Evaluating citizen science data for forecasting species responses to national forest management
Full text
368
|
www.ecole ol.o g Ecology and E olu ion 2017; 7: 368–378
Recei ed: 21 Sep embe 2016
|
Re ised: 13 Oc obe 2016
|
Accep ed: 20 Oc obe 2016
DOI: 10.1002/ece3.2601
ORIGINAL RESEARCH
E alua ing ci izen science da a o o ecas ing species
esponses o na ional o es managemen
Louise Mai 1 | Philip J. Ha ison1 | Ma i Jönsson1 | Swan je Löbel1,2 | Jenni
No dén3,4 | Juha Sii onen5 | Tomas Lämås6 | Ande s Lunds öm6 | To d Snäll1
This is an open access a icle unde he e ms o he C ea i e Commons A ibu ion License, which pe mi s use, dis ibu ion and ep oduc ion in any medium,
p o ided he o iginal wo k is p ope ly ci ed.
© 2016 The Au ho s. Ecology and E olu ion published by John Wiley & Sons L d.
1Swedish Species In o ma ion Cen e,
Swedish Uni e si y o Ag icul u al Sciences
(SLU), Uppsala, Sweden
2Depa men o En i onmen al Sys em
Analysis, Ins i u e o Geoecology, Technical
Uni e si y B aunschweig, B aunschweig,
Ge many
3Depa men o Resea ch and
Collec ions, Na u al His o y
Museum, Uni e si y o Oslo, Oslo, No way
4No wegian Ins i u e o Na u e Resea ch,
Oslo, No way
5Na u al Resou ces Ins i u e Finland, Van aa,
Finland
6Depa men o Fo es Resou ce
Managemen , Swedish Uni e si y o
Ag icul u al Sciences (SLU), Umeå, Sweden
Co espondence
To d Snäll, Swedish Species In o ma ion
Cen e, Swedish Uni e si y o Ag icul u al
Sciences (SLU), Uppsala, Sweden.
Email: o [email p o ec ed]
Funding in o ma ion
FORMAS, G an /Awa d Numbe : 2012-991
and 2013-1096
Abs ac
The ex ensi e spa ial and empo al co e age o many ci izen science da ase s (CSD)
makes hem appealing o use in species dis ibu ion modeling and o ecas ing.
Howe e , a equen limi a ion is he inabili y o alida e esul s. He e, we aim o as-
sess he eliabili y o CSD o o ecas ing species occu ence in esponse o na ional
o es managemen p ojec ions ( ep esen ing 160,366 km2) by compa ison agains
o ecas s om a model based on sys ema ically collec ed coloniza ion–ex inc ion da a.
We i ed species dis ibu ion models using ci izen science obse a ions o an old-
o es indica o ungus Phellinus e ugineo uscus. We applied i e modeling app oaches
(gene alized linea model, Poisson p ocess model, Bayesian occupancy model, and wo
MaxEn models). Models we e used o o ecas changes in occu ence in esponse o
na ional o es managemen o 2020- 2110. Fo ecas s o species occu ence om
models based on CSD we e cong uen wi h o ecas s made using he coloniza ion–ex-
inc ion model based on sys ema ically collec ed da a, al hough di e en modeling
me hods indica ed di e en le els o change. All models p ojec ed inc eased occu -
ence in se - aside o es om 2020 o 2110: he p ojec ed inc ease a ied be ween
125% and 195% among models based on CSD, in compa ison wi h an inc ease o
129% acco ding o he coloniza ion–ex inc ion model. All bu one model based on
CSD p ojec ed a decline in p oduc ion o es , which a ied be ween 11% and 49%,
compa ed o a decline o 41% using he coloniza ion–ex inc ion model. All models hus
highligh ed he impo ance o p o ec ed old o es o P. e ugineo uscus pe sis ence.
We conclude ha models based on CSD can ep oduce o ecas s om models based
on sys ema ically collec ed coloniza ion–ex inc ion da a and so lead o he same o es
managemen conclusions. Ou esul s show ha he use o a sui e o models allows
CSD o be eliably applied o land managemen and conse a ion decision making,
demons a ing ha widely a ailable CSD can be a aluable o ecas ing esou ce.
KEYWORDS
deadwood-dependen ungi, o es y, global biodi e si y in o ma ion acili y, habi a change,
land use change, oppo unis ic da a, olun ee eco ding
|
369
MAIR e Al.
1 | INTRODUCTION
Species dis ibu ion models (SDMs) ha e been ex ensi ely applied in
o ecas ing species esponses o u u e habi a and clima e change
(Eli h & Lea hwick, 2009). The empo al and spa ial ex en o such
s udies can be expanded h ough he inc easingly popula use o ci i-
zen science da a (CSD) (De ic o , Whi ake , & Bel ame, 2010). CSD
p o ide an inexpensi e sou ce o species obse a ion da a, pa icu-
la ly as he online colla ion o da a is becoming common p ac ice o
many egions o he wo ld (Sil e own, 2009). This g ea ly expands
he po en ial scope o SDM o ecas ing s udies. Fo ecas s can p o ide
aluable insigh s in o possible u u e condi ions, allowing land use
manage s and conse a ionis s o make in o med decisions (Mouque
e al., 2015).
A d awback o CSD is ha hey a e equen ly p esence- only
obse a ions, which canno be modeled using es ablished p es-
ence–absence amewo ks such as gene alized linea models (GLMs).
New me hods ha e he e o e been de eloped speci ically o model
p esence- only da a; o emos o hese is MaxEn (Phillips, Ande son,
& Schapi e, 2006). MaxEn has been shown o ou pe o m o he
me hods when p edic ing species’ dis ibu ions and has been ex en-
si ely es ed agains p esence–absence me hods such as GLMs (e.g.,
Eli h e al., 2006). MaxEn has been widely applied o CSD and used
o add ess a di e se ange o opics, including conse a ion applica-
ions (Eli h e al., 2011). Ye , MaxEn has o en been misunde s ood o
misused (Yackulic e al., 2013). The e o e, any in e ences made om
model p ojec ions mus be ca e ully assessed, pa icula ly in a man-
agemen con ex .
A second d awback is ha CSD o en su e om spa ial eco ding
biases (Dickinson, Zucke be g, & Bon e , 2010). Volun ee eco de s
may disp opo iona ely isi si es close o home o oads, o may a o
species- ich habi a s (Dennis & Thomas, 2000). I obse a ion da a a e
p esence- only, hen sepa a ing ou species–habi a associa ions om
olun ee - habi a p e e ences can be di icul (Ba bosa, Pau asso, &
Figuei edo, 2013). Spa ial o en i onmen al il e ing o eco ds can e-
duce bias and imp o e model pe o mance (Bo ia, Olson, Goodman, &
Ande son, 2014); howe e , such me hods in ol e h owing away da a.
Al e na i ely, spa ial eco ding bias can be explici ly modeled using a
small amoun o p esence–absence da a (Fi hian, Eli h, Has ie, & Kei h,
2015). This educes he in es men equi ed in ob aining p esence–
absence da a while making use o ex ensi e p esence- only da ase s.
This app oach pe o med well on one species g oup (Fi hian e al.,
2015), bu has ye o be widely es ed.
Thi dly, he impe ec de ec ion o species in he ield is a gen-
e al ea u e o obse a ion da a, ye is a ely accoun ed o in SDMs
(Lahoz- Mon o , Guille a- A oi a, & Win le, 2014). The de ec abili y o
a species ( he p obabili y ha an indi idual is obse ed whe e p es-
en ) may a y among si es and/o o e ime ( an S ien, an Swaay, &
Ke y, 2011). In he con ex o ci izen science, de ec ion may also a y
among eco de s due o di e ing iden i ica ion skills o sea ch e o .
We hence o h use he e m “occupancy model” o join modeling
o occu ence and de ec abili y (MacKenzie e al., 2002). Occupancy
models we e ini ially de eloped o accoun o impe ec de ec ion
using epea - su ey da a, bu ha e ecen ly been applied o ad hoc
CSD, success ully eco e ing expec ed ends in species’ dis ibu-
ions ( an S ien, an Swaay, & Te maa , 2013). Mo eo e , occupancy
models iden i ied biologically easonable species–habi a associa ions
when applied o spa ially biased da a, in con as o con en ional
eg ession models (Higa e al., 2015). The applica ion o occupancy
models o spa ially biased and/o ad hoc da a is as ye e y limi ed,
howe e , and u he es ing is equi ed o de e mine whe he in e -
ences om a di e si y o da ase s a e eliable.
The e a e hus a b oad a ie y o modeling app oaches a ailable
and p e ious wo k has concluded ha no single me hod consis en ly
p oduced he mos accu a e esul s (Qiao, Sobe ón, & Pe e son, 2015).
Mo eo e , di e en app oaches o deal wi h eco ding biases can p o-
duce di e en conclusions (Isaac, an S ien, Augus , de Zeeuw, &
Roy, 2014). A u he sou ce o a ia ion s ems om he inc easingly
popula echnique o combining co ela i e and mechanis ic compo-
nen s in species dis ibu ion modeling. The combina ion o co ela i e
and mechanis ic componen s, such as physiological cons ain s o
popula ion dynamics, has been ad oca ed o imp o e he biological
ealism o models (Kea ney & Po e , 2009). Howe e , he inclusion
o mechanisms can quan i a i ely change p ojec ed ends (Swab,
Regan, Ma hies, Becke , & B uun, 2015), implying ye ano he sou ce
o a ia ion among me hods. The e o e, i may in ac be p e e able
o apply mul iple me hods in o de o add ess sou ces o unce ain y
(Qiao e al., 2015).
A limi a ion o many modeling s udies ha apply CSD is he lack
o alida ion agains independen models based on sys ema ically
collec ed da a. I CSD a e o be widely applied in a eas such as land
managemen and conse a ion decision making, hen he abili y o
models based on CSD o p oduce o ecas s ha a e cong uen wi h
o ecas s om models based on sys ema ically collec ed da a should
be demons a ed. Cong uence would p o ide con idence in applying
cheap, widely a ailable CSD o a ange o o ecas ing ques ions, which
would inc ease he scope o o ecas ing s udies and a oid he need o
cos ly, ime- consuming da a collec ion by expe s.
In his s udy, we aimed o assess he eliabili y o species oc-
cu ence o ecas s om models based on CSD. We es ed whe he
i e di e en occu ence models based on open access CSD p o-
duced o ecas s ha we e cong uen wi h o ecas s om a dynamic
model based on coloniza ion–ex inc ion da a ha we e sys ema i-
cally collec ed by expe s. We hus compa ed o ecas s om models
based on di e ing quali y o da a (in e ms o ci izen scien is e -
sus expe collec ion) and di e ing biological in o ma ion con en
(occu ence CSD e sus dynamic coloniza ion–ex inc ion da a). We
p ojec ed changes in he occu ence o Phellinus e ugineo uscus,
an old- o es indica o ungus, in esponse o na ional o ecas s o
o es managemen in Sweden. All i e species dis ibu ion mod-
els based on CSD u ilized p esence- only and/o p esence–absence
da a collec ed by olun ee eco de s and we e selec ed o encom-
pass a di e se ange o da a equi emen s and assump ions abou
eco ding biases.
370
|
MAIR e Al.
2 | METHODS
2.1 | S udy species
Phellinus e ugineo uscus is a polypo us species associa ed wi h
No way sp uce, Picea abies. Polypo us ungi a e impo an dead-
wood decompose s and many species a e nega i ely a ec ed by o -
es managemen (No dén, Pen ilä, Sii onen, Tomppo, & O askainen,
2013). The occu ence o P. e ugineo uscus is de e mined by
deadwood a ailabili y and connec i i y old sp uce- domina ed o -
es (Jönsson, Edman, & Jonsson, 2008). Phellinus e ugineo uscus
is classi ied as nea h ea ened (NT) in Sweden due o o es y
(A da abanken, 2015). I has been widely used as an old- o es indi-
ca o species in na u e conse a ion in en o ies in he No dic coun-
ies (Niemelä, 2005). Phellinus e ugineo uscus is easy o ind and
iden i y in he ield.
2.2 | Ci izen science species obse a ion da a
Ci izen science da a o P. e ugineo uscus we e downloaded om he
Swedish open access Li ewa ch websi e (www.analysispo al.se) o
he pe iod 2000–2013 a he 100 m g id cell esolu ion. Obse a ions
we e p esence- only, and he species was eco ded in 5,317 cells
(Figu e 1). The Li ewa ch websi e is a po al ha compiles obse a ion
da a om mul iple sou ces. The p ima y sou ce o ungal obse a-
ions is he Swedish Species Obse a ion Sys em (www.a po alen.
se). Da a uploaded o he Species Obse a ion Sys em come om
many di e en eco de s anging om ama eu en husias s o ained
ield wo ke s ca ying ou in en o ies o o es y companies. Da a
may be comple e species checklis s o single species obse a ions;
howe e , as eco de s a e no equi ed o egis e species absences,
his in o ma ion is unknown.
To ob ain a p esence–absence da ase o P. e ugineo uscus, we
in e iewed eco de s o wood- dependen ungi. Each eco de was
asked he same ques ions abou hei ield me hods. I ield sea ches
we e ho ough and consis en (see Appendix S1 in Suppo ing
In o ma ion), hen obse a ion eco ds om ha eco de we e com-
piled o c ea e a p esence–absence da ase . Among hese, he p es-
ence o species o he han he a ge species was aken o indica e he
absence o he a ge species. Da a om eigh eco de s we e used
co e ing 15,508 g id cells (Appendix S1).
2.3 | En i onmen al da a
We hypo hesized ha P. e ugineo uscus occu ence p obabili y in-
c eased wi h li ing sp uce olume and o es s and age. Fo es da a
we e based on es ima es which combine sa elli e images and g ound-
u hing; “kNN- Sweden” (h p://skogska a.slu.se; Reese e al., 2003;
o de ails, see Appendix S2). Du ing model de elopmen , i became
clea ha eco ding e o was biased owa d olde o es . The e o e,
o es age was excluded in o de o a oid modeling eco ding bias
a he han species occu ence.
The kNN da a we e also used o es he hypo hesis ha species
occu ence inc eased wi h connec i i y o old o es , which e lec s he
po en ial dispe sal sou ces o he species in he su ounding landscape.
We used a connec i i y calcula ion adap ed om No dén e al. (2013)
(de ailed in Appendix S2). We es ed h ee alues o he dispe sal pa-
ame e ep esen ing a mean dispe sal dis ance o 1, 5, and 10 km.
We hypo hesized ha P. e ugineo uscus occu ence was nega-
i ely ela ed o empe a u e and p ecipi a ion, gi en he no he n bo-
eal dis ibu ion o he species. We also hypo hesized ha he e was
an in e ac i e e ec as he e ec o high wa e a ailabili y on ungal
ac i i y is lowe a colde empe a u es due o educed me abolic a es
(Boddy e al., 2014). G idded me eo ological da a we e ob ained om
he EURO4M Mesan da ase (Landelius, Dahlg en, Goll ik, Jansson,
& Olsson, 2016). We used mean annual empe a u e and seasonally
accumula ed p ecipi a ion om May o No embe , bo h a e aged o e
he pe iod 1989–2010 (see Appendix S2 o de ails). This ime ame
includes he 10 yea s p io o he species obse a ion da a as ui ing
FIGURE1 Obse ed 100 m g id cell esolu ion occu ences o
Phellinus e ugineo uscus 2000–2013 (N = 5,317) ob ained om
Swedish Li ewa ch (analysispo al.se)
N
0100 200
Kilome e s
|
371
MAIR e Al.
bodies obse ed om 2000 onwa ds may e lec coloniza ion se e al
yea s ea lie .
We calcula ed a we ness index and a a iable which e lec ed he
s eepness and o ien a ion o a g id cell using a digi al ele a ion map
(Swedish land su ey se ice; www.lan ma e ie .se; calcula ions in
Appendix S2). The hypo hesis was peak occu ence a in e media e
we ness, which ep esen s he op imum condi ions o he species’
p ima y habi a . Fo he a iable e lec ing s eepness and o ien a ion,
we hypo hesized a linea ela ionship e lec ing inc eased occu ence
on s eepe , no h- acing slopes due o lowe sun exposu e.
One o he modeling app oaches we applied accoun ed o spa-
ial biases in he collec ion o p esence- only da a (Fi hian e al., 2015).
We used he a iables popula ion densi y (numbe o people pe km2
in 2010; S a is ics Sweden, www.scb.se), log popula ion densi y, dis-
ance o small oads, dis ance o main oads, dis ance o he i e la g-
es ci ies, dis ance o all ci ies, and dis ance o owns ( oad and u ban
a ea da a om he Swedish land su ey se ice). All a iables we e
ans o med om polygon da a o 100 m g id cells. We es ed o bo h
linea and quad a ic e ec s o each bias a iable.
2.4 | Occu ence models based on ci izen
science da a
The complexi y o models was cons ained o imp o e compa a i e
abili y among models, o allow e alua ion o he biological plausibili y
o he species’ esponse cu es, and o a oid o e i ing (Me ow e al.,
2014). To acili a e assessmen o he ela i e impo ance o co a i-
a es, all a iables we e s anda dized (di ision wi h he s anda d de ia-
ion) p io o modeling. All modeling based on CSD was ca ied ou a
he 100 m g id cell esolu ion and he occu ence da a we e u ilized
as a single snapsho .
2.4.1 | GLM
A gene alized linea model wi h a binomial dis ibu ion and logi link
was i ed o he p esence–absence da a. We i s i ed a model using
li ing sp uce olume as he explana o y a iable. Model complexi y
was hen assessed using AIC (Bu nham & Ande son, 2002) o ensu e
ha model i was imp o ed wi h he inclusion o u he co a ia e o
in e ac ion e ms, see En i onmen al da a abo e. Models we e i ed
using R e sion 3.1.0 (R Co e Team, 2014).
2.4.2 | MaxEn
MaxEn is a maximum en opy model which makes use o species
p esence- only obse a ions and a backg ound sample (Eli h e al.,
2011; Phillips e al., 2006). The backg ound sample may also be e-
e ed o as “pseudo- absence” da a. We used wo app oaches o
ob ain he backg ound sample. Fi s ly, we sampled 40,000 g id cells
andomly om he s udy a ea, excluding cells wi h p esence- only e-
co ds o he ocal species. Secondly, in o de o accoun o eco d-
ing biases, we applied he a ge - g oup backg ound (TGB) me hod
(Phillips & Dudik, 2008), whe e backg ound cells we e selec ed based
on he p esence o species wi h simila eco ding biases (bu no he
ocal species). We selec ed wood- dependen ungal species (N = 202;
S okland & Meyke, 2008) as he a ge g oup. This ga e 34,430 back-
g ound cells (downloaded om Swedish Li ewa ch o 2000–2013 a
100 m esolu ion).
In o de o p e en he inclusion o spu ious in e ac ions o qua-
d a ic e ms wi h no biological jus i ica ion, we c ea ed all in e ac ions
and quad a ic e ms and en e ed hem in o MaxEn as so- called lin-
ea ea u es. All o he MaxEn ea u es we e swi ched o (Phillips &
Dudik, 2008). Va iable selec ion was ca ied ou by main aining only
he co a ia es which had an impo ance o con ibu ion g ea e han
ze o. AUC was calcula ed on he p esence–absence da a o ensu e
ha no loss in p edic i e abili y occu ed when a iables we e e-
mo ed. Models we e i ed using MaxEn e sion 3.3.3 un om R
using he dismo package e sion 1.5 (Hijmans, Phillips, Lea hwick, &
Eli h, 2014).
2.4.3 | PA/PO model
We also applied an inhomogeneous Poisson poin - p ocess model
which combines p esence- only and p esence–absence species’ obse -
a ion da a ( e med he e “PA/PO model”; Fi hian e al., 2015). The
app oach models species occu ence agains en i onmen al a iables
while explici ly modeling spa ial bias in eco ding e o , by combining
a species occu ence componen and a eco ding bias componen . The
model equi es p esence- only da a o mul iple species, a small sample
o p esence–absence da a, and a backg ound sample.
We used p esence- only and p esence–absence da a o ou s udy
species and six o he sp uce- associa ed deadwood- dependen ungi
(Amylocys is lapponica, Fomi opsis osea, Lep opo us mollis, Phellinus
ch ysoloma, Phellinus nig olimi a us, and Phlebia cen i uga). Fo he
backg ound sample, we andomly sampled 40,000 cells ac oss he
s udy a ea. We es ed he en i onmen al and bias a iables desc ibed
in En i onmen al da a abo e. Va iable selec ion was based on AIC o
P. e ugineo uscus. Models we e i ed in R using he package mul ispe-
ciesPP e sion 1.0.
2.4.4 | Occupancy model
Es ima ing species de ec abili y using occupancy modeling elies on
da a om epea isi s o si es wi hin a closed pe iod. We es ablished
a de ec ion/nonde ec ion da ase o P. e ugineo uscus using he
p esence- only ci izen science da a. We i s iden i ied o he old- o es
indica o species o deadwood- dependen ungi which, based on ou
knowledge, ci izen scien is s in e es ed in P. e ugineo uscus we e
highly likely o also sea ch o and eco d when ound (N = 35; see
Appendix S3). We used de ec ions o indica o species o he han ou
ocal species o indica e he nonde ec ion o he ocal species. A small
p opo ion o g id cells had wo o mo e species obse a ion eco ds
occu ing on di e en days wi hin he same calenda yea , and we
u ilized hese obse a ions as epea - isi da a. We used a calenda
yea as he de ini ion o a closed pe iod as he species’ ui ing body
li e span is 1–2 yea s. The da a consis ed o 29,615 g id cells, o which
372
|
MAIR e Al.
807 g id cells ecei ed wo o mo e isi s (o hese, maximum numbe
o isi s = 7, median = 2).
We o mula ed he occupancy model in a Bayesian amewo k.
The p obabili y o occu ence and he p obabili y o de ec ion we e
modeled as a logis ic unc ion, essen ially as in Ké y, Ga dne , and
Monne a (2010). Obse ed da a a e a esul o he in e ac ion be-
ween he ue occu ence and he de ec abili y o he species. T ue
occu ence was modeled as a unc ion o he en i onmen al a iables.
De ec abili y was assumed o a y among eco de s (and he e o e o
a y among si es and isi s depending on he eco de p esen ) and
was modeled agains he o al numbe o days each indi idual eco de
had submi ed eco ds o wood- li ing indica o species du ing he
s udy pe iod. Fo a discussion o he de ec abili y a iables conside ed,
see Appendix S4.
Va iable selec ion o species occu ence was based on he pos e-
io dis ibu ions o he pa ame e s ( he use o DIC is no app op ia e
o mix u e/hie a chical models; Hoo en & Hobbs, 2014). I he 95%
c edible in e al o he pa ame e es ima e did no include ze o, hen
he a iable was conside ed o be signi ican . We s a ed wi h a model
which included li ing sp uce olume as he explana o y a iable o
occu ence and an in e cep - only de ec ion model. Complexi y was
inc eased by adding one a iable a a ime and assessing signi icance.
Once he species occu ence model was es ablished, he de ec abil-
i y model was i ed. The models we e i ed using OpenBUGS (Lunn,
Spiegelhal e , Thomas, & Bes , 2009) h ough R using he packages
R2OpenBUGS and BRugs. We an wo chains wi h 80,000 i e a ions
hinned by wo, a e a bu n- in o 20,000 i e a ions. The BUGS code
o he inal model is gi en in Appendix S5.
2.5 | Coloniza ion–ex inc ion model based on
sys ema ically collec ed ield da a
Occu ence models based on CSD we e compa ed agains a dynamic
model i ed o sys ema ically collec ed da a on coloniza ion– ex inc ion
e en s (Ha ison, P.J, Mai , L, No dén, J, Sii onen, J, Lunds öm, A,
Kind all, O, Snäll, T, in p epa a ion). To ob ain coloniza ion– ex inc ion
da a, we conduc ed esu eys in 2014 o 174 o es s ands in Finland
ha we e ini ially su eyed in 2003–2005 (No dén e al., 2013). In
bo h ime pe iods, we in en o ied all deadwood objec s wi h a diam-
e e a b eas heigh (DBH) ≥5 cm and leng h ≥1.3 m wi hin a ixed
su ey plo (usually 20 m × 100 m) inside each s and. Deadwood cha -
ac e is ics (used as explana o y a iables in addi ion o hose desc ibed
in En i onmen al da a abo e) and polypo e p esences we e eco ded.
We modeled he cu and noncu s ands sepa a ely. We used o -
wa d s epwise model selec ion and a iables we e e ained based on
he pos e io dis ibu ions o he pa ame e s. We i s de ine Zj, as he
ue occupancy s a e o plo j du ing su ey pe iod . We assume ha
Zj, ~ Be noulli(ψj, ). Fo he second su ey pe iod:
whe e
c∗
j,
and
e∗
j,
gi e, espec i ely, he coloniza ion and ex inc ion
p obabili ies, which ha e been o se o co ec o he di e en num-
be s o yea s be ween he su eys and he di e en plo a eas, such
ha :
whe e nj, gi es he numbe o yea s be ween he su eys di ided by
10 (i.e., scaled by he ypical numbe o yea s), and aj, gi es he plo
a ea di ided by 0.2 (i.e., scaled by he ypical plo size in hec a es). I
he o es in he plo had been clea - cu (ei he be o e he i s su ey
o be ween he wo su ey e en s), hen cloglog(cj, ) = δ1 and clo-
glog(ej, ) =
ε1
. We chose o use he complemen a y log–log link unc-
ion, cloglog, as due o i s asymme ical na u e i is be e sui ed han
he mo e con en ional logis ic link unc ion o cases whe e he p oba-
bili ies a e e y la ge o e y small. I he o es in he plo had no been
clea - cu , we assumed ha :
and ha cloglog(ei,j, ) =
ε2
. Due o da a limi a ions, we could no in-
clude co a ia es in he models o he ex inc ion p obabili ies o he
coloniza ion p obabili y on clea - cu cells (in e cep - only models we e
used in hese cases). The l co a ia es used in he model o he colo-
niza ion p obabili y a e gi en by Xl,j, wi h co esponding pa ame e
alues βl. Finally, we de ine Yj, as he obse ed occupancy s a e o plo
j du ing su ey pe iod . Fo he obse a ion model, we assume ha
Yi,j, ~ Be noulli(Zj, p) whe e p gi es he de ec ion p obabili y. This de-
ec ion p obabili y was es ima ed as 0.9 based on an in ensi e con ol
s udy. No coloniza ion e en s occu ed on cu si es and so hei colo-
niza ion p obabili y was se o ze o. In o de o ini ialize he models
used o simula e he u u e dynamics o he polypo e species, we used
a model i ed o he occu ence da a om 2014.
2.6 | Tempo al o ecas s o species occu ence in
esponse o o es managemen
In o de o es whe he he occu ence models based on CSD p oduced
o ecas s ha we e cong uen wi h o ecas s om he coloniza ion–
ex inc ion model based on sys ema ically collec ed dynamics da a, we
used he models o p ojec species occu ence in esponse o a o es
managemen scena io. Fo es p ojec ion da a we e a ailable om he
Swedish na ionwide Fo es Scena io Analyses 2015 (FSA 15; Claesson,
Du emo, Lunds öm, & Wikbe g, 2015; E iksson, Snäll, & Ha ison,
2015). Using he Heu eka sys em (Wiks öm e al., 2011), p ojec ions
we e made o he Na ional Fo es y In en o y (NFI) plo s (F idman e al.,
2014) o e e y i h yea om 2020 o 2110. We used da a o a o al o
17,383 NFI plo s om he whole bo eal egion o Sweden (160,366 km2
o p oduc i e o es ). Da a on p ojec ed changes in li ing and dead-
wood sp uce olume and o es age we e a ailable ( o da a de ails see
Appendix S6 and o calcula ion o connec i i y see Appendix S7). We
used a scena io which assumes ha 84% o he land is used o wood
p oduc ion and 16% is se - aside om o es y. The aim o se - aside o -
es is o imp o e biodi e si y conse a ion wi hin he o es ed landscape.
P ojec ions o species esponse o o es managemen we e based
on a space– ime subs i u ion, such ha we p ojec ed he occu ence
o he species ac oss he NFI plo s a each ime s ep, and so ob ained
ψ
j, =
(1
−Zj, −1
)
c
∗
j,
+Zj, −1
(1
−e
∗
j, )
c∗
j,
=
1
−
(1
−
c
j,
)n
j,
a
j,
e
∗
j,
=1−(1−ej, )
n
j,
aj,
cloglog(
cj, )=δ
2+
∑
l
βlXl,j,
|
373
MAIR e Al.
he change in species occu ence o e ime. The p ocedu e was as ol-
lows. Sepa a ely o each o he models, we p edic ed he p obabili y
o species’ occu ence a each NFI plo o each ime s ep. Mechanis ic
assump ions we e hen inco po a ed in o he p ojec ions. The species
could no occu whe e no deadwood was p esen (i is a deadwood-
dependen species), o whe e o es age was 25–64 yea s (due o
deadwood u no e on cu si es; see Appendix S8 o de ails). The al-
ues p edic ed a each plo we e hen scaled o e lec he p opo ion
o he o al coun y ha each plo ep esen s (densi y o plo s a ies
ac oss he coun y and hus he a ea ha each plo ep esen s a ies).
Scaled p obabili ies we e summa ized ac oss he whole egion and
sepa a ed in o p oduc ion and se - aside o es . Tempo al p ojec ions
using he models based on CSD we e compa ed agains p ojec ions
using he coloniza ion–ex inc ion model. We also calcula ed he el-
a i e change in species occu ence o e ime. Finally, we a e aged
p ojec ions o ela i e change ac oss all i e models based on CSD in
o de o es an ensemble modeling app oach.
We in es iga ed he sensi i i y o he esul s o he mechanis-
ic assump ions ou lined abo e. We compa ed p ojec ions om he
models based on CSD including (i) no mechanis ic assump ions; (ii) he
o es age h eshold assump ion alone; (iii) he deadwood p esence
assump ion alone; and (i ) bo h mechanis ic assump ions oge he .
2.7 | Spa ial p edic ion o cu en occu ence
To assess he spa ial accu acy o p edic ions o cu en species’ occu -
ence om he models based on CSD, we used block c oss- alida ion
and calcula ed he a ea unde he ecei e ope a ing cu e (AUC; see
Appendix S9 o de ails). We also used he models o p edic he cu -
en dis ibu ion o P. e ugineo uscus in Sweden a he 10 km g id
cell esolu ion. Species p obabili ies o occu ence we e p edic ed
ac oss he 100 m esolu ion sample o andom backg ound poin s
and agg ega ed o 10 km esolu ion using he mean. We applied he
mechanis ic assump ion ela ing o o es age, bu could no apply he
deadwood assump ion as no na ional GIS laye on deadwood occu -
ence exis s. Maps we e compa ed isually.
3 | RESULTS
3.1 | Tempo al p ojec ions: o es managemen
scena io
Fo ecas s om he occu ence models based on CSD we e gene -
ally cong uen wi h o ecas s om he coloniza ion–ex inc ion model
based on sys ema ically collec ed da a (Figu es 2 and 3). All models
p ojec ed p obabili y o occu ence o P. e ugineo uscus (o sui abil-
i y in he case o MaxEn ) o be lowe in p oduc ion o es han in
se - aside o es se - aside (Figu e 3). P obabili y o occu ence was
p ojec ed o inc ease o e ime in se - asides, bu o decline in p o-
duc ion o es acco ding o all bu one o he models based on CSD
(MaxEn TGB p ojec ed a sligh inc ease).
FIGURE2 Fo ecas s o mean p obabili y o Phellinus
e ugineo uscus occu ence in esponse o p ojec ed o es
managemen o e he coming cen u y om he coloniza ion–
ex inc ion model based on sys ema ically collec ed da a. Mean
p obabili y o occu ence is p esen ed o all o es and o
p oduc ion and se - aside o es sepa a ely. The ela i e changes in
p obabili y o occu ence (%) om 2020 o 2110 a e gi en o se -
aside and p oduc ion o es
0.0 0.1 0.2 0.30.4
Yea
Mean p obabili y o occu enc
e
2020 2050 2080 2110
To al
P oduc ion
Se −aside
+ 129 %
− 41 %
FIGURE3 Fo ecas s o mean p obabili y o Phellinus e ugineo uscus occu ence (o sui abili y) in esponse o p ojec ed o es managemen
o e he coming cen u y using models based on ci izen science da a. Models used we e (a) GLM; (b) PA/PO model; (c) occupancy model; (d)
MaxEn andom backg ound; and (e) MaxEn TGB. Mean p obabili y o occu ence is p esen ed o all o es and o p oduc ion and se - aside
o es sepa a ely. The ela i e changes in p obabili y o occu ence (%) om 2020 o 2110 o each model ype a e gi en o se - aside and
p oduc ion o es
0.00.1 0.20.3 0.4
Yea
Mean p obabili y o occu enc
e
2020 2050 2080 2110
To al
P oduc ion
Se −aside
(a)
+ 195 %
−34 %
0.0 0.10.2 0.30.4
Yea
Mean p obabili y o occu enc
e
2020 2050 2080 2110
(b)
+ 191 %
−49 %
0.00.1 0.20.3 0.4
Yea
Mean p obabili y o occu enc
e
2020 2050 2080 2110
(c)
+ 132 %
−11 %
0.00.1 0.20.3 0.4
Yea
Mean sui abili
y
2020 2050 2080 2110
(d)
+ 115 %
2 %
0.00.1 0.20.3 0.4
Yea
Mean sui abili
y
2020 2050 2080 2110
(e)
+ 147 %
+ −22 %
374
|
MAIR e Al.
Al hough all models p ojec ed compa able ends, di e en models
p ojec ed di e en amoun s o change o e ime. The inc ease om
2020 o 2110 in p obabili y o occu ence in se - asides a ied be-
ween 115% and 195% among models based on CSDs, compa ed o
an inc ease o 129% p ojec ed by he coloniza ion–ex inc ion model.
In p oduc ion o es , only he MaxEn TGB model p ojec ed a sligh in-
c ease in p obabili y o occu ence o 2%, while he emaining models
based on CSD p ojec ed declines o 11% o 49%. The coloniza ion–
ex inc ion model p ojec ed a decline o 41%.
P ojec ed ends in ela i e change o e ime we e e y simila
be ween he coloniza ion–ex inc ion model and he a e aged models
based on CSD, al hough he la e p ojec ed la ge inc eases in se - aside
o es (Figu e 4). A e aging ac oss models based on CSD ga e an in-
c ease o 162% in se - asides and decline o 20% in p oduc ion o es .
3.2 | Spa ial p edic ions: species dis ibu ions maps
Simila AUC sco es on bo h aining and wi hheld es ing da a we e
ob ained o all models based on CSD (Appendix S9), sugges ing ha
he di e en app oaches all achie ed good i s. The mean aining
AUC was 0.83–0.84 and mean es ing AUC was 0.78–0.79.
All i e app oaches highligh ed cen al Sweden as ha ing he high-
es p obabili y o P. e ugineo uscus occu ence (Figu e 5). The GLM,
PA/PO model, and occupancy model di e ed in absolu e p obabili-
ies, wi h he occupancy model p edic ing gene ally highe alues. The
MaxEn model p edic ions o ela i e sui abili y we e ypically also
highe alues.
3.3 | Key en i onmen al a iables in models based
on ci izen science da a
Final models had a ying s uc u es bu no able simila i ies (Appendix
S10). All models iden i ied li ing sp uce olume as he a iable wi h
he s onges posi i e ela ionship wi h P. e ugineo uscus occu -
ence. The a iable wi h he second s onges and posi i e e ec was
connec i i y. Fi ed lines illus a ing he e ec s o he ou mos im-
po an a iables (sp uce olume, connec i i y, empe a u e, and p e-
cipi a ion) indica ed ha he MaxEn TGB model iden i ied a weake
FIGURE4 Fo ecas s o ela i e change
in Phellinus e ugineo uscus occu ence in
esponse o p ojec ed o es managemen
o e he coming cen u y om (a) he
coloniza ion–ex inc ion model based
on sys ema ically collec ed da a and (b)
a e aged p ojec ions om he models
based on ci izen science da a (mean ± SD).
Rela i e change is p esen ed o all o es
(“ o al”) and o p oduc ion and se - aside
o es sepa a ely
Yea
Rela i e change (%)
To al
P oduc ion
Se –aside
(a)
050 100 150 200
050 100 150 200
Yea
Rela i e change (%)
2020 2050 2080 2110 2020 2050 2080 2110
(b)
FIGURE5 Maps o he p edic ed p obabili y o Phellinus e ugineo uscus cu en occu ence (o p edic ed sui abili y in he case o
MaxEn models) a 10 km g id cell esolu ion o (a) GLM, (b) PA/PO model, (c) occupancy model, (d) MaxEn andom backg ound, and (e)
MaxEn TGB
0.77−0.80
0.36−0.40
>0−0.04
P obabili y o occu ence
(a)
0.77−0.80
0.36−0.40
>0−0.04
P obabili y o occu ence
(b)
0.77−0.80
0.36−0.40
>0−0.04
P obabili y o occu ence
(c)
0.77−0.80
0.36−0.40
>0−0.04
P edic ed sui abili y
(d)
0.77−0.80
0.36−0.40
>0−0.04
P edic ed sui abili y
(e)
|
375
MAIR e Al.
e ec o sp uce olume ela i e o he o he modeling app oaches
(Appendix S10).
The a iables explaining spa ial eco ding biases iden i ied by he
PA/PO model we e popula ion densi y and dis ance o small oads
(Appendix S10). The eco ding bias was highes a in e media e densi-
ies (a ound 2220 people pe km2), alling o e y low eco ding p ob-
abili ies a he ex emes o popula ion densi y. Reco ding bias was
highes a sho dis ances om small oads.
The sensi i i y analysis showed ha he o e all p obabili y o oc-
cu ence (o sui abili y) was educed wi h he inclusion o mechanis ic
assump ions (Appendix S10). The inclusion o he deadwood p esence
assump ion esul ed in a g ea e educ ion in p obabili y o occu ence
han inclusion o he o es age assump ion. The inclusion o mecha-
nis ic assump ions esul ed in bo h g ea e inc eases o e ime in se -
aside o es and mo e nega i e ends in p oduc ion o es ela i e o
p ojec ions ha did no inco po a e mechanis ic assump ions.
3.4 | Coloniza ion–ex inc ion model
F om he Finnish plo - le el da a, we obse ed nine ex inc ion e en s
( ou on noncu si es and i e on cu si es) and wel e coloniza ion
e en s (all o which occu ed on he noncu si es). Only s and age was
selec ed as he a iable explaining he coloniza ion p obabili y o non-
cu si es (Ha ison e al. in p ep).
4 | DISCUSSION
Species dis ibu ion models buil using ci izen science da a o e-
cas changes in P. e ugineo uscus occu ence in esponse o o es
managemen ha we e quali a i ely cong uen wi h o ecas s om
a coloniza ion–ex inc ion model buil using sys ema ically collec ed
da a (Ha ison e al. in p ep). The i e modeling app oaches we applied
(GLM, PA/PO model, Bayesian occupancy model, MaxEn andom
backg ound, and MaxEn TGB) all p ojec ed an inc ease in p obabil-
i y o occu ence o e ime in o es se - aside om p oduc ion. All
bu one model (MaxEn TGB) p ojec ed a decline in he al eady e y
low p obabili y o occu ence in p oduc ion o es . Thus, he ange o
modeling app oaches applied he e p oduced concu en o es man-
agemen conclusions, highligh ing he impo ance o se - aside o es s
o he pe sis ence o P. e ugineo uscus. Ou esul s demons a e
ha CSD can be a use ul o ecas ing esou ce, wi h he po en ial o
eliably in o m land managemen and conse a ion decision making.
All models based on CSD achie ed good spa ial i and p edic ed
dis ibu ion maps indica ed ag eemen ha cen al Sweden was he
mos sui able o P. e ugineo uscus. Ne e heless, he e was quan i a-
i e a ia ion among model o ecas s. Thus, model pe o mance may
a y depending on whe he i is assessed spa ially o empo ally (Smi h
e al., 2013). The MaxEn models p ojec ed he smalles amoun o
change o e ime and, in pa icula , he TGB me hod ailed o cap u e
he decline in sui abili y in p oduc ion o es ha was p ojec ed by
all o he models. P e ious wo k has ound ha , o spa ially biased
da a in MaxEn , selec ing backg ound poin s (some imes e e ed o
as “pseudo- absences”) based on he p esence o o he ecologically
simila species ( he a ge - g oup backg ound (TGB) me hod) esul ed
in be e model pe o mance han aking a andom backg ound sample
(Phillips e al., 2009); he e o e, he poo e pe o mance o he TGB
app oach was unexpec ed. The TGB model es ima ed a weake e ec
o sp uce olume on species occu ence compa ed o he o he mod-
els, which may explain he di e ing p ojec ion ends. I is likely he e-
o e ha he selec ion o species o he TGB sample is impo an in
de e mining model pe o mance. Mo eo e , ou esul s demons a e
ha p e iously es ed me hods o educe p oblems o spa ial eco d-
ing bias a e no necessa ily uni e sally applicable (S ola & Nielsen,
2015). Thus, he compa ison o mul iple di e en models in o de o
es ablish ag eemen has he po en ial o imp o e eliabili y and is likely
o be o pa icula impo ance when ex ending s udies o new egions
and species.
P e ious wo k has sugges ed ha , in o de o imp o e o ecas ing,
a ia ion among models can be deal wi h by using an ensemble ap-
p oach (A aújo & New, 2007; Ma mion, Pa iainen, Luo o, Heikkinen,
& Thuille , 2009). Indeed, a e aging ac oss p ojec ions om he mod-
els based on CSD esul ed in o ecas s o ela i e change ha we e
quan i a i ely simila o o ecas s om he coloniza ion–ex inc ion
model. Ne e heless, o e all he models based on CSD ended o
o e p edic inc eases in se - aside o es s and unde p edic declines
in p oduc ion o es compa ed o he coloniza ion–ex inc ion model
based on sys ema ically collec ed da a. By cap u ing he slow dynam-
ics o ce ain species, coloniza ion–ex inc ion models a e expec ed o
yield mo e in o ma i e p edic ions o species occu ences han s a ic
SDMs (Yackulic, Nichols, Reid, & De , 2015). Da a on species dynamics
a e a e, howe e , and ou esul s show ha simila quali a i e conclu-
sions can be eached using occu ence models based on widely a ail-
able ci izen science occu ence da a.
The use o p esence–absence, a he han p esence- only, da a
is o en conside ed p e e able o species dis ibu ion modeling
(B o ons, Thuille , A aújo, & Hi zel, 2004). Ou esul s suppo his as-
se ion as he models which used p esence–absence da a (GLM and
PA/PO model) p ojec ed la ge declines in p oduc ion o es , which
we e mo e acquiescen wi h he coloniza ion–ex inc ion model o e-
cas s. Ou esul s addi ionally suppo he PA/PO model (Fi hian e al.,
2015) as a p omising ad ance in he e icien use o a ailable da a, due
o he good pe o mance demons a ed he e and he equi emen o
only a small amoun o p esence–absence da a. Ob aining p esence–
absence da a o his s udy was a ime- consuming bu wo hwhile en-
dea o , as he use o p esence–absence da a a oids eco ding biases
being modeled as species’ habi a associa ions (Yackulic e al., 2013).
Howe e , his also highligh s he bene i o asking ci izen scien is s o
p o ide in o ma ion on hei me hodologies du ing da a uploading. A
sligh inc ease in in o ma ion p o ided can g ea ly imp o e he alue
o ad hoc obse a ion da a; o example, comple e species lis s can be
used o asce ain absences (Isaac e al., 2014).
Occupancy modeling has been ad oca ed as a pa icula ly use ul
ool o ex ac ing obus conclusions om ci izen science da a (Bi d
e al., 2014). We applied p esence- only da a o he occupancy ame-
wo k, which is a ela i ely no el app oach (bu see Ké y, Royle, e al.
376
|
MAIR e Al.
(2010) and an S ien, Te maa , G oenendijk, Mensing, and Ke y (2010)
o ea ly examples). P e ious wo k has ound ha species lis s mus be
comp ehensi e in o de o p oduce eliable ends ( an S ien e al.,
2010). Howe e , based on ou esul s, we sugges ha bo h sho
and long species lis s can be used oge he , along wi h an in o ma i e
de ec abili y a iable e lec ing eco de expe ience, in o de o make
use o all a ailable obse a ion da a. One limi a ion o ou app oach
was ha he occu ence o he ocal species was modeled ela i e o a
wide g oup o ecologically simila species. As a esul , ou p edic ions
we e o he occu ence o P. e ugineo uscus gi en he p esence o
o he old- o es indica o ungi, which explains he high p obabili ies
o occu ence in he p edic ed dis ibu ion maps. Ne e heless, p o-
jec ions o ela i e change we e easonable, sugges ing ha eliable
esul s can be ob ained e en o spa ially biased da a, suppo ing con-
clusions by Higa e al. (2015).
O impo ance in gene a ing easonable p ojec ions was he in-
clusion o mechanis ic assump ions. The inco po a ion o mechanis ic
assump ions in o co ela i e models can p o ide no el insigh s in o
he p ocesses a ec ing species dynamics (Swab e al., 2015). The in-
co po a ion o mechanis ic assump ions he e imp o ed he biological
ealism o he models, by cap u ing aspec s o P. e ugineo uscus ecol-
ogy which we e no included in he co ela i e s uc u es and educing
he likelihood o o e p edic ing species occu ence.
This s udy is one o he ew o apply species dis ibu ion models o
CSD o a sessile species (bu see Ma mion e al. (2009) o a s udy on
plan s). Deadwood- dependen ungi a e a less well- s udied o ganism
g oup ela i e o he popula bi ds and bu e lies; howe e , such ses-
sile species could in ac be pa icula ly appealing o ci izen science
ini ia i es, gi en he oppo uni y o ime o be aken o e iden i i-
ca ion. Mo eo e , deadwood- dependen ungi a e a unc ionally e y
impo an g oup (O osson e al., 2015), and hei success ul model-
ing could acili a e he conside a ion o di e en ace s o ecosys em
unc ioning in o es o ecas ing. Fo example, P. e ugineo uscus is a
ed- lis ed species and i s p esence is likely o indica e a ela i ely na -
u al o es and he p esence o o he deadwood (sp uce)- dependen
species. The esul s p esen ed he e open up he oppo uni y o CSD
on o he sessile o ganism g oups, such as lichens and b yophy es, o
also be used in modeling and o ecas ing.
We ha e shown ha models based on ci izen science da a p o-
jec ed ends in P. e ugineo uscus occu ence in esponse o o es
managemen ha we e cong uen wi h ends om a model based on
sys ema ically collec ed ield da a on coloniza ion–ex inc ion e en s.
Applying a ange o app oaches based on di e en assump ions and
achie ing ag eemen among hem s eng hened con idence in he e-
sul s. Ci izen science da a hold he po en ial o be eliably applied in
o ecas ing species esponses o land use scena ios, opening up he
possibili y ha such ex ensi e da a could be use ul o conse a ion
and o es managemen planning.
ACKNOWLEDGMENTS
We hank Ke s in Be gelin, Ö jan F i z, Janolo He mansson, Olli
Manninen, Kjell Ma hson, Pe - E ik Mukka, Dan Olo sson, Ani a
S id all, So ia Sunds öm, and Tony S ensson o ag eeing o be in-
e iewed. We hank T. Landelius and he EURO4M eam (Eu opean
g an ag eemen no.: 242093) o ea ly access o he EURO4M Mesan
da ase . We hank he many eco de s con ibu ing species obse a-
ion da a. LM, PJH, and TS we e unded by FORMAS g an 2012- 991
and TS by 2013- 1096.
CONFLICT OF INTEREST
None decla ed.
DATA ACCESSIBILITY
Species obse a ion da a a e a ailable om he Swedish Li ewa ch
websi e; www.analysispo al.se. Na ional o es da a, “kNN- Sweden,”
a e a ailable om h p://skogska a.slu.se. The EURO4M Mesan
da ase (clima e da a) is publicly a ailable h ough he Ea h Sys em
G id Fede a ion (ESGF), o example, h p://esg-dn1.nsc.lui.se and
sea ch om “mesan.”
REFERENCES
A aújo, M. B., & New, M. (2007). Ensemble o ecas ing o species dis ibu-
ions. T ends in Ecology & E olu ion, 22, 42–47.
A da abanken (2015). Rödlis ade a e i S e ige 2015 [The 2015 Swedish Red
Lis ]. Uppsala: A da abanken SLU.
Ba bosa, A. M., Pau asso, M., & Figuei edo, D. (2013). Species–people co -
ela ions and he need o accoun o su ey e o in biodi e si y anal-
yses. Di e si y and Dis ibu ions, 19, 1188–1197.
Bi d, T. J., Ba es, A. E., Le check, J. S., Hill, N. A., Thomson, R. J., Edga ,
G. J., … F ushe , S. (2014). S a is ical solu ions o e o and bias
in global ci izen science da ase s. Biological Conse a ion, 173,
144–154.
Boddy, L., Bün gen, U., Egli, S., Gange, A. C., Heegaa d, E., Ki k, P. M., …
Kause ud, H. (2014). Clima e a ia ion e ec s on ungal ui ing. Fungal
Ecology, 10, 20–33.
Bo ia, R. A., Olson, L. E., Goodman, S. M., & Ande son, R. P. (2014). Spa ial
il e ing o educe sampling bias can imp o e he pe o mance o eco-
logical niche models. Ecological Modelling, 275, 73–77.
B o ons, L., Thuille , W., A aújo, M. B., & Hi zel, A. H. (2004). P esence-
absence e sus p esence- only modelling me hods o p edic ing bi d
habi a sui abili y. Ecog aphy, 27, 437–448.
Bu nham, K. P., & Ande son, D. R. (2002). Model selec ion and mul imodel
in e ence. A p ac ical in o ma ion- heo e ic app oach, 2nd edn. New Yo k,
NY: Sp inge -Ve lag.
Claesson, S., Du emo, K., Lunds öm, A., & Wikbe g, P. E. (2015). Fo es im-
pac analysis 2015 - SKA 15 (Skogliga konsek ensanalyse - SKA 2015).
Swedish Fo es Agency, Repo 10.
Dennis, R. L. H., & Thomas, C. D. (2000). Bias in bu e ly dis ibu ion maps:
The in luence o ho spo s and eco de ’s home ange. Jou nal o Insec
Conse a ion, 4, 73–77.
De ic o , V., Whi ake , R. J., & Bel ame, C. (2010). Beyond sca ci y:
Ci izen science p og ammes as use ul ools o conse a ion biogeog-
aphy. Di e si y and Dis ibu ions, 16, 354–362.
Dickinson, J. L., Zucke be g, B., & Bon e , D. N. (2010). Ci izen science as
an ecological esea ch ool: Challenges and bene i s. Annual Re iew o
Ecology, E olu ion, and Sys ema ics, 41, 149–172.
Eli h, J., G aham, C. H., Ande son, R. P., Dudík, M., Fe ie , S., Guisan, A.,
… Zimme mann, N. E. (2006). No el me hods imp o e p edic ion o
species’ dis ibu ions om occu ence da a. Ecog aphy, 29, 129–151.