
*Co espondence: sa [email p o ec ed]
Recei ed: 7 Augus 2019; Accep ed: 4 Sep embe 2019; Published: 6 Sep embe 2019
Abs ac :
Global en i onmen al changes a e a ec ing bo h he dis ibu ion and abundance o species
a an unp eceden ed a e. To assess hese e ec s, species dis ibu ion models (SDMs) ha e been g ea ly
de eloped o e he las decades, while species abundance models (SAMs) ha e gene ally ecei ed
less a en ion e en hough hese models p o ide essen ial in o ma ion o conse a ion managemen .
Wi h popula ion abundance de ined as an essen ial biodi e si y a iable (EBV), SAMs could o e
spa ially explici p edic ions o species abundance ac oss space and ime. Sa elli e-de i ed ecosys em
unc ioning a ibu es (EFAs) a e known o in o m on p ocesses con olling species dis ibu ion, bu
hey ha e no been es ed as p edic o s o species abundance. In his s udy, we assessed he use ulness
o SAMs calib a ed wi h EFAs (as p ocess- ela ed a iables) o p edic local abundance pa e ns o a
a e and h ea ened species ( he na ow Ibe ian endemic ‘Ge
ê
s lily’ I is boissie i; p o ec ed unde he
Eu opean Union Habi a s Di ec i e), and o p ojec in e -annual luc ua ions o p edic ed abundance.
We compa ed he p edic i e accu acy o SAMs calib a ed wi h clima e (CLI), opog aphy (DEM), land
co e (LCC), EFAs, and combina ions o hese. Models i ed only wi h EFAs explained he g ea es
a iance in species abundance, compa ed o models based only on CLI, DEM, o LCC a iables.
The combina ion o EFAs and opog aphy sligh ly inc eased model pe o mance. P edic ions o he
in e -annual dynamics o species abundance we e ela ed o in e -annual luc ua ions in clima e,
which holds impo an implica ions o acking global change e ec s on species abundance. This
s udy unde lines he po en ial o EFAs as obus p edic o s o biodi e si y change h ough popula ion
size ends. The combina ion o EFA-based SAMs and SDMs would p o ide an essen ial oolki o
species moni o ing p og ams.
Keywo ds:
ecosys em unc ioning a ibu es (EFAs); essen ial biodi e si y a iables (EBVs); I is
boissie i; a e species; sa elli e emo e sensing; species abundance models (SAMs); species dis ibu ion
models (SDMs)
Remo e Sens. 2019,11, 2086; doi:10.3390/ s11182086 www.mdpi.com/jou nal/ emo esensing
Remo e Sens. 2019,11, 2086 2 o 16
1. In oduc ion
Human-induced global en i onmen al change is h ea ening biodi e si y globally a an
unp eceden ed a e [
1
]. These en i onmen al changes a e expec ed o g ea ly impac bo h he
dis ibu ion and he abundance o species. In ha sense, bo h ‘species dis ibu ion’ and ‘popula ion
abundance’ a e po en ial candida es o be conside ed as essen ial biodi e si y a iables (EBVs) in
he EBV class ‘species popula ions’ [
2
,
3
]. P edic ions om species abundance models (SAMs) can
deli e ele an in o ma ion bo h on species dis ibu ions and on popula ion sizes o biodi e si y
moni o ing and conse a ion managemen [
4
]. The in eg a ion o modeling ools and biodi e si y
moni o ing p o ocols has inc eased ou capaci y o quan i y and an icipa e he esponse o species o
en i onmen al change ac oss scales [
5
–
7
]. Model-assis ed moni o ing o dis ibu ion and abundances
is he e o e essen ial o assess he e ec i eness o conse a ion policies globally [8–10].
Howe e , while species dis ibu ion models (SDMs) ha e been g ea ly de eloped o e he las
decades [
11
,
12
] (and e e ences he ein), SAMs ha e gene ally ecei ed less a en ion [
13
–
15
], bu
see [
16
] and [
17
,
18
]. Mo eo e , hese modeling echniques ha e been adi ionally based on clima e
and s a ic habi a ea u es [
19
], ailing o conside o he en i onmen al in o ma ion ha is ecologically
ele an o species. Thus, he de elopmen and applica ion o comp ehensi e en i onmen al da ase s
and e icien modeling echniques a e needed o p edic and o moni o he geog aphic dis ibu ion o
species and abundance, along wi h hei ela ed en i onmen al d i e s [
20
]. In his ega d, sa elli e
emo e sensing ime-se ies o e con inuous and cos -e ec i e measu es o he en i onmen ac oss space
and ime [
21
], allowing nea eal- ime assessmen o species dis ibu ion and popula ion dynamics [
22
].
This is especially ele an o a e and endange ed species ha o en hold high habi a speci ici y,
na ow niche b ead h, limi ed dispe sal abili y, and inc easing habi a agmen a ion.
As desc ip o s o exchange o ma e and ene gy (i.e., ecosys em unc ioning [
23
,
24
]), and he e o e
p ocesses ha a e key o species and hei popula ions, sa elli e-de i ed ecosys em unc ioning
a ibu es (EFAs) o e an in eg a i e and ea lie iew o ecosys em esponses o en i onmen al changes
han clima e and landscape s uc u al o composi ional a ibu es [
25
]. Sa elli e-de i ed EFAs can,
he e o e, be used as p ocess- ela ed p edic o s in biodi e si y models, since hey p o ide in o ma ion
abou ecosys em p ocesses and p ope ies ela ed o he ca bon (e.g., ege a ion g eenness) and wa e
cycles (e.g., e apo anspi a ion) o wi h ene gy balance (e.g., land su ace empe a u e and albedo).
In ac , sa elli e-de i ed EFAs a e being also es ed as candida e EBVs ela ed o he ca bon cycle,
ene gy, and adia ion balance capable o in o ming abou ecosys em componen s linked o species
conse a ion s a us [
26
,
27
]. Thus, he inco po a ion o EFAs in o SDMs was ound o inc ease hei
p edic i e powe and ans e abili y [
28
], o e ing he oppo uni y o cos -e ec i ely moni o mul iple
endange ed species [
27
,
29
], a di e en spa ial and empo al scales [
30
]. S ill, despi e hese ad an ages,
he p edic i e abili y o EFAs in abundance models as well as o assess in e -annual popula ion
dynamics o a e species emains la gely un es ed.
The main goal o his s udy is o assess he po en ial o SAMs calib a ed wi h sa elli e-de i ed
EFAs (as p ocess- ela ed model p edic o s) o p edic spa ial pa e ns o abundance and o suppo
p edic ions o he in e -annual abundance dynamics o a e plan s, using as es species he na ow
Ibe ian endemic ‘Ge
ê
s lily’ (I is boissie i; p o ec ed unde he EU Habi a s Di ec i e). Fu he mo e,
we assess whe he EFAs (as EBVs om he Ecosys em unc ion class [
2
]) also hold he po en ial o
in o m on species abundance dynamics (Species popula ions EBV class) o biodi e si y moni o ing
and epo ing [
4
,
26
]. Sequen ially, we: 1) assessed he abili y o sa elli e-de i ed EFAs o suppo
eliable p edic ions o I is abundance; 2) analyzed he added- alue o EFAs o eplace o complemen
‘ adi ional’ clima e, land co e o opog aphic da a in species abundance models; 3) p ojec ed ( h ough
back- and o e-cas ing) he bes model o ob ain yea ly abundance p edic ions; and 4) compa ed hose
p edic ed abundances wi h dis u bance ac o s ( i e, clima e anomalies) hypo hesized o explain he
p edic ed in e -annual abundance a ia ions. The amewo k p oposed he e is lexible enough o
be applicable o o he species and socio-en i onmen al con ex s globally, and i is expec ed o assis
conse a ion decision-making p ocesses ela ed o biodi e si y moni o ing and epo ing schemes.
Remo e Sens. 2019,11, 2086 3 o 16
2. Ma e ials and Me hods
2.1. Tes Species and S udy A ea
We es ed ou app oach wi h he ‘Ge
ê
s lily’ (I is boissie i Hen iq. =Xiphion boissie i (Hen iq.)
Rodion), a na ow- anged endemic plan holding a ‘c i ically endange ed’ conse a ion s a us [
31
],
and p o ec ed unde he Eu opean Habi a s Di ec i e (he ea e HD) o which EU membe -s a es hold
egula epo ing obliga ions (unde A icle 17 o he HD/Annex II and IV) [
32
]. This species is es ic ed
o moun ainous a eas o he no hwes Ibe ian Peninsula (Figu e 1a), whe e Po ugal concen a es
he la ges popula ions o he species, especially in he Peneda-Ge
ê
s Na ional Pa k (Figu e 1b). This
moun ain-p o ec ed a ea ep esen s a ansi ional a ea be ween he Medi e anean and Eu o-Sibe ian
and Alpine biogeog aphic egions and a mosaic o mixed ege a ion ypes cha ac e izes i . A e age
annual empe a u es ange om 17
◦
C o 20
◦
C, wi h some a ia ion along he al i udinal g adien [
33
].
Highlands ha e an a e age empe a u e o abou 10
◦
C, anging om 4 o 14
◦
C, while a eas in he
alleys ha e milde clima es wi h an a e age empe a u e o 14
◦
C, anging om 8 o 20
◦
C. To al mean
ain all eaches 3000 mm pe yea (wi h mo e han 130 ainy days pe yea ) and snow all is equen in
he moun ain ops.
Remo e Sens. 2019, 11, x FOR PEER REVIEW 3 o 18
2. Ma e ials and Me hods
2.1. Tes Species and S udy A ea
We es ed ou app oach wi h he ‘Ge ês lily’ (I is boissie i Hen iq. = Xiphion boissie i (Hen iq.)
Rodion), a na ow- anged endemic plan holding a ‘c i ically endange ed’ conse a ion s a us [31],
and p o ec ed unde he Eu opean Habi a s Di ec i e (he ea e HD) o which EU membe -s a es
hold egula epo ing obliga ions (unde A icle 17 o he HD/Annex II and IV) [32]. This species is
es ic ed o moun ainous a eas o he no hwes Ibe ian Peninsula (Figu e 1a), whe e Po ugal
concen a es he la ges popula ions o he species, especially in he Peneda-Ge ês Na ional Pa k
(Figu e 1b). This moun ain-p o ec ed a ea ep esen s a ansi ional a ea be ween he Medi e anean
and Eu o-Sibe ian and Alpine biogeog aphic egions and a mosaic o mixed ege a ion ypes
cha ac e izes i . A e age annual empe a u es ange om 17 °C o 20 °C, wi h some a ia ion along
he al i udinal g adien [33]. Highlands ha e an a e age empe a u e o abou 10 °C, anging om 4
o 14 °C, while a eas in he alleys ha e milde clima es wi h an a e age empe a u e o 14 °C, anging
om 8 o 20 °C. To al mean ain all eaches 3000 mm pe yea (wi h mo e han 130 ainy days pe
yea ) and snow all is equen in he moun ain ops.
Figu e 1. Loca ion o he s udy a ea ( he Peneda-Ge ês Na ional Pa k) in he no hwes Ibe ian
Peninsula (a). Dis ibu ion o he sampled uni s o I is boissie i in 2006 o each g id uni (250 × 250 m)
in he s udy a ea (b).
Figu e 1.
Loca ion o he s udy a ea ( he Peneda-Ge
ê
s Na ional Pa k) in he no hwes Ibe ian
Peninsula (
a
). Dis ibu ion o he sampled uni s o I is boissie i in 2006 o each g id uni (250
×
250 m)
in he s udy a ea (b).
The highes ocky ou c ops wi h s eep opog aphy and low accessibili y a e he main habi a o
he ‘Ge
ê
s lily’ [
31
]. The species ypically occu s in open sc ublands and c e ices o g ani e ou c ops a
ele a ions be ween 500 and 1500 m abo e sea le el. Many o hese habi a a eas we e main ained by
he adi ional g azing sys em, which in ol ed equen small-scale bu ning o pas u e main enance
Remo e Sens. 2019,11, 2086 4 o 16
and was he e o e an impo an con ol o ege a ion enc oachmen . This long- e m land managemen
sys em has been declining o decades due o se e e u al depopula ion and o he socioeconomic
changes. As a esul , widesp ead ege a ion enc oachmen has changed habi a condi ions as well as
he local i e egime, po en ially a ec ing he conse a ion o I is boissie i and o he plan species o
high conse a ion conce n.
2.2. Species Abundance Da ase
The o iginal da ase consis ed o abundance es ima es (coun s o he numbe o indi iduals),
pe o med in si u in he Peneda-Ge
ê
s Na ional Pa k (Po ugal) in 2006 by Pa k bo anis s. The ield
su ey was conduc ed by isi ing all p e iously known I is occu ence si es as well as he neighbo ing
a eas. I is clus e s sepa a ed om each o he by mo e han 100 me e s we e conside ed as dis inc local
popula ions o pe o m he coun s.
To ma ch abundance da a o he EFAs esolu ion (ca. 232 m MODIS pixel size; see below),
abundance alues we e agg ega ed o each cell ollowing an a ea-p opo ional alloca ion ule, i.e., he
o al numbe o plan s wi hin each sampling a ea (i.e., a local popula ion de ined by a polygon) was
spli be ween con iguous g id cells depending on he p opo ionally occupied a ea by he polygon in
each cell. The inal abundance da ase consis ed o 61 cells (o he 232 m g id; Figu e 1b), wi h he
numbe o I is plan s pe cell anging om 1 o 285.
2.3. Modeling F amewo k
To assess he p edic i e abili y o sa elli e-de i ed EFAs in abundance models, we compa ed
SAMs based on EFAs agains SAMs based on clima ic, opog aphic, o landscape a iables. The
bes p edic o s we e hen combined o es whe he mo e obus SAMs (and hus be e abundance
p edic ions) could be ob ained.
2.3.1. En i onmen al P edic o s
We selec ed candida e p edic i e a iables exp essing en i onmen al and ecological ac o s ha
a e known o in luence plan physiology, dis ibu ion, and local habi a p e e ences o e a pe iod
(de ails below) as close as possible o he da e o he in-si u abundance su ey (2006).
Based on a mon hly ime se ies o ai empe a u e (mean maximum and mean minimum
empe a u e) and o al p ecipi a ion—which we e calcula ed om me eo ological s a ions in Spain and
Po ugal by using mul iple eg ession echniques (see de ails in [
34
])—we de i ed 19 (bio-) clima e
a iables a 200 m spa ial esolu ion o each yea om 2000 o 2010 by using he unc ion ‘bio a s’
a ailable in he R package ‘dismo’, e sion 1.1.
Land use/co e da a was ob ained by a hyb id classi ica ion p ocedu e o Landsa -7 ETM +op ical
and he mal image y (19 May 2010 and 30 July 2010; 30 m esolu ion), which combines unsupe ised
and supe ised echniques (see de ails in [
28
]). F om he classi ied land co e /use maps, we ob ained
he ac ional co e o o es , sh ubland, and ocky ou c op ypes o use as landscape composi ion
a iables a a spa ial esolu ion o 232 m.
F om he digi al ele a ion model (DEM) o e Eu ope (EU-DEM based on SRTM and ASTER
GDEM; h ps://www.eea.eu opa.eu/da a-and-maps/da a/eu-dem) p oduced a 30 m esolu ion, we
de i ed opog aphic a iables a 232 m g id size by using he unc ion ‘ e ain’ a ailable in he R
package ‘ as e ’, e sion 2.8–19.
To compu e EFAs, we used he MODIS Enhanced Vege a ion Index (EVI) (MOD13Q1. 006; 232 m
pixel e e y 16 days) o he 2001–2016 pe iod. Fo ha , we used Google Ea h Engine [
35
] o de i e he
ollowing eigh summa y me ics o he EVI seasonal dynamics: EVI annual mean (su oga e o annual
p ima y p oduc ion), EVI annual maximum and minimum (indica o s o he annual ex emes), EVI
seasonal s anda d-de ia ion (desc ip o o seasonali y), and sine and cosine o he da es o maximum
and minimum EVI [25,36] (indica o s o phenology).
Remo e Sens. 2019,11, 2086 5 o 16
All en i onmen al a iables we e p ocessed o ma ch he pixel size o he EFAs esolu ion
(ca. 232 m g id size) a he geog aphical coo dina e sys em WGS84, using he R package ‘ as e ’ ( e sion
2.8–19). The ini ial da ase included 44 candida e p edic o s (Table S1—Suppo ing In o ma ion).
To a oid including highly co ela ed a iables in model i ing, we conduc ed a mul icollinea i y
analysis by es ing Pea son pai -wise co ela ions and Va iance In la ion Fac o s (VIF). Based on hese
mul icollinea i y analyses, we inally e ained h ee independen p edic o s o each compe ing model
(wi h Pea son’s co ela ion coe icien s o <0.7 and VIF o <4; Table S2; Figu e S1).
2.3.2. Model Fi ing
Fo each model, he esponse a iable (I is abundance pe g id cell; 61 eco ds) was ela ed o
p edic o a iables using gene alized linea models (GLMs) calib a ed wi h h ee p edic o s. GLMs
a e lexible s a is ical me hods o analyzing ecological ela ionships since hey ake in o accoun
he in e ac i e beha io o a iables ha can be poo ly ep esen ed by classical Gaussian (linea )
dis ibu ions [
37
]. Fu he mo e, ecen li e a u e ecommends using models based on Poisson and
nega i e binomial dis ibu ions ins ead o log- ans o ming he da a [
38
] wi h he main ad an age ha
he pa ame e s a e es ima ed on he o iginal scale and hus back- ans o ma ions a e a oided [
39
]. Since
his is an inc easingly-used app oach o coun (abundance) da a, we ela ed I is abundance alues o
p edic o s by compu ing a nega i e binomial dis ibu ion o e o s (a e checking o o e dispe sion
e ec s in Poisson models and obus es ima ion ia weigh ed likelihood), and applying a loga i hmic
link unc ion by using he unc ion ‘glm.nb’ a ailable in he R package ‘MASS’, e sion 7.3–51.1. Based
on a mul i-model in e ence (MMI) amewo k [
40
], we es ed and anked he indi idual compe ing
models i ed wi h h ee a iables o abundance p edic ions pe o mance (Table S2).
We calib a ed i e GLMs as ollows: (1) models based on p ocess- ela ed (i.e., ecosys em
unc ioning) p edic o s (‘EFA’)—using only emo ely sensed EFAs ( hei in e -annual mean o he
2001–2006 pe iod); (2) clima e-based models (‘CLI’)—using a iables ex ac ed om he mon hly
clima e da ase ( he in e -annual mean o he 2000–2010 pe iod); (3) models based on land co e
(landscape composi ion) p edic o s (‘LCC’)— om land co e maps (yea 2010); (4) models based on
opog aphic a iables (‘DEM’); and (5) a bes model, combining he mos signi ican h ee p edic o s
selec ed om all possible combina ions o p e ious g oups. The Akaike In o ma ion C i e ion wi h a
co ec ion o ini e sample size (AICc) was used o ank he compe ing models [
41
]. Complemen a ily,
we used he explained de iance and Spea man’s co ela ion be ween obse ed and p edic ed alues o
assess model i ing. Models ha we e no signi ican , o wi h low pe o mance in AICc o in explained
de iance we e excluded om he analysis. We also es ed he p edic i e accu acy o each model by
calcula ing he mean absolu e scaled e o (MASE) [
42
]. A MASE highe han 1 implies ha he ac ual
o ecas is wo se han a nai e o ecas as a e e ence me hod calcula ed as a sample, while a MASE
lowe han 1 implies ha he cu en o ecas beha io is be e han he nai e me hod. Fo ha , we
used he ´accu acy´ unc ion h ough he R package ´ o ecas ´ ( e sion 8.9).
2.4. In e -Annual P edic ions o I is Abundance
The bes model was backcas ed and o ecas ed (i.e., p ojec ed) o he en i onmen al condi ions
o each yea o ob ain a ime-se ies o annual abundance p edic ions. Speci ically, we used he
mos pa simonious and wi h g ea e p edic i e accu acy (‘bes ’) SAM ( he one calib a ed wi h he
in e -annual (2001–2006) a e age alues o EFAs plus opog aphy; see Resul s) o ob ain annual
backwa d ( om 2005 o 2001) and o wa d ( om 2007 o 2016) yea ly p ojec ions. We also calcula ed
he in e -annual mean (
¯
X
), he in e -annual s anda d de ia ion (SD), and he in e -annual coe icien o
a ia ion (CV; %) o p edic ed abundance a he g id cell le el o he en i e 2001–2016 ime-pe iod as a
spa io empo al p oxy o he in e -annual a iabili y o I is popula ions.
Finally, we explo ed possible ela ions be ween he in e -annual a iabili y o species p edic ed
abundance and he occu ence o dis u bances, namely hose ela ed o i e and clima e, by applying
quan ile eg ession analysis. Unlike o dina y leas squa e eg ession, quan ile eg ession is a
Remo e Sens. 2019,11, 2086 6 o 16
dis ibu ion-agnos ic me hodology, which makes no assump ions abou he dis ibu ion o he esiduals,
and he e o e no assume a pa ame ic dis ibu ion o he esponse [
43
]. Thus, we explo ed di e en
aspec s o he ela ionship (as signals and e ec s) be ween he dependen a iable (p edic ed abundance)
and he independen a iables (clima e a iables and anomalies). To do so, we calcula ed he o al
bu ned a ea pe yea om o icial wild i e maps based on Landsa -7 ETM+images [
44
] o he
2001–2016 pe iod. Clima ic means and annual anomalies we e de i ed om Te aClima e [
45
] a 4 km
esolu ion o he same pe iod: o al annual p ecipi a ion (mm
·
yea
−1
, PpT); a e age annual minimum
empe a u e (
◦
C; Tmin); and a e age annual maximum empe a u e (
◦
C; Tmax). Since clima e da a, in
gene al, ha e much mo e spa ial au oco ela ion, we esampled hese clima ic a iables o ca. 232 m.
The I is abundance alues showed a igh -skewed dis ibu ion, and a p e alence o low abundance
alues makes i di icul o ex ac ela ions be ween popula ion size and p edic o s in abundance
end analysis (he e oskedas ic esponse [
46
,
47
]). To check he ela ionships and signals be ween he
esponse a iable (p edic ed abundance) and independen a iables (clima e and i s anomalies), we
applied quan ile eg ession es s (‘quan eg’ R package) i ed o he median (quan ile =0.5). To es
he goodness-o - i o quan ile eg ession models, we used he pseudo-R
2
measu e sugges ed by [
48
].
Fu he mo e, we es ed he same associa ions be ween in e -annual p edic ed abundances and i e
and clima e a iables, bu o di e en le els o popula ion size (quan iles). Fo ha , we applied he
same quan ile eg ession es whe e abundances we e sepa a ed by 5% in e als (i.e., in quan iles) in
o de o o e come limi a ions ela ed o he igh -skewed dis ibu ion and he e oskedas ic esponse o
abundance da a. To check signi ican di e ences ac oss quan iles, we used he es o equali y o slopes
by using he ´ANOVA´ unc ion implemen ed by ‘quan eg’ de elope s.
All analyses we e conduc ed using he R en i onmen o s a is ical compu a ion [
49
]. QGIS
e sion 2.18.11 and A cGIS e sion 10.2 we e used o managing and ep esen ing spa ial da a
and p ojec ions.
3. Resul s
3.1. Ranking o Models and P edic o s
O e all, models ha include EFAs as p edic o s showed he highes pe o mance (explained
de iance up o 0.3), whe eas he LCC-based model a ained he lowes pe o mance (Table 1).
Table 1.
Resul s om he gene alized linea models (GLM) model selec ion and mul i-model in e ence
explaining local abundance pa e ns o I is boissie i. The compe ing models a e lis ed in descending
o de om he bes o leas i ed model de e mined by hei mean absolu e scaled e o (MASE)
alues. Fo compa ison, a baseline ‘null model’ con aining a single in e cep e m is used. Loglik:
loglikelihood. AICc: Akaike In o ma ion C i e ion alue.
∆
AICc: ep esen s he di e ences be ween
he AICc alues o he bes model conside ed and o he compe ing models. wi: ep esen s he Akaike
weigh s and indica es he p obabili y ha a pa icula model is he bes among hose conside ed.
Explained de iance: he p opo ion o he explained de iance by he model. Spea man Co ela ion:
Spea man co ela ion be ween obse ed and p edic ed alues. MASE: Mean absolu e scaled e o as a
measu e o he p edic i e accu acy, and anges om 0 =highes accu acy, o 1 =lowes accu acy.
Compe ing
Model P edic o s LogLik AICc ∆AIC wi Explained
De iance
Spea man
Co ela ion MASE
EFAs +DEM
- EVImin
- EVIdmic
- SLO
−239.68 487.95 0.00 0.79 0.33 0.44 0.73
EFAs +CLI
- EVImin
- EVIdmic
- BIO5
−240.36 490.71 2.14 0.19 0.32 0.43 0.75
EFAs
- EVImin
- EVImn
- EVIdmic
−240.87 492.85 2.76 0.20 0.31 0.43 0.77
Remo e Sens. 2019,11, 2086 7 o 16
Table 1. Con .
Compe ing
Model P edic o s LogLik AICc ∆AIC wi Explained
De iance
Spea man
Co ela ion MASE
DEM
- ELE
- SLO
- ASP
−243.89 498.52 8.43 0.01 0.24 0.34 0.90
CLI
- BIO5
- BIO15
- BIO19
−247.62 505.97 15.88 0.00 0.15 0.31 0.96
LCC
- DeFo
- OSh
- RoA
−248.77 508.28 18.19 0.00 0.12 0.09 1
Null model - −253.38 510.97 20.88 0.00 0.00 - 1
The mos pa simonious model (based on AICc) and wi h he g ea es p edic i e abili y (based on
MASE) was ob ained when combining EFAs and opog aphic a iables (EFAs +DEM model; Table 1).
The co ela ion be ween species abundance and p edic o a iables in his model (EVIdmic, EVImin,
and SLO) was nega i e o he wo EFAs and posi i e o SLO (Spea man co ela ions we e
−
0.13,
−
0.37,
and 0.10, espec i ely); howe e , only EVImin showed a signi ican co ela ion wi h I is abundance
(Figu e 2).
Remo e Sens. 2019, 11, x FOR PEER REVIEW 8 o 18
Figu e 2. Rela ion be ween obse ed abundance (numbe o indi iduals) o I is boissie i and each o
he p edic i e a iables included in he mos pa simonious species abundance model (SAM)
(Ecosys em unc ioning a ibu es (EFAs) + digi al ele a ion model (DEM)); (a) cosine o he da e o
minimum enhanced ege a ion index (EVI) (EVIdmic as minimum ege a ion g eenness); (b) EVI
annual minimum (EVImin as indica o o minimum pho osyn he ic ac i i y); (c) Slope (SLO as an
indica o o e ain la ness). A signi ican Spea man co ela ion (R) was only ound o he
ela ionship be ween obse ed abundance and EVImin.
3.2. In e -Annual Va iabili y o I is Abundance
On a e age (conside ing all pixels in he s udy a ea), he median and median absolu e de ia ion
(MAD) o abundance p edic ions based on EFAs p ojec ions o e he ull 2001–2016 ime-pe iod we e
o 17.43 (12 o obse ed alues) and 12.04 (14.08) indi iduals, espec i ely, anging om 12.79 (11.96
MAD) in yea 2006 o 23.85 (11.84 MAD) in yea 2007 (Figu e S2). O e all, he in e -annual abundance
p edic ions we e spa ially consis en wi h he known loca ions o I is (Figu e S3).
The in e -annual mean o he p edic ed abundances o he 61 obse ed sampled cells based on
he p edic ed spa ial X
, SD and CV we e 25.02, 8.06 and 33.06 indi iduals, espec i ely. Spea man
co ela ion analysis showed a high and posi i e signi ican co ela ion be ween he X
and SD ( =
0.83, p < 0.05) (Figu e S4). The e was also a high co ela ion o X wi h slope ( = 0.81, p < 0.05) and
wi h EVImin ( = -0.33, p < 0.05), bu no co ela ion wi h EVIdmic. The CV showed no co ela ion
wi h slope ( = -0.07, p < 0.05), bu a co ela ion wi h EVImin ( = 0.60, p < 0.05) and EVIdmic ( = -
0.23, p < 0.05). Those g id cells ha combine highe alues o he X
, and lowe alues o he CV,
he e o e ep esen hose a eas whe e condi ions a e be e o main ain la ge popula ions (Figu es 3a,
b).
Figu e 2.
Rela ion be ween obse ed abundance (numbe o indi iduals) o I is boissie i and each o he
p edic i e a iables included in he mos pa simonious species abundance model (SAM) (Ecosys em
unc ioning a ibu es (EFAs) +digi al ele a ion model (DEM)); (
a
) cosine o he da e o minimum
enhanced ege a ion index (EVI) (EVIdmic as minimum ege a ion g eenness); (
b
) EVI annual minimum
(EVImin as indica o o minimum pho osyn he ic ac i i y); (
c
) Slope (SLO as an indica o o e ain
la ness). A signi ican Spea man co ela ion (R) was only ound o he ela ionship be ween obse ed
abundance and EVImin.
3.2. In e -Annual Va iabili y o I is Abundance
On a e age (conside ing all pixels in he s udy a ea), he median and median absolu e de ia ion
(MAD) o abundance p edic ions based on EFAs p ojec ions o e he ull 2001–2016 ime-pe iod we e
Remo e Sens. 2019,11, 2086 8 o 16
o 17.43 (12 o obse ed alues) and 12.04 (14.08) indi iduals, espec i ely, anging om 12.79 (11.96
MAD) in yea 2006 o 23.85 (11.84 MAD) in yea 2007 (Figu e S2). O e all, he in e -annual abundance
p edic ions we e spa ially consis en wi h he known loca ions o I is (Figu e S3).
The in e -annual mean o he p edic ed abundances o he 61 obse ed sampled cells based on
he p edic ed spa ial
¯
X
, SD and CV we e 25.02, 8.06 and 33.06 indi iduals, espec i ely. Spea man
co ela ion analysis showed a high and posi i e signi ican co ela ion be ween he
¯
X
and SD ( =0.83,
p<0.05) (Figu e S4). The e was also a high co ela ion o
¯
X
wi h slope ( =0.81, p <0.05) and wi h
EVImin ( =
−
0.33, p <0.05), bu no co ela ion wi h EVIdmic. The CV showed no co ela ion wi h
slope ( =
−
0.07, p <0.05), bu a co ela ion wi h EVImin ( =0.60, p <0.05) and EVIdmic ( =
−
0.23,
p<0.05). Those g id cells ha combine highe alues o he
¯
X
, and lowe alues o he CV, he e o e
ep esen hose a eas whe e condi ions a e be e o main ain la ge popula ions (Figu e 3a,b).
Remo e Sens. 2019, 11, x FOR PEER REVIEW 9 o 18
Figu e 3. Spa ial p ojec ions o p edic ed I is boissie i abundance ep esen ing he a iabili y o he
species popula ions based on: (a) in e -annual mean; and (b) in e -annual coe icien o a ia ion (CV)
o he whole 2001–2016 ime-pe iod based on he EFAs + SLO model. Blank cells ep esen alues <
30.
The o e all mean o he s udy a ea o he annual p edic ed abundance ma ched he indica o o
minimum pho osyn he ic ac i i y (EVImin) (Figu e 4a), and was highe a e ainy yea s ( o al
a e age annual p ecipi a ion; PpT), and mainly ma ched chilly yea s (lowe alues o maximum
(Tmax) and minimum (Tmin) annual empe a u e) (Figu e 4b). Mo e speci ically, clima e condi ions
o he yea wi h he lowes p edic ed abundance (16.79; 2002) we e PpT = 1724.97 mm.yea
-1
, Tmax =
15.69 °C, and Tmin = 7.38 °C, while o he yea wi h he la ges p edic ed abundance (23.85; 2007)
we e PpT = 1087.49 mm yea -1, Tmax = 15.88 °C, and Tmin = 7.21 °C. Howe e , wild i es ( o al bu n
a ea; TBA) did no ha e a signi ican e ec on p edic ed I is abundances (R
1
(pseudoR
2
) = 0.025; Figu e
4c; Figu e S5).
Figu e 3.
Spa ial p ojec ions o p edic ed I is boissie i abundance ep esen ing he a iabili y o he
species popula ions based on: (
a
) in e -annual mean; and (
b
) in e -annual coe icien o a ia ion (CV)
o he whole 2001–2016 ime-pe iod based on he EFAs +SLO model. Blank cells ep esen alues <30.
The o e all mean o he s udy a ea o he annual p edic ed abundance ma ched he indica o o
minimum pho osyn he ic ac i i y (EVImin) (Figu e 4a), and was highe a e ainy yea s ( o al a e age
annual p ecipi a ion; PpT), and mainly ma ched chilly yea s (lowe alues o maximum (Tmax) and
minimum (Tmin) annual empe a u e) (Figu e 4b). Mo e speci ically, clima e condi ions o he yea
wi h he lowes p edic ed abundance (16.79; 2002) we e PpT =1724.97 mm
·
yea
−1
, Tmax =15.69
◦
C,
and Tmin =7.38
◦
C, while o he yea wi h he la ges p edic ed abundance (23.85; 2007) we e
PpT =1087.49 mm yea -1, Tmax =15.88
◦
C, and Tmin =7.21
◦
C. Howe e , wild i es ( o al bu n a ea;
TBA) did no ha e a signi ican e ec on p edic ed I is abundances (R
1
(pseudoR
2
)=0.025; Figu e 4c;
Figu e S5).
The quan ile eg ession analysis be ween a e age annual p edic ed abundance and in e -annual
clima e luc ua ions (in abundance le els ac oss ime), and anomalies (changes in clima ic a iables in
ela ion o he mean), showed nega i e ela ionships o o al p ecipi a ion (PpT; R
1
(pseudoR
2
)=0.30),
and i s anomalies (PpTa; R
1
=0.30), as well as minimum empe a u e (Tmin; R
1
=0.27) and i s
anomalies (Tmina; R1=0.28) (Figu e S6).
Remo e Sens. 2019,11, 2086 9 o 16
Remo e Sens. 2019, 11, x FOR PEER REVIEW 10 o 18
Figu e 4. Yea ly a e age alues o p edic ed abundance o I is boissie i (black line) in he s udy a ea
o he ime-pe iod 2001–2016 agains annual a iables: (a) EFAs h ough EVI minimum (EVIm) and
da e o minimum EVI; (b) clima e h ough o al p ecipi a ion (Pp), maximum (Tmax) and minimum
(Tmin) empe a u e; (c) o al amoun o bu ned a ea (TBA). Da k g ey ba s ep esen he calib a ion
o he pe iod 2001–2006. Red lines ep esen lag esponse o he species o i e e en s.
Figu e 4.
Yea ly a e age alues o p edic ed abundance o I is boissie i (black line) in he s udy a ea o
he ime-pe iod 2001–2016 agains annual a iables: (
a
) EFAs h ough EVI minimum (EVIm) and da e
o minimum EVI; (
b
) clima e h ough o al p ecipi a ion (Pp), maximum (Tmax) and minimum (Tmin)
empe a u e; (
c
) o al amoun o bu ned a ea (TBA). Da k g ey ba s ep esen he calib a ion o he
pe iod 2001–2006. Red lines ep esen lag esponse o he species o i e e en s.
Remo e Sens. 2019,11, 2086 16 o 16
54.
Pea son, R.G.; Dawson, T.P. P edic ing he impac s o clima e change on he dis ibu ion o species: a e
bioclima e en elope models use ul? Glob. Ecol. Biogeog . 2003,12, 361–371. [C ossRe ]
55.
Deblauwe, V.; D oissa , V.; Bose, R.; Sonk
é
, B.; Blach-O e gaa d, A.; S enning, J.-C.; Wie inga, J.J.;
Ramesh, B.R.; S
é
a , T.; Cou eu , T.L.P. Remo ely sensed empe a u e and p ecipi a ion da a imp o e
species dis ibu ion modelling in he opics. Glob. Ecol. Biogeog . 2016,25, 443–454. [C ossRe ]
56.
Lassueu , T.; Joos , S.; Randin, C.F. Ve y high esolu ion digi al ele a ion models: Do hey imp o e models o
plan species dis ibu ion? Ecol. Modell. 2006,198, 139–153. [C ossRe ]
57.
Co d, A.F.; Meen emeye , R.K.; Lei
ã
o, P.J.; V
á
cla
í
k, T. Modelling species dis ibu ions wi h emo e sensing
da a: b idging disciplina y pe spec i es. J. Biogeog . 2013,40, 2226–2227. [C ossRe ]
58.
Co d, A.F.; Klein, D.; Ge nand , D.S.; de la Rosa, J.A.P.; Dech, S. Remo e sensing da a can imp o e p edic ions
o species ichness by s acked species dis ibu ion models: A case s udy o mexican pines. J. Biogeog .
2014
,
41, 736–748. [C ossRe ]
59.
Alca az, D.; Pa uelo, J.; Cabello, J. Iden i ica ion o cu en ecosys em unc ional ypes in he ibe ian peninsula.
Glob. Ecol. Biogeog . 2006,15, 200–212. [C ossRe ]
60.
Hog e e, K.R.; Pa il, V.P.; Ru h au , D.R.; Meixell, B.W.; Budde, M.E.; Hupp, J.W.; Wa d, D.H. No malized
di e ence ege a ion index as an es ima o o abundance and quali y o a ian he bi o e o age in A c ic
Alaska. Remo e Sens. 2017,9, 1234. [C ossRe ]
61.
Ames, G.M.; Wall, W.A.; Hohmann, M.G.; W igh , J.P. T ai space o a e plan s in a i e-dependen ecosys em.
Conse . Biol. 2017,31, 903–911. [C ossRe ]
62.
Pausas, J.G.; Keeley, J.E. Wild i es as an Ecosys em Se ice. F on . Ecol. En i on.
2019
,17, 289–295. [C ossRe ]
63.
Renwick, A.R.; Massimino, D.; Newson, S.E.; Chambe lain, D.E.; Pea ce-Higgins, J.W.; Johns on, A. Modelling
changes in species’ abundance in esponse o p ojec ed clima e change. Di e s. Dis ib.
2012
,18, 121–132.
[C ossRe ]
©
2019 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/).
