This is a sel -a chi ed e sion o an o iginal a icle. This e sion
may di e om he o iginal in pagina ion and ypog aphic de ails.
Au ho (s):
Ti le:
Yea :
Ve sion:
Copy igh :
Righ s:
Righ s u l:
Please ci e he o iginal e sion:
CC BY 4.0
h ps://c ea i ecommons.o g/licenses/by/4.0/
The gian panda is c yp ic
© The Au ho (s) 2021
Published e sion
Nokelainen, Ossi; Sco -Samuel, Nicholas E.; Nie, Yonggang; Wei, Fuwen; Ca o,
Tim
Nokelainen, O., Sco -Samuel, N. E., Nie, Y., Wei, F., & Ca o, T. (2021). The gian panda is c yp ic.
Scien i ic Repo s, 11, A icle 21287. h ps://doi.o g/10.1038/s41598-021-00742-4
2021
1
Vol.:(0123456789)
Scien i ic Repo s | (2021) 11:21287 | h ps://doi.o g/10.1038/s41598-021-00742-4
www.na u e.com/scien i ic epo s
The gian panda is c yp ic
Ossi Nokelainen1*, Nicholas E. Sco ‑Samuel2, Yonggang Nie3,4, Fuwen Wei3,4 & Tim Ca o5,6*
The gian panda (Ailu opoda melanoleuca) is an iconic mammal, bu he unc ion o i s black‑and‑
whi e colo a ion is mys e ious. Using pho og aphs o gian pandas aken in he wild and s a e‑o ‑
he‑a image analysis, we con i m he coun e in ui i e hypo hesis ha hei colo a ion p o ides
camou lage in hei na u al en i onmen . The black u blends in o da k shades and ee unks,
whe eas whi e u ma ches oliage and snow when p esen , and in e media e pelage ones ma ch
ocks and g ound. A longe iewing dis ances gian pandas show high edge dis up ion ha b eaks
up hei ou line, and up close hey ely mo e on backg ound ma ching. The esul s a e consis en
ac oss acui y‑co ec ed canine, eline, and human ision models. We also show quan i a i ely ha he
species animal‑ o‑backg ound colou ma ching alls wi hin he ange o o he species ha a e widely
ecognised as c yp ic. Thus, hei colo a ion is an adap a ion o p o ide backg ound ma ching in he
isual en i onmen in which hey li e and simul aneously o a o d dis ance‑dependen dis up i e
colo a ion, he la e o which cons i u es he i s compu a ional e idence o his o m o p o ec i e
colo a ion in mammals.
Mos mammals ha e d ab colo a ion, gene ally showing b own ones, bu he e a e a small numbe o well-known
excep ions ha demand e olu iona y explana ion1,2. One is he gian panda (Ailu opoda melanoleuca), an iconic
lagship species o conse a ion biology which is amilia o a g ea many people, bu he unc ion o i s black-
and-whi e colo a ion has p o en puzzling. Se e al hypo heses o i s s ange appea ance ha e been sugges ed.
These include in aspeci ic signalling3,4, hea managemen 3, aposema ism5,6, and backg ound ma ching5,7, he
la e wo o a oid p eda ion by ige s (Pan he a ig is), leopa ds (Pan he a pa dus) and dholes (Cuon alpinus),
all o which a e sugges ed o p ey on gian pandas, especially young ones3,4. To da e howe e , he adap i e sig-
ni icance o i s seemingly conspicuous black-and-whi e colo a ion emains un esol ed.
Wi h espec o camou lage8, colo a ion may acili a e c ypsis (i.e., p e en ing de ec ion), o example h ough
bo h backg ound ma ching9, whe eby an animal’s appea ance ma ches he colou , ligh ness and pa e n o i s
isual backg ound, and dis up i e colo a ion, whe eby an animal’s appea ance c ea es alse edges o bounda ies
hinde ing ecogni ion o i s shape10. Ca o and colleagues11 used a compa a i e phylogene ic analysis o ca ni o es
o p opose ha he gian panda has c yp ic body pelage (while he black ma kings a ound he eyes and on he
ea s could be used in communica ion). Using compa a i e me hods o sco e colo a ion, hey showed ha ligh e
u colou co ela es wi h he p esence o snow ac oss e es ial ca ni o e species, making i plausible ha whi e
a eas o he gian panda’s body a e c yp ic agains he seasonal snowy backg ound cha ac e is ic o he species’
habi a 11. In con as , da k a eas a e associa ed wi h shade in ca ni o es and u sids, sugges ing hese pa s o he
gian panda’s pelage migh keep he animal camou laged in shaded o es ed a eas12. This dual unc ion colo a-
ion could be d i en by he need o his species o a e se wo e y di e en isual backg ounds (win e and
summe habi a ) du ing he cou se o he yea , because hei die p e en s hem om hibe na ing as do some
o he U sidae11. While backg ound ma ching was implica ed in hese analyses11, dis up i e colo a ion13,14 was
poo ly suppo ed. Howe e , he pelage displays wo c i ical ea u es o dis up i e colo a ion: highly con as ing
colou pa ches in e sec ing he animal’s ou line, and colo a ion ha ma ches a leas some isual elemen s o he
backg ound—maximum dis up i e con as and di e en ial blending, espec i ely8,15,16.
We explo e he hypo hesis ha he gian panda is c yp ic in i s na u al moun ain o es habi a , despi e he
ac ha he species’ black-and-whi e colo a ion appea s conspicuous when iewed close up (Fig.1). To achie e
his, we examined di icul - o-ob ain pho og aphs o 15 gian pandas aken in he wild using he Mul ispec al
Image Calib a ion and Analysis (MICA) oolbox17 and he Quan i a i e Colou and Pa e n Analysis (QCPA)
amewo k18, which oge he p o ide s a e-o - he-a image analysis17–20, o es ablish whe he o no gian pan-
das a e camou laged ei he h ough backg ound ma ching o dis up i e colo a ion, o bo h. We no e ha he
OPEN
1Depa men o Biological and En i onmen al Science, Uni e si y o Jy äskylä, P.O. Box 35, 40014 Jy äskylä,
Finland. 2School o Psychological Science, Uni e si y o B is ol, B is ol, UK. 3Key Labo a o y o Animal Ecology
and Conse a ion Biology, Chinese Academy o Sciences, Beijing, China. 4Cen e o Excellence in Animal E olu ion
and Gene ics, Chinese Academy o Sciences, Kunming, China. 5School o Biological Sciences, Uni e si y o B is ol,
B is ol BS8 1TQ, UK. 6Cen e o Popula ion Biology, Uni e si y o Cali o nia, 1 Shields A enue, Da is, CA 95616,
USA. *email: [email p o ec ed]; [email p o ec ed]
2
Vol:.(1234567890)
Scien i ic Repo s | (2021) 11:21287 | h ps://doi.o g/10.1038/s41598-021-00742-4
www.na u e.com/scien i ic epo s/
Figu e1. Gian panda colo a ion and i s isual backg ound. Fi s ow: animals pho og aphed in a zoo (le
Beijing zoo, igh Chengdu zoo, Wikimedia commons, he es in na u e by FW), exempli ying how black-
and-whi e colo a ion s ands ou when iewed close up and/o agains an a i icial backg ound. Second ow:
In i s na u al habi a , he colo a ion is less conspicuous e en a close ange. Thi d ow: he camou lage is mo e
e iden a in e media e dis ances and om a ious angles, and helps he gian panda o blend in o i s na u al
en i onmen . Bo om ow: om u he away, he black-and-whi e colo a ion b eaks up he ou line o he
animal due o inc eased edge dis up ion. No e ha hese igu es a e mean illus a i e— he images a e no
no malised da a.
3
Vol.:(0123456789)
Scien i ic Repo s | (2021) 11:21287 | h ps://doi.o g/10.1038/s41598-021-00742-4
www.na u e.com/scien i ic epo s/
sample size is no la ge, bu i een eplica es a e adequa e o add ess some o he basic ques ions ela ed o hei
mys e ious appea ance om a isual ecology pe spec i e.
Fi s , using he QCPA amewo k and i s acui y-co ec ed clus e analysis (RNL clus e ing), we s udied how
closely he black-and-whi e pelage o he gian panda clus e s wi h he isual pa chiness o he backg ound. This
app oach uses spa ial il e ing o de e mine which colou and/o luminance elemen s clus e oge he in he
image (Fig.2). We assumed ha he pelage would ha e coun e pa s (i.e., simila b igh ness elemen s) p esen
in he isual backg ound i hey we e o se e a backg ound ma ching unc ion. We modelled he image da a
wi h dich oma ic (canine and elid—p eda o su oga es) and ich oma ic (human—a poin o e e ence) ision
sys ems and hei co esponding isual acui ies. Second, using Gabo il e ing o quan i y he a io o alse edges
ha un o hogonal o an animal’s ue ou line19, we p obed whe he he gian panda demons a es dis up i e
colo a ion ope a ing as a unc ion o dis ance. We hypo hesised ha edge dis up ion, i i occu s, would ope a e
mo e easily on animals u he away because, as he dis ance be ween p eda o and p ey inc eases, he la e ’s
ou line will become less isible han he highly salien dis up i e ma kings15,21,22. A he limi o objec de ec ion
(e.g., a he maximum dis ance a which any pa o he panda can s ill be seen by a po en ial p eda o ), bo h
dis up i e ma kings and he animal’s ou line should become unde ec able. Finally, o debunk he my h o con-
spicuousness o gian pandas in hei na u al en i onmen , a common misconcep ion, we asked whe he gian
panda colo a ion ma ches he backg ound colou a ia ion as assessed by a colou map compa ison echnique (a
componen o QCPA) using human ision as a p oxy18. We compa ed his simila i y- o-backg ound me ic in he
gian panda o a a ie y o o he species examined objec i ely along he “c ypsis-conspicuousness spec um”. Ou
Figu e2. Quan i a i e colou and pa e n analysis (QCPA) was used o in es iga e he gian panda colo a ion
in i s na u al habi a . We s udied how closely he black-and-whi e pelages o he gian panda clus e s wi h he
isual pa chiness o he backg ound. The RNL clus e ing me hod uses spa ial il e ing o de e mine which isual
elemen s g oup oge he in he image. He e, one example o he gian panda images is shown and modelled
using dich oma ic canine (domes ic dog) and elid (ca ) ision, and ich oma ic human ision. The i s ow
shows images con e ed o ecei e dependen cone ca ch images, he second ow shows acui y-co ec ed
(canine—dog 12cpd, elid—ca 10cpd, human 72cpd,) images, he hi d ow shows RNL ank- il e ed
images and he ou h ow shows RNL-clus e ed iew. These examples we e gamma co ec ed o be e image
sc eening.
4
Vol:.(1234567890)
Scien i ic Repo s | (2021) 11:21287 | h ps://doi.o g/10.1038/s41598-021-00742-4
www.na u e.com/scien i ic epo s/
esul s p o ide in o ma ion on how conspicuous black-and-whi e colo a ion in animals may p o ide concealmen
in he isual en i onmen in which hey li e and simul aneously a o d dis ance-dependen dis up i e colo a ion.
Resul s
Quan i ica ion o backg ound ma ching: how does he pa chiness o he black‑and‑whi e col‑
o a ion blend in o he isual en i onmen ? The gian panda pelage pa ches we e signi ican ly associ-
a ed wi h isual elemen s ound in hei na u al habi a (Chi-Squa e es o independence, χ2 = 89.01, d = 10,
p < 0.001). The e we e no pa ches ha we e unique o he gian panda and no ound in he backg ound in e ms
o luminance. The black pa ches clus e ed wi h shadows and na u ally da k ee unks, whi e pa ches clus e ed
wi h oliage (due o specula ligh e lec ing om he lea es unde high i adiance) and snow when p esen , and
mid ones clus e ed wi h g ound and ocks. Al hough people end o conside he gian panda as dicho omously
black-and-whi e, we de ec ed an addi ional isual clus e o o -whi e u , which cha ac e izes mid ones in e ms
o luminance (i.e., no being da k o ligh ).
Mo e speci ically, backg ound isual elemen s (i.e., shade, ees, g ound, snow, oliage, ocks) clus e ed
be ween he black, mid ones and whi e colou s o gian pandas’ pelage as ollows (Fig.3). On a e age ac oss he
h ee isual models, he black pa ches we e associa ed wi h da k ee unks (57.6%) and shadows (30.7%), bu
also wi h g ound (3.8%), ocks (3.8%) and oliage (3.8%). The whi e pa ches we e g ouped wi h snow (on a e -
age 60.0% o he images, bu always when snow was p esen ), waxy b igh oliage (30.0%) and ocks (10.0%).
The mid ones we e clus e ed wi h g ound (35.8%), oliage (28.2%), shady backg ound a eas (17.9%), and also
wi h ocks (17.9%).
QCPA showed simila clus e ing esul s o backg ound ma ching o ROI compa isons (i.e., egions o in e -
es : he panda, he backg ound, he en i e image) as ega ds numbe o isual clus e s o med, clus e luminance
and ela i e pa ch co e age (Fig.4, TablesS3–S7). Also, he majo i y o Visual Con as Analysis (VCA) me ics
o e lapped when compa ing alues ac oss he gian panda, he backg ound and he en i e image (Fig.S3). How-
e e , wo a iables om dich oma ic elid (ca ) and canine (dog) ision models showed a ma ked di e ence om
highe human ision model alues: weigh ed mean o RNL pa e n con as (VCA:MSL) and weigh ed s anda d
de ia ion o RNL luminance pa e n con as (VCA:sSL). The compa ison o clus e ing esul s be ween seasons
(summe , win e ) did no show ma ked di e ences (Fig.S4).
Quan i ica ion o dis up i e colo a ion. We ound ha gian panda edge dis up ion was poo a close
p oximi y (Figs.5 and S5), compa ed o edge dis up ion ha was measu ed u he away. The minimum (o
poo es ) edge dis up ion was obse ed a app oxima ely 12m (a 8 ela i e body leng hs dis ance). We specula e
ha maximal edge dis up ion should be close o 50 pe cen , because beyond his he ‘ alse edges’ become p e-
dominan , leading o he same ‘p oblem’ o ha ing an ou line. A a dis ance o app oxima ely 50m (app oxima ed
as 34 ela i e gian panda body leng hs), howe e , edge dis up ion inc eased s eeply o dich oma s, a e which
i le elled ou a e en longe dis ances (edge dis up ion, dis ance by ision model in e ac ion, F2,430 = 39.77,
p < 0.001, Fig.4). Signi ican ly, canine and elid ision sys ems had s eepe edge dis up ion by dis ance depend-
0.00
0.25
0.50
0.75
1.00
FOLIAGEGROUNDROCKS SHADESNOW TREES
backg ound isual elemen
equency
pelage blacks mid ones whi es
Canine ision − RNL clus e ed
A
0.00
0.25
0.50
0.75
1.00
FOLIAGEGROUNDROCKS SHADESNOWTREES
backg ound isual elemen
equency
pelage blacks mid ones whi es
Felid ision − RNL clus e ed
B
0.00
0.25
0.50
0.75
1.00
FOLIAGEGROUNDROCKS SHADESNOWTREES
backg ound isual elemen
equency
pelage blacks mid ones whi es
Human ision − RNL clus e ed
C
Figu e3. Backg ound ma ching o he gian panda. The clus e ing me hod uses spa ial il e ing o es which
isual elemen s clus e oge he in he image. The ision modelling was pe o med using acui y-co ec ed cone-
ca ch images. We used h ee ision models: (A) canine ision (dog 12cpd) and (B) elid ision (ca 10cpd), as
he dich oma ic p eda o su oga es, and (C) he ich oma ic human ision (72cpd) o he sake o compa ison
and o p o ide a poin o e e ence. The clus e ing esul s show ha he da k elemen s (e.g., shadows, ee
unks) in he backg ound a e g ouped wi h black pelage pa ches, b igh es elemen s (e.g., snow, b igh
lea es in oliage and ocks) a e g ouped oge he wi h whi e pelage pa ches, whe eas elemen s wi h mid ones
(e.g., g ound, bu also ocks, lea es and shades) a e g ouped wi h in e media e o -whi e pelage pa ches. The
equency o each subg oup is ep esen ed allowing a compa ison o he con ibu ion o hei p opo ion o he
whole. The equencies (0–1) a e addi i e o 100% wi hin he isual backg ound compa ison g oup.

5
Vol.:(0123456789)
Scien i ic Repo s | (2021) 11:21287 | h ps://doi.o g/10.1038/s41598-021-00742-4
www.na u e.com/scien i ic epo s/
ency han human ision, which may sugges ha edge dis up ion wo ks pa icula ly well agains isual sys ems
wi h lowe acui y.
Debunking he my h o conspicuousness: how simila is he colou o he gian panda o i s
na u al habi a and in compa ison o o he species? The gian panda is appa en ly conspicuous o
he human eye when seen a sho iewing dis ances. In i s na u al habi a , howe e , on a e age 50% o all col-
ou s wi hin an image a e pe cep ually simila o human ision be ween a gian panda and i s isual backg ound
(i.e., simila i y- o-backg ound ch oma ic o e lap, n = 15 images, s.d. = 0.1) as judged om colou space his o-
g ams (Figs.6 and S2). Colou space o e lap compa isons o 15 species, whose adap i e signi icance has been
demons a ed quan i a i ely, con i med he expec a ion ha he highes deg ee o a e age ch oma ic simila i y-
o-backg ound was ound among well-camou laged dese oden s, whe eas he lowes deg ee o a e age ch o-
ma ic simila i y was ound among aposema ic poison ogs (simila i y index by species, F14,134 = 18.99, p < 0.001).
Gian pandas all be ween je boas and sho e c abs, species classically assumed o be c yp ic in he ield (Fig.6).
These esul s debunk he common misconcep ion ha he gian panda would be conspicuous o human ision
in hei na u al habi a .
Discussion
We ha e shown ha he iconic black-and-whi e colo a ion o he gian panda is c yp ic in i s na u al en i on-
men . Fu he mo e, he gian panda shows dis ance-dependen dis up i e colo a ion21,23. Al hough dis up i e
colo a ion has been discussed in ela ion o o he mammalian axa be o e24,25, his is he i s ime o ou knowl-
edge ha dis up i e colo a ion has been shown o occu in mammals using a quan i a i e se o compu a ional
image analysis echniques.
Fi s , we demons a ed ha gian panda u pelage sha es luminance appea ance (i.e., has simila coun e -
pa s) wi h i s na u al isual en i onmen . The black pelage pa ches blend wi h da k shadows and ee unks,
hus p o iding concealmen agains da ke elemen s in he isual scene whe eas whi e pa ches ma ch oliage and
snow (when p esen ). The b igh ness a ia ion o he backg ound highligh s explains why whi e is associa ed wi h
oliage, which a e snow-co e ed du ing win e and ha e a waxy ou e cu icle (causing specula ligh ing) du ing
2.5
5.0
7.5
10.0
12.5
backg ound pandawhole−image
ROI
Clus e ID
Vision Ca Dog Human
Numbe o clus e s
A
0.00
0.25
0.50
0.75
1.00
backg oundpanda whole−image
ROI
Luminance
Vision Ca Dog Human
Clus e luminance
B
0e+00
2e+05
4e+05
6e+05
backg ound pandawhole−image
ROI
A ea (px)
Vision Ca Dog Human
Clus e size
C
0
25
50
75
100
backg oundpanda whole−image
ROI
Co e age
Vision Ca Dog Human
Rela i e co e age
D
Figu e4. Backg ound ma ching o he gian panda. The igu e shows he ela ionship be ween he measu ed
egion o in e es (ROI) and ision model used. Clus e ing esul s a e p esen ed o (A) numbe o clus e s
o med, (B) clus e luminance dis ibu ion, (C) clus e size measu ed in pixels ( hus, de ac o a ea is a bi a y
and ep esen s be ween g oup di e ences), and (D) he ela i e co e age o clus e size wi hin he measu ed
ROI (backg ound, panda, whole image). The clus e ing me hod uses spa ial il e ing o es which isual
elemen s clus e oge he in he image. The ision modelling was pe o med using acui y-co ec ed cone-ca ch
images (canine ision—dog 12cpd and elid ision—ca 10cpd, he dich oma ic p eda o su oga es, and he
ich oma ic human ision 72cpd).
6
Vol:.(1234567890)
Scien i ic Repo s | (2021) 11:21287 | h ps://doi.o g/10.1038/s41598-021-00742-4
www.na u e.com/scien i ic epo s/
summe . Thus, i seems plausible ha whi e pa ches can p o ide concealmen in well illumina ed o es gaps e en
wi hou snow, when small gaps in he canopy allow b igh ligh o pene a e he o es . In addi ion, in equen
pelage mid ones ma ch he g ound and ocks. We a gue ha hese mid ones enhance backg ound ma ching in
he gian panda by p o iding an in e media e colo a ion, which b idges he gap be ween he e y da k and e y
ligh isual elemen s in i s habi a . Plausibly, he mid ones allow mo e e ec i e backg ound ma ching han i he
panda we e jus black-and-whi e. This p o ides adap i e insigh in o he b own-and-whi e colo a ion o Sichuan
and Qinling gian panda indi iduals, because simila (b own) mid ones eplace black pelage on indi iduals in
hese popula ions. In he na u al isual en i onmen , which mos ly consis s o b igh illuminan small gaps,
la ge gaps and pa ial woodland shade12, he con as ing black-and-whi e colo a ion hus acili a es backg ound
ma ching. The esul s ob ained om bo h dich oma ic and ich oma ic ision models we e consis en , al hough
wo me ics (weigh ed mean o RNL pa e n con as and weigh ed s anda d de ia ion o RNL luminance pa e n
con as ) sugges ha dich oma s may pe cei e he gian panda’s colo a ion mo e uni o mly han ich oma s.
O e all, ou da a con i m p e iously epo ed indings (based on an independen and di e en compa a i e
me hodology) ha gian panda colo a ion ma ches i s backg ound11.
Second, using Gabo il e ing19 and h ee ision models, we ha e shown ha he gian panda demons a es
dis up i e colo a ion ope a ing as a unc ion o dis ance: edge dis up ion is g ea e a longe dis ances in com-
pa ison wi h sho e iewing dis ances, whe e he e may be mo e eliance on backg ound ma ching. The la e
would always be expec ed o ope a e as a unc ion o dis ance: when an animal is a enough away bo h i and i s
backg ound will be blu ed, a which poin backg ound ma ching will ob ain ega dless o he spa ial s uc u e
o indi idual colou s. In his con ex , dis up i e colo a ion is pa icula ly in e es ing because he e appea s o
be an in e media e s age when he spa ial s uc u e o he dis up i e elemen s can be esol ed e en when he
a ge shape canno . In e es ingly, we ound ha dich oma ic isual sys ems wi h lowe isual acui y han human
ision showed s eepe edge dis up ion by dis ance. Dis up i e colo a ion may he e o e wo k pa icula ly well
agains isual sys ems wi h low acui y, especially when iewed o e longe dis ances. Ou esul s suppo he
idea ha edge dis up ion would be easie o de ec on animals u he away and wi h low isual acui y, because
as he dis ance be ween p eda o and p ey inc eases, he la e ’s ou line becomes less isible han he dis up i e
ma kings. Plausibly, dis up ion o he gian panda’s ou line can no longe be esol ed by he isual sys em o a
po en ial p eda o a some ecologically ele an iewing dis ances, because spa ial isual acui y is o en poo in
la ge ca ni o es26, and leads o dis ance dependen pa e n blending21,23. Fo example, spo ed hyaenas (C ocu a
c ocu a) canno esol e he wides s ipes on a plains zeb a (Equus bu chelli) beyond 50m dis ance unde
0.2
0.3
0.4
0.5
0.6
0.7
5813 21 34 55 89 144233 377
Dis ance
Edge dis up ion
(10 cpd)
Dich oma ic ca ision
0.2
0.3
0.4
0.5
0.6
0.7
5813 21 34 55 89 144233 377
Dis ance
Edge dis up ion
(12 cpd)
Dich oma ic dog ision
0.2
0.3
0.4
0.5
0.6
0.7
5813 21 34 55 89 144233 377
Dis ance
Edge dis up ion
(72 cpd)
T ich oma ic human ision
Figu e5. Gian panda edge dis up ion. The me hod uses Gabo il e ing o quan i y he a io o alse edges
ha un o hogonal o he animal’s ue ou line. Edge dis up ion was modelled h ough h ee isual models
and hei espec i e acui ies: dich oma ic elid (ca —10cpd) and canine (dog—12cpd), as well as ich oma ic
human ision (72cpd) o compa ison. The edge dis up ion inc eases as a unc ion o dis ance be o e le elling
ou a longe dis ances. The boxplo shows minimum and maximum (whiske s), median ho izon al line and he
in e qua ile ange o he simula ed da a.
7
Vol.:(0123456789)
Scien i ic Repo s | (2021) 11:21287 | h ps://doi.o g/10.1038/s41598-021-00742-4
www.na u e.com/scien i ic epo s/
ligh ing condi ions when hey hun 26. A la ge iewing dis ances, we belie e ha bo h he dis up i e ma kings
and he animal’s ou line will be unde ec able21,27,28. We suspec ha he gian panda’s colo a ion shows ea u es o
bo h maximum dis up i e colo a ion and di e en ial blending depending on who iews hem and om which
dis ance8,15,16.
Thi d, ou da a bus ed he my h o conspicuousness o gian pandas. Al hough a close dis ances, as in cap i e
se ings, he species appea s conspicuous o humans, gian pandas sha e mo e han hal o hei colou s wi h
hei na u al habi a . In e ms o a simila i y- o-backg ound index (i.e., ch oma ic animal e sus backg ound
o e lap) using he colou map echnique, he gian panda g oups oge he wi h se e al species widely conside ed
as camou laged.
Any appea ance in na u e is dependen on iewing condi ions and pe cep ual sensi i i ies, as well as he isual
acui y o he obse e . In ligh en i onmen s which mos ly consis o b igh illuminan small gaps, la ge gaps
and pa ial woodland shade12, con as ing black-and-whi e colo a ion can se e as bo h backg ound ma ching
and dis up i e colo a ion, and his possibili y should be explo ed u he in o he animals wi h conspicuous
black-and-whi e ma kings. I is plausible hough ha he black-and-whi e colo a ion is conspicuous close up,
and we canno he e o e discoun he possibili y ha isual signalling, whe he o conspeci ics o he e ospeci -
ics, would be impo an o he gian panda29, a conclusion eached by Ca o and colleagues11 o he black eye
Figu e6. Simila i y o backg ound measu ed as o e lap in colou space. The ch oma ici y diag am using
human ision shows ha 50% o all colou s wi hin images we e pe cep ually simila be ween he gian panda
and i s backg ound. The leas simila species o hei espec i e backg ounds a e hose usually conside ed as
being aposema ic, whe eas he o he end o he spec um consis s o he camou laged animals. The gian panda
alls a he middle o his ‘conspicui y spec um’, and among o he species ha a e conside ed as camou laged,
sugges ing ha gian panda pelage colo a ion does no s and ou agains he backg ound mo e han a concealed
species would. All inse pic u es: Wikimedia commons open access license (CC BY-SA 3.0).
8
Vol:.(1234567890)
Scien i ic Repo s | (2021) 11:21287 | h ps://doi.o g/10.1038/s41598-021-00742-4
www.na u e.com/scien i ic epo s/
and ea ma kings o his species. In summa y, he gian panda uses black-and-whi e pelage as a o m o c ypsis
o a oid de ec ion in i s na u al habi a and i s appa en conspicuousness o he human eye is an a e ac o he
sho iewing dis ances ound in pho og aphy and zoos.
Me hods
Pho og aphs o he gian pandas in hei na u al habi a . The gian panda li es in agmen ed,
humid and emo e moun ain o es s in Sichuan, Shaanxi and Gansu P o ince in sou h-cen al China, which
makes i di icul o ob ain da a on his endange ed animal in i s na u al habi a 11. Fi een encoun e s (assumed
o be o di e en indi iduals) we e pho og aphed om 2007 o 2014 by FW in Sanguanmiao Field S a ion, Fop-
ing Na u e Rese e, Shaanxi P o ince. The pho og aphic da a we e collec ed om a na ow geog aphic ange,
bu we ha e no eason o expec ha ou esul s would be limi ed o only wi hin ou sampling egion.
Animals we e pho og aphed om a ious dis ances (es ima ed o be app oxima ely 5–150m) as hey we e
encoun e ed in hei en i onmen . The iming o he pho og aphs a ied ac oss seasons: some (7) we e pho-
og aphed when he e was snow on he g ound whe eas o he s (8) we e pho og aphed in summe condi ions.
The en i onmen in hese a eas comp ises se e al dis inc i e isual elemen s such as ocks, g ound subs a e,
small s eams, p ima y o es ee unks, seconda y g ow h sh ubs (including bamboo), oliage and a mix u e
o al e na ing b igh sunli spo s and da k shady a eas. The o es geome y plausibly co esponds o ligh p o ile
o o es shade wi h small ligh gaps12. As i was no easible o explo e he i adiance p o iles o hese di e en
o es pa ches he e, we used s anda d D65 illumina ion in he pho og aphs as a p oxy. Fo de ails o pho og aphy
and using he micaToolbox as well as QCPA see appendices (TableS1).
Image analysis echniques. We examined pho og aphs o 15 gian pandas using he Mul ispec al Image
Calib a ion and Analysis (MICA) oolbox in Image J ( e sion 1.52k/Ja a 1.8.0_172 (64-bi )17 and i s ex ension,
he Quan i a i e Colou and Pa e n Analysis (QCPA) amewo k18—micaToolbox e sion 15.3.2019. Pho o-
g aphs we e p ocessed in digi al nega i e o ma (.dng) and we used a Canon EOS400D wi h ki speci ica ions
(18–55mm lens) in quan i a i e colou and pa e n analysis amewo k ( o de ails see TableS1). Each image
was con e ed in o mul ispec al image s acks as ollows. The pho og aphs’ whi e balance was adjus ed h ough
he MICA oolbox plugin wi h ImageJ so wa e and was no malised o 7 pe cen black using he gian panda
black (da k ones) and 63 pe cen whi e e lec ance using he whi e (ligh ones) pelage pa ches and 97 pe cen
snow e lec ance as p oxies (TableS1). The e lec ance alues o he pelages we e con i med p io o analyses.
Fo his, wo cap i e gian pandas we e pho og aphed in Äh ä i Zoo, Finland, nex o calib a ed pho og aphic
s anda ds (Fig.S1). No e ha he exac alues o u e lec ance a e no c ucial, bu i is impo an o no malise
he dynamic ligh ange ac oss he pho og aphs30. We used panda size o s anda dise scale ac oss he pho o-
g aphs: leng h was used as ‘a measu ing s ick’ o app oxima e ela i e dis ance in he edge dis up ion analysis
(see below). As egions o in e es (ROIs), he en i e animal ou line (animal) and he emaining su ounding
isual en i onmen excluding he animal (backg ound) we e selec ed.
Quan i ica ion o backg ound ma ching. We es ed how closely he pelage pa ches o he gian panda clus e s
wi h he isual pa chiness o he backg ound. The RNL clus e ing me hod uses spa ial il e ing o de e mine
which luminance elemen s g oup oge he in he image (i.e., he me hod clus e s pixels acco ding o hei
pe cei ed colou o luminance con as —ou app oach o his ques ion akes accoun o luminance only,
TableS1—as measu ed in Euclidian dis ance be ween pixels in he RNL colou space). Fo de ails see18. Acui y-
co ec ed cone ca ch images we e used wi hou dis ance scaling o h ee ision models: domes ic dog and ca ,
and human ision. De ails a e epo ed in TableS1. The ob ained clus e s we e coun ed as dummy a iables
(i.e., dicho omous: ue o alse) wi h espec o whe he o no each pelage g ouped wi h speci ied backg ound
isual ea u es; associa ions be ween gian panda pelage pa ches and he g ound, lea es, ocks, shades, snow
and ee unks we e de e mined using ela i e equencies (i.e., numbe o associa ions di ided by he o al
numbe o samples). We also used he QCPA’s Visual Con as Analysis (VCA) o de ailed luminance analysis
(TableS2), and o con i m he alidi y o human ision modelled da a in compa ison o dich oma ic ision
sys ems (domes ic dog and ca ) wi h he ollowing isual acui ies: dog 12 cpd31, ca 10 cpd32,33, and human 72
cycles pe deg ee—cpd27.
Quan i ica ion o dis up i e colo a ion. We es ed whe he ee-li ing gian pandas show dis up i e colo a ion
ope a ing as a unc ion o dis ance, using Gabo il e ing o ach oma ic edge dis up ion which quan i ies he
a io o alse edges ha un o hogonal o an animal’s ue ou line19. This app oach allowed us o es edge dis-
up ion using only he luminance channel. The pic u es we e i s s anda dised o ixed pixels pe dis ance a io.
The dis ance was modelled as ela i e body leng hs ( he panda body leng h was se as 1, which co esponds o
app oxima ely 1.5m). We modelled he dis ance o a ge using he acui y iew ool in he oolbox18 wi h Fibo-
nacci sequence inc emen s o ob ain exponen ial g ow h o longe dis ances (s a ing a 5). A GabRa Sigma
alue o 3 was used19. We modelled edge dis up ion h ough all h ee ision sys ems explo ed he e (ca , dog,
human) and used hese as a p oxy o he gian panda’s would-be p eda o s26,34, because he la e ’s exac spa ial
acui ies emain unce ain.
Exposing he my h o conspicuousness. We es ed whe he gian panda colo a ion ma ches na u al backg ound
colou a ia ion as judged by a colou map compa ison echnique18. These colou maps can be conside ed as a
non-pa ame ic his og am ep esen a ion o he spec al a ia ion cap u ed in any image egion using ecep o
noise limi ed colou space and acui y co ec ion. The colou map ool in QCPA quan i ies ch oma ic con as
as ΔS in he ch oma ic Recep o Noise Limi ed (RNL) space18. As such, i ca ego ically igno es luminance35